Changelog

New: AI in Radio, a strand page and a guide

• AI in Radio: Voice Tracking and Scheduling maps three tools by what the makers state and shows what we could not read.
• AI Voice Tracking and Scheduling for Radio covers what ENCO and RCS state about AI-generated links between songs and plain-language scheduling, the fact that no page we read says who reviews generated copy, and a trial to run off air.
• Labelled Experimental and not tested by us. ENCO's own page and two trade articles were read. RCS's own page and the Futuri study and CoHostAI coverage returned errors, so nothing is claimed from them beyond their existence. This completes the four professional strands.

New: AI in Broadcast, a strand page and three guides

• AI in Broadcast: Live Captions, Live Translation and Quality Control maps six tools by what the makers state and shows which makers' pages we could not read.
• Live Captioning for Broadcast covers what Verbit and AI-Media state, the absence of any accuracy figure, and a test on your own programmes.
• Live Subtitles and Dubbing in Other Languages covers the EBU's release on Eurovision Sport and CAMB.AI, which names no languages and no accuracy, and a trial with a native reader.
• AI Quality Control and Archive Search for Broadcasters covers Interra BATON, Telestream Qualify and Moments Lab, and how to pilot them on files whose faults you know.
• All three are labelled Experimental and not tested by us. We read the makers' own pages for Verbit, CAMB.AI, Interra, Telestream's blog and Moments Lab. AI-Media's pages, the regulator's caption rules and two Interra and Telestream releases were unreachable or seen only as search summaries, and each guide says so.
• Moments Lab's page does not mention Dalet, so an earlier note of ours that it integrates with Dalet is unconfirmed and is not repeated.

New guide: digital replicas and synthetics under the SAG-AFTRA 2026 TV/Theatrical summary

• Digital Replicas and Synthetics follows the union's own Summary of Tentative Agreement for the 2026 TV/Theatrical contracts, dated 11 May 2026, and cites its section numbers. It covers consent to scan and the business reason for asking, what counts as a scripted alteration, consent for dubbing with a replica on pictures that begin principal photography on or after 1 July 2027, pay for independently created replicas, replica security and transfer, fixed limits, and the notice and bargaining duties around synthetic performers.
• It is a summary of a tentative agreement. It predates the ratification vote, does not report the result, and is not the contract, so the guide says what that leaves open. It is not legal advice, and it does not cover other union contracts, non-union work or law outside the United States.
• Pages that said we had not read the 2026 union terms now say what was read and what was not. They are the AI Tools hub, the audio post and cinema and TV strand pages, the synthetic voice guide and the dialogue cleanup in post guide.

New: AI in Cinema and Television Production, with three guides

• A second professional strand, AI in Cinema and Television Production, covers the cutting room, VFX and restoration. Its three guides are AI in the cutting room, AI in VFX and AI in film restoration.
• We tested none of these tools. Statements come from the makers' own pages, read on 8 October 2026. Media Composer and Autodesk Flow Studio are described from a trade report and an unopened search summary, because the makers' pages returned access errors, and the guides say so.
• There is no guide on digital replicas and synthetic performers. It depends on the 2026 SAG-AFTRA agreement, whose own page returned an access error twice, and the secondary summaries we found include some that describe the 2023 deal. We will not write about its terms until the agreement itself has been read, and the strand page says so.
• Two claims in our research notes did not appear on the makers' pages and are not repeated: IntelliScript in DaVinci Resolve 21, and a Nuke plug-in for Beeble. The AI Tools hub gained three jobs and links to both strands.

New: AI in Professional Audio Post-Production, with four guides

• A new strand page, AI in Professional Audio Post-Production, covers film, television and broadcast audio post. It asks four questions a professional tool has to pass: real time or offline, whether it loads in your workstation, whether it is steady across a reel, and whether the delivery specification and performer consent allow it.
• Four new guides cover dialogue cleanup in the post chain, splitting a mix into dialogue, music and effects, stereo to surround upmix and transcription inside Pro Tools.
• We tested none of these tools. Statements come from the makers' own pages, read on 8 October 2026. The Pro Tools guide rests on trade coverage because Avid's page returned an access error, and says so.
• The AI Tools hub gained two jobs, dialogue-music-effects separation and stereo upmix, and a link to the new strand.

New: AI Tools for Creative Work, with eight job guides

• A new hub, AI Tools for Creative Work, starts from the job you need done and lists tools at four budgets: free, starter, pro and studio. It has a finder, routes by who you are, and a table of every tool with where it runs and what hardware the maker states. It is labelled Experimental.
• Eight new guides cover dialogue cleanup, live noise removal, stem separation, transcription, synthetic voice, video upscaling, image fill and video generation.
• We tested none of these tools. Each statement about a tool comes from the maker's own page, read on 8 October 2026, and each page says so. Six tools are listed from press coverage because their makers' pages could not be read, and the hub marks them "Not yet" under "Checked against maker".
• The earlier roundup, AI Tools for Media Creators in 2026, is folded into the hub and now redirects to it. Its judgement on where AI is mature and where the hype outpaces the result moved to the hub. AI Tools is a new link in the menu on every page.

Mobile menu no longer cuts off its last links

• The burger menu on phones was capped at 320 pixels tall, which is room for about six of its nine links. Blog, About and Search were clipped on every page. The menu is now as tall as its links need and scrolls on a very short screen.

Blog added to the menu on the course pages

• The 51 Application Intensive course and module pages left the Blog link out of both the desktop navigation and the mobile menu behind the burger button. It is now between Accessibility and About, as on every other page.

Worked examples on four more tools

• Pitch Analyzer, Speaking Pace Analyzer, Stereo Width Checker and Silence Remover. Each has a worked examples section computed from the formula in the tool's own script, covering note frequencies and rounding, peaks to words per minute, correlation against side level, and threshold in decibels against the linear level.
• Streaming Bitrate & Bandwidth Calculator. The worked examples table did not match the tool. It listed 720p60 at 3,500 kbps where the tool uses 4,000, showed a 480p row the tool does not offer, left out 1440p60 and 4K60, and used a 25 per cent headroom where the tool uses 1.5 times. It is rebuilt from the tool's own calculation.

Three tools that did not run, repaired, and fifteen worked examples rebuilt from the tools

• Depth of Field Calculator, Video Bitrate Calculator and Synthetic Sound Generator. In each, a comment line had swallowed the start of the next statement, so the tool stopped with an error when it ran (the first two on load, the third when generating a sound). The statements are back on their own lines, and a browser load of every tool page now finds no script errors.
• Audio Denoise. The drop area's accessible name did not contain its visible text. The label is removed so the name comes from the text on screen.
• Worked examples rebuilt from each tool's own calculation. Fifteen sections on the cable loss, latency and buffer, limiter and true peak, monitor distance, export file size, gain staging, checksum offload, lux and beam angle, lighting power, lens field of view, RAID, offload storage, music royalties, YouTube retention and frame rate tools did not match what the tool computes. Some used invented presets or ratings, one rounded the number of fixtures on a circuit up instead of down, and one listed a retention table with a stray paragraph inside it. Each is now computed from the tool's script, with every quoted figure checked in the generating script.
• Rounding. Four tables rounded an exact tie to the even digit where the tool rounds up. BPM Delay Calculator, Kinetic Timing and Lens Field of View now match what the tool displays.

Flash checker no longer clears a video it sampled too thinly

• Flash Checker. A test with a real five-per-second flashing clip in a browser showed that at the old default of 4× analysis speed the tool read only about 7 frames per second and reported no flashing. The default is now 1×, which read the same clip at about 30 frames per second and flagged it. Below 20 sampled frames per second the tool now says it could not clear the video instead of reporting a pass.

Four history and future claims corrected

• Mechanical sound makers. The page no longer places Kircher's Organum Mathematicum in the 1650 Musurgia Universalis or calls it a music composer. The records read describe a rod-based calculating machine, described by Kaspar Schott in 1668.
• Radio broadcasting. KDKA is now described in the IEEE's wording, a world pioneer of commercial radio broadcasting that began scheduled programming on 2 November 1920, rather than as the first licensed station.
• Foley sound recording. The page no longer dates Jack Foley's Show Boat episode to 1927. The AFI Catalog dates the start of production to July 1928 and CineMontage dates the problem to October 1927, so the page gives no date.
• Content provenance. The page no longer says several camera makers have implemented C2PA. It names the standard's steering members, as listed on the C2PA site.

Storage calculator repaired, three worked examples added

• Recording Storage Calculator. The calculator did not run at all. A comment line had swallowed the start of its bitrate table, so the script failed to load and the page showed nothing. It is fixed, and a check of every inline script on the site found no other case. The page also has worked examples now, computed from the tool's own table.
• Podcast CPM & Sponsorship Revenue Calculator and VO Rate Calculator. Each has a worked examples section computed from the formula in the page.

Sources sections on the future guides

• All 26 future guides now carry a Sources section. They are forecasts, so each one says which present-day claim it rests on and whether a record was read for it. The content provenance and media creation guides cite the C2PA overview, the neural codecs guide cites the AV1 specification page, and the others say plainly that no record was read and name the record that would settle the claim.

Sources added to nineteen history guides

• The animation, film, film censorship, esports broadcasting, streaming video, online video, podcasting, radio broadcasting, visual effects, mechanical sound makers, Foley and live broadcast history guides now carry a Sources section naming the patents, press releases, museum and institute records that were read, with what each establishes and what was not checked. The broadcast journalism, camcorder, color film, game audio, music video, sampling and video editing guides carry a Sources section that says no primary record was reached and names the record that would settle each date.

Four smaller claims corrected after testing

• Audio Restoration Beyond Noise Removal. A notch filter no longer "removes hum with far less collateral damage" than noise reduction. It removes less hum as it gets narrower, and only works on an exact frequency.
• Media Metadata Standards: EXIF, XMP, and Sidecars. Step 5 now says that -map_metadata -1 emptied an MP4 and a WAV file but left a JPEG's EXIF and GPS, so the output must be checked.
• Audio Compression Explained for Beginners. The tip about 4:1 then 2:1 now says it only helps with separate thresholds and timing, since at one threshold it is the same curve as 8:1.
• Why Your Live Stream Looks Bad, and How to Fix It. Step 5 now says a lower resolution scored better only at 1000 kbps in our test, and 1080p scored higher from 2500 kbps.

Flash checker now counts to the definition

• Flash & Photosensitivity Checker. The checker counted every change of direction as a flash, so 2 flashes per second read as 4, judged only the luminance of the whole frame, so it missed saturated red against grey and a white patch of 8 percent of the frame, and ignored the area allowance. It now follows small blocks of the picture separately, counts pairs of opposing transitions, counts flashing only where at least about 2.8 percent of the frame shares it (the definition's quarter of a 10 degree field), and tests saturated red by the definition's red ratio and chromaticity change. The page text says so. Flashing Content and Photosensitivity has the new results on its test clips. Tested in Node on generated frames, not yet on real video in a browser.

Video metadata claim softened and sourced

• Verifying Found and User-Submitted Video. Step 3 and the FAQ said metadata is commonly or frequently stripped entirely by social platforms. The Berkeley Protocol says some metadata may be stripped and that platforms often re-encode uploads, so the metadata describes the new file. The page now says that, and a Sources section names the Protocol and says no platform was tested.

IAB measurement section corrected and sourced

• Podcast Hosting, RSS Feeds & Dynamic Ad Insertion Guide. Section 4 cited version 2.1 of the IAB Tech Lab's podcast measurement guidelines, which are now at 2.3, and called them a standard when they are guidelines. The Apple Siri example is gone, since neither version has it. The 24 hour window moved from the one minute rule to de-duplication, where the guidelines put a calendar day or a rolling period. The $18 to $45 CPM figure was removed because no IAB document gives prices. A Sources section names both versions.

Log footage exposure targets sourced

• Log Footage Explained: S-Log, C-Log, V-Log. The unsourced "40 to 55 percent for most log profiles" became the makers' own grey points: 41 percent on S-Log3, 42 on V-Log and about 33 on Canon Log 3, with light skin about a stop above grey, roughly 43 to 50 percent. None of the three makers publishes a skin target. The 50 to 60 percent figure after the LUT is now called a starting point. A Sources section names the three documents.

Sample rate conversion claim removed and sourced

• Audio Sample Rate Conversion & Anti-Aliasing Filters Guide. The sentence saying iZotope RX and Weiss SARACON upsample to multiples in the tens of megahertz with over 170 dB of stopband rejection is gone. Neither vendor's documentation says it, and the common multiple of 44,100 and 48,000 Hz is 7.056 MHz. It now says what the two documents do say, and a Sources section names them.

Remote podcast platform details corrected and sourced

• How to Record a Remote Podcast Guest. Zoom had it backwards: local recording works on a free account in the desktop app, and cloud recording needs a paid plan. Skype was retired by Microsoft on 5 May 2025 and is no longer listed. SquadCast is now part of Descript. A Sources section lists the pricing and support pages read and what was not checked. The Riverside and SquadCast prices were confirmed and did not change.

Stream bitrate figures corrected and sourced

• Why Your Live Stream Looks Bad, and How to Fix It. YouTube's live encoder table now gives 720p at 30 or 60 frames per second a 3 Mbps minimum and an 8 Mbps recommendation for H.264, and 2 and 6 Mbps for AV1 and H.265, and publishes no maximum. The page had 8 Mbps as a maximum, 6 Mbps as the recommendation and 4 Mbps for 720p at 30. The 2500 kbps point now says it sits below the H.264 minimum. A Sources section names the help page and what was not checked.

Shared storage networking figures corrected and sourced

• 10GbE Shared Storage & NAS Networks for Video Editing. OM4 fibre reaches 400 meters on 10GBASE-SR, not 300, which is OM3, and no fibre is "zero-loss". 10GBASE-T runs to 100 meters on unshielded Cat 6a, not only shielded, and to 55 meters on Cat 6. The unsourced 4 W per port figure gave way to Cisco's 2.5 W 10GBASE-T module rating and Lenovo's 11.2 W for a dual-port card. ProRes 422 HQ is now 88 to 111 MB/s per UHD stream at 24 to 30 fps, drive speeds are labelled as outer-track best cases, RAID 6 parity is described as spread across the drives, and the cache paragraph no longer says a ZFS SLOG or L2ARC absorbs project saves. A Sources section names the documents read.

Wireless microphone band rules and cable figures corrected and sourced

• Wireless Audio: RF Frequency Coordination & Squelch Guide. The page said microphones are now illegal across 614 to 698 MHz in North America and called 941 to 960 MHz an exclusive band. Under the current US rules (47 CFR 15.236 and 74.802), 614 to 616 and 657 to 663 MHz stay open to unlicensed microphones and 653 to 657 MHz to licensed ones, and Part 74 licensees are not granted exclusive assignments in four segments between 941.5 and 959.85 MHz. Paddle gain is now given as about 6 dBi, RG-58 loss follows Belden's figures by frequency, RG-8X is no longer grouped with low-loss cable, the squelch passage no longer says high squelch halves the range or names a trade name, and a Sources section names the documents read.

Surround calibration levels corrected and sourced

• 5.1 Surround & 7.1.4 Dolby Atmos Calibration Guide. Section 3 said Dolby and the EBU recommend 79 dBC per speaker with the LFE reading 89 dBC. The EBU sets 78 dBC at -18 dBFS, the only document read that gives 79 is Netflix's near field spec (79 or 82 dB SPL), and no Dolby small room guideline was reached. The section now gives Netflix's, the EBU's and ATSC A/85's levels by room size, gives the LFE as +10 dB in band on a real-time analyser (Dolby's cinema spec and the EBU), drops the unsupported "severe ear fatigue" claim, and a Sources section names the four documents read.

Broadcast loudness targets corrected and sourced

• Meeting Broadcast Loudness Requirements. The page gave -27 LUFS as a theatrical target and -24 LUFS ±2 as the EBU R 128 figure. The -27 LKFS figure is Netflix's dialog-gated target for near field (home) mixes, and Netflix sets no loudness target for theatrical mixes. EBU R 128 is -23.0 LUFS ±0.5 LU, and -24 LKFS is the ATSC A/85 recommendation, whose ±2 dB is measurement tolerance rather than a range to aim at. The unsourced -13 LUFS laptop figure and the "Dolby 7" reference were removed, the room levels now follow A/85 and Netflix, and a Sources section names the three documents read.

A Spotify podcast loudness claim corrected

• LUFS for Podcasts. The FAQ said Spotify normalizes all podcasts to -16 LUFS, which contradicted the page's own step 2 and Sources. Spotify's loudness article gives -14 LUFS for music tracks and does not mention podcasts, and the -16 LUFS figure Spotify publishes is for podcast ads, so the page now says Spotify publishes no target for podcast episodes. The step, the Platform Standards list and the Sources section say the same, and Spotify's podcast ad requirement is added as a source.

A mastering ceiling claim corrected

• Audio Mastering Signal Chain. Section 4 said a -1.0 dBFS ceiling guarantees clean transcoding on Spotify and Apple Music. In our test, FFmpeg's native AAC encoder at 128 kbps decoded a master limited to -1 dBTP at up to +3.4 dBTP, so the page now says the ceiling leaves headroom but guarantees nothing, and that the encoders those services use were not tested. The ceiling is also given in dBTP, the unit a true peak limiter uses.

Faststart claim qualified on the browser video page

Browser Video Playback said that without faststart the browser "must download the entire file". In our test with a server that answers byte range requests, Chrome and Firefox showed a first frame after about 5 to 6 percent of the file. The step and the FAQ now say the whole download happens when the server cannot answer range requests, and that otherwise faststart saves an extra request for the file’s tail.

Timecode drift figures corrected on two pages

• Multicam Genlock and Timecode Sync. Section 4 said a standard clock drifts "1 frame every 2 hours (around 20 PPM)". 20 ppm is 144 ms in 2 hours, 3.5 frames at 24 fps. It now says so. The 0.5 ppm line now reads about 43 ms, roughly one frame at 24 fps, over 24 hours, instead of "less than 1 frame", and no longer promises mesh systems prevent "any drift whatsoever".
• SMPTE Timecode, LTC and Jam Sync. The claim that timecode boxes "guarantee less than 1 frame drift in 24 hours (±0.5 ppm)" now says what ±0.5 ppm works out to, about 43 ms or one frame at 24 fps.

Two flicker-safe shutter lists corrected

• Indoor Sports Lighting. The page listed 1/200 s at 50 Hz and 1/240 s at 60 Hz as flicker-safe. Each is half a light cycle, and in our model the light varied by about 2.2 stops across frames at full modulation depth. The steps, the explanation and both FAQ answers now give 1/50 and 1/100 s at 50 Hz and 1/60 and 1/120 s at 60 Hz, and say that faster speeds still band under lamps that pulse deeply. This now agrees with the flicker and shutter calculator.
• Video Frame Rates, Shutter Angle and Motion Blur. Section 5 gave "1/50s or 1/100s (172.8° shutter angle)". Only 1/50 s at 24 fps is 172.8°. 1/100 s is 86.4°, a stop darker. Each now has its own angle.

A new post on building a sound design reel

What a Sound Design Reel Has to Prove in Ninety Seconds is advice for sound designers applying for a first job in post-production or games. It covers how long the main reel should run, a running order to cut to, why dialogue cleanup and production sound work better as separate breakdowns, and what the website behind the reel has to carry.

The loudness targets it names were checked against the documents themselves: -23 LUFS in EBU R 128, -24 LKFS in ATSC A/85, and -27 LKFS for home viewing mixes in Netflix’s sound mix specification. For footage to redesign, the post points to Blender Studio’s films, which carry a Creative Commons Attribution licence, rather than to downloaded YouTube clips, which YouTube’s terms do not allow. Advice on reel length and running order is marked as working practice, since no published study of how long reviewers watch a reel was found.

Formulas and symbols that showed as raw code on 11 pages, and one author name everywhere

• One author name. 57 tutorials and 15 blog posts still credited "Infinite Creation Editorial Team", while every other page credits "Infinite Creation". All 72 now say "Infinite Creation" in the visible byline, the page metadata and the structured data. Nothing else on those pages changed.
• Raw maths markup removed. Eleven pages showed formula markup as text, with dollar signs and backslashes where the page meant a symbol such as 180°, because the site loads no maths renderer. All 41 cases now read as plain text with ordinary symbols (×, °, ±, →, Ø). The wording and figures are unchanged. Pages: Audio Latency Buffer Calculator, Light Beam Angle and Lux Calculator, anamorphic vs spherical, dialogue editing, drum recording phase, synthesizer history, sample rate conversion, SMPTE timecode, Steadicam balance, Atmos speaker calibration and wireless RF coordination.

A calculated section on the depth of field guide

• Depth of Field for Video: Aperture, Bokeh, and Focus. New "Calculated" section that works out depth of field with the standard formulas. It scales with the f-number, falls with the square of the focal length at the same distance, and rises with the square of the distance, as the page says. At the same framing, though, it is nearly the same for every focal length, which the page's "telephoto gives shallower DOF" leaves out. A telephoto does blur the background more at the same framing, which is the effect people usually mean. Full frame is about 1.5 times shallower than APS-C and 2 times shallower than Micro Four Thirds at the same distance, focal length equivalent and f-stop.

A calculated section on the three point lighting guide

• 3-Point Lighting & Cinematic Contrast Ratios. New "Calculated" section that works out the page's contrast ratios. 2:1 to 32:1 are 1 to 5 stops, as the page says. Under the page's own definition (key plus fill over fill) the key alone is only 1, 3, 7, 15 and 31 times the fill, and the fill sits 1.4, 1.7, 2.6, 3.9 and 5.6 times as far from the subject as an equal fixture would for the same ratio. Read the other common way, key over fill, 4:1 is 2.3 stops.

A calculated section on the live sound mixing guide

• Live Sound Mixing: FOH, Stage Monitors & Feedback Control. New "Calculated" section that works out three claims. 150 feet is 133 ms, as the page says, but a swing from 15 to 30 C moves that by about 3.4 ms, most of the 5 ms margin. A supercardioid null falls at 126 degrees, as the page says, and a wedge 45 degrees off the rear gives about 22 dB of rejection against 12 dB directly behind. A two source cardioid sub model keeps its rear rejection near 30 dB at 80 Hz for a 0.1 ms delay error.

A calculated section on the studio monitor placement guide

• Studio Monitor Placement: Equilateral Triangle & SBIR Guide. New "Calculated" section that works out the SBIR null frequency, c divided by four times the wall distance. 2 feet from the wall is 141 Hz, the top of the page's 100 to 140 Hz range, and 8 inches is 422 Hz, above its 300 Hz. The 1 to 3 foot danger zone spans 94 to 281 Hz. Our assumed desk geometry gives a bounce delay of about 0.6 ms, shorter than the 1 to 2 ms the page states.

A measured section on the drum recording phase guide

• Drum Recording: Mic Techniques, Phase Alignment & Bleed. New "Measured" section built on the comb filter test. A 3.4 ms offset puts the first notch at about 146 Hz, but the cancellation is only complete when the late copy is as loud as the first: at half the level the dip is 6 dB, at a tenth it is under 1 dB. Two overheads summed to mono with a 50 mm distance mismatch have a first notch near 3.4 kHz, so a millimetre is far more precision than the sum needs. A snare bottom summed without inverting it lost about 7 dB at 100 Hz in the model and gained about 4 dB when inverted.

A measured section on the lav and boom phase alignment guide

• Dialogue Editing & Phase Alignment: Lav vs Boom Guide. New "Measured" section that sums noise with a delayed copy in FFmpeg and reads the notches. The first notch is at 1 divided by twice the delay, so 2 and 3 ms put it at 250 and 167 Hz, with a notch every 500 and 333 Hz. A 2 to 3 ms offset is 96 to 144 samples at 48 kHz, not several hundred, and the page's own geometry gives a path difference nearer 1.2 to 2.2 ms.

A tested section on the vocal tuning guide

• Vocal Tuning & Pitch Correction: Melodyne vs Auto-Tune. New "Tested" section with two checks. A synthetic vowel shifted four semitones by speeding up playback moved its formants by the same ratio as its pitch, which is the chipmunk effect the page describes. A first order model of retune speed left about 65 percent of a 5.5 Hz vibrato at a 25 ms time constant and about 87 percent at 50 ms. The formant-locking algorithms themselves were not tested.

A calculated section on the surround and Atmos calibration guide

• 5.1 Surround & 7.1.4 Dolby Atmos Calibration Guide. New "Calculated" section that works out the time of flight rule for a 7.1.4 room. 1.13 feet per millisecond is right at 20 C and moves by about 1 percent across 15 to 25 C, the top speakers in a 2.7 m ceiling room sit about 0.9 m closer than ear level speakers and need about 2.6 ms of delay, and one sample at 48 kHz is 20.8 microseconds and 7.2 mm, so "the exact same microsecond" is a target and not a result.

A calculated section on the wireless audio RF coordination guide

• Wireless Audio: RF Frequency Coordination & Squelch Guide. New "Calculated" section that works out third order intermodulation for real channel plans. Six carriers evenly spaced at 400 kHz have a third order product on every carrier, and a random greedy search on a 25 kHz grid found a large group in 6 MHz that keeps every carrier 350 kHz apart and every product 100 kHz clear of any carrier. The formula in the page is right and the 350 kHz spacing is not enough on its own.

A calculated section on the film set power guide

• Film Set Electrical Power Calculations & Location Distribution. New "Calculated" section that recomputes the page's figures. Watt's law, the 80 percent limits for 15 A, 20 A, 13 A and 16 A circuits and the generator leg current all check out. The voltage drop guidance needs a caveat: 12 AWG carrying 20 A over 100 feet drops 5.3 percent at 120 V, just over the page's 5 percent limit for lighting and above its 3 percent limit for camera electronics.

A calculated section on the multi-camera timecode guide

• Multi-Camera Production: Genlock, Timecode Jamming & Sync. New "Calculated" section that turns the page's ppm figures into frames. 20 ppm is about 3.5 frames every 2 hours at 24 fps, not one, and one frame every 2 hours at 24 fps is about 5.8 ppm. 0.5 ppm is 43 ms over 24 hours, 1.04 frames at 24 fps and 1.3 at 29.97, so "less than 1 frame" holds only below 24 fps.

A checked section on the 10GbE shared storage guide

• 10GbE Shared Storage & NAS Networks for Video Editing. New "Checked" section that recomputes the page's link figures and measures a ProRes rate. Header overhead limits TCP goodput to about 118 MB/s on 1GbE and 1,177 MB/s on 10GbE at an MTU of 1500, and 1,238 MB/s with jumbo frames, so the page's 100 to 110 and 1,150 to 1,200 MB/s sit at or just under the ceiling. FFmpeg's ProRes 422 HQ encoder wrote a clean UHD test pattern and a noisy one at different rates around the page's 92 MB/s figure, so the streams per link depend on the picture.

A tested section on the EQ basics guide

• EQ Basics: What Every Control Does. New "Tested" section that reads FFmpeg's EQ filters from their impulse responses. A 10 dB bell at 1 kHz was 2.5 octaves wide between its half-gain points at Q 0.5 and 0.3 at Q 5. A 6 dB high shelf at 8 kHz gave half its boost at 8 kHz and 5.4 dB at 16 kHz, and a high pass passed 0 dB above its corner and removed the lows.

A calculated section on the frame rates and shutter angle guide

• Video Frame Rates & 180-Degree Shutter Angle Rule. New "Calculated" section that recomputes the page's shutter angle arithmetic. Most of it holds: 60 and 120 fps on a 24 fps timeline are 2.5 and 5 times slower, and 1/24 s at 24 fps is 360 degrees and one stop more than 180. One figure does not: 1/100 s at 24 fps is 86.4 degrees, not the 172.8 degrees the page gives for 1/50 s and 1/100 s together.

A calculated section on the indoor sports lighting guide

• Shooting Indoor Sports: Gyms, Pools, and Flicker. New "Calculated" section that models discharge light pulsing at twice the mains frequency. Exposures of 1/100 s and 1/50 s in a 50 Hz region gathered the same light at every start time, but 1/200 s, which the page lists as safe, left the light varying by about 2.2 stops at full modulation depth and 0.6 stops at a mild depth of 0.3. The same is true of 1/240 s in a 60 Hz region. Only whole numbers of light cycles are safe.

A calculated section on the tempo maps and hit points guide

• Tempo Maps and Hit Points: Writing Music to Land on Picture. New "Calculated" section that recomputes the page's tempo arithmetic. An 8 second gap takes 120 beats per minute at 16 beats and 90 at 12, as the page says, and 105 at 14 and 135 at 18. A tempo 4 beats per minute slow over the 16 beat gap arrives 0.28 s late, 6.6 frames at 24 fps, so "four frames late" needs only about 2.4 beats per minute of error. Moving from 120 to 123 across eight bars buys 0.20 s, about 4.7 frames at 24 fps.

A calculated section on the sharpness guide

• Why Your Photos Are Not Sharp, and Which Cause You Have. New "Calculated" section with diffraction and subject blur worked out. For a perfect lens at 550 nm, contrast at the sensor's Nyquist frequency on a 24 megapixel full frame body falls to 16 percent at f/16 and to nothing at f/22, and a small sensor gets there earlier, as the page says. A walking subject at 10 m through an 85 mm lens smears about 8 pixels at 1/250 s, so "around a two-hundred-and-fiftieth" stops obvious blur and not pixel-level blur.

A tested section on the broadcast loudness guide

• Meeting Broadcast Loudness Requirements. New "Tested" section using ebur128. A very quiet section did not lower integrated loudness at all because of the gate, so the page's "quiet passages and loud passages both contribute" needs a limit. A gain of 6.02 dB moved loudness by 6.1 LU and left the loudness range at 7 LU. Downmixing a 5.1 programme to stereo changed loudness by 1.7 LU, and a surround pair polarity-inverted against an identical front pair cost 15 LU. The page's cinema and streaming target figures were not checked.

A tested section on the thumbnail and preview pipeline guide

• Building a Thumbnail and Preview Pipeline. New "Tested" section on a generated clip that opens on ten seconds of black. The first frame was black, while FFmpeg's thumbnail filter and scoring 20 sampled frames on brightness and variance both picked real frames. A 100 thumbnail sprite sheet is one file. Seeking to 5.0 s in a file with a keyframe every 10 s returned the right frame with accurate seeking and the frame from 4.8 s with keyframe-only seeking.

A checked section on the 360 video guide

• 360 Video Basics: Shooting and Publishing Immersive Video. New "Checked" section. A 5760 pixel wide equirectangular frame gives only 1440 pixels across a 90 degree view, less than a 1920 pixel viewport, so the page's "per-viewport sharpness disappoints" is arithmetic. FFmpeg's v360 placed the same visual size across 214 frame pixels at the equator and 438 at 60 degrees latitude, which is where the frame's pixels go near the poles.

A tested section on the audio compression guide

• Audio Compression Explained for Beginners. New "Tested" section that measures FFmpeg's compressor. The Knowledge Base example holds, with -16 dB in giving -18.9 dB out at 4:1 and -20 dB. A 4:1 compressor followed by a 2:1 one at the same threshold measured exactly like 8:1, so the tip about two light compressions is about thresholds and timing and not about the ratio. With an 8:1 compressor the first 5 ms of a burst peaked 6.7, 11.6, 15.0 and 16.1 dB above the settled level at attack times of 1, 5, 20 and 50 ms.

A checked section on the phone video guide

• Cinematic Video on a Phone. New "Checked" section. A simulated bar moving at 960 pixels per second left a 20 pixel trail at 1/48 s and 1.6 pixels at 1/480 s, so the 180 degree rule gives more than ten times the blur of a fast shutter. Recomputing step 4 from the sunny 16 rule, a phone lens at f/1.8 in full sun needs about 7.4 stops of ND to reach 1/48 s at ISO 100, so ND64 (6 stops) leaves it more than a stop over.

A tested section on the voice EQ guide

• Voice EQ Settings for Content Creators. New "Tested" section that reads FFmpeg's high pass and peaking filters from their impulse responses. An 80 Hz high pass at 12 dB per octave took 12 dB off a 40 Hz rumble and 2.1 dB off a 90 Hz voice fundamental, and at 24 dB per octave 25 dB and 4.2 dB. A 2 dB cut at 250 Hz was 1.9 octaves wide at Q 0.7 and 0.4 octaves at Q 4.

A tested section on the remote podcast guest guide

• How to Record a Remote Podcast Guest. New "Tested" section on the clap sync step, with two synthetic recordings whose clocks differ by 100 ppm. Aligned on the opening clap, they were 28 ms apart after four and a half minutes, and the drift extrapolates to about 375 ms after an hour. Resampling one by the measured ratio brought the error under 1 ms. The page's "they'll stay in sync for the whole interview" holds only for short recordings or matched clocks.

A tested section on the audio interface guide

• Audio Interfaces Explained. New "Tested" section that writes a quiet take as 16-bit and 24-bit WAV through FFmpeg and raises it by 30 dB. The 24-bit version stayed clean and the 16-bit version's quantisation noise came up with it to about -71 dBFS. The 24-bit file was half as big again as the 16-bit one, so "minimal" is a matter of scale. No interface was tested.

A checked section on the sports slow motion guide

• Slow Motion and Replay for Sports. New "Checked" section that recomputes the page's slow motion arithmetic and runs its conform advice in FFmpeg. Conforming 120 fps footage to 30 fps kept all 120 frames distinct at quarter speed, while slowing a 30 fps clip four times by repeating frames gave 480 frames of which only 30 were different, and interpolation gave 449 distinct frames of which at most 29 were ever captured.

A tested section on the green screen guide

• Green Screen & Chroma Keying: Lighting, Spill & Mattes. New "Tested" section that keys generated green screen frames with FFmpeg's chromakey and compares the key with a known alpha. Chroma subsampling made little difference on a smooth edge but 4:2:0 lost 36 percent of two-pixel hair strands against 25 percent at 4:4:4, and 8-bit against 10-bit made no difference. A tight key showed no background leakage up to half a stop of unevenness and 21 percent at a stop, and a looser setting showed none at a stop and a half. A screen two stops brighter clipped and was not keyed at all. Spill, despill and the dual matte workflow were not tested.

A tested section on the subwoofer calibration guide

• Subwoofer Calibration & Bass Management Guide. New "Tested" section of electrical and arithmetic checks with FFmpeg and numpy. FFmpeg's 4th order Linkwitz-Riley crossover put each output at -6.0 dB at 80 Hz with a flat sum. The Butterworth +3 dB bump needs a 4th order filter summed in phase, or a 2nd order one with a polarity flip. The polarity-flip method needs the offset within about 0.1 ms for a 30 dB null. At 80 Hz a wavelength is 14.1 feet, and a corner adds 9 dB at fixed power or 18 dB at fixed driver motion. No speaker or room was measured.

A tested section on the mastering signal chain guide

• Audio Mastering Signal Chain & Processing Flow. New "Tested" section using FFmpeg on synthetic masters. Masters at a sample peak of -0.1 dBFS measured +0.5 and +2.8 dBTP. At a true peak of -1 dBTP, MP3 and Opus came back under 0 dBTP, but FFmpeg's AAC at 128 kbps decoded at +1.3 and +3.4 dBTP, so -1 does not guarantee clean transcoding. Folding the bass below 90 Hz to mono removed a 10 dB loss in the mono sum, and FFmpeg's linear-phase EQ delayed every frequency equally, by 50 ms.

A tested section on the media app testing guide

• Automated Testing for Media Applications. New "Tested" section that runs nine encoders repeatedly and reads back hashes, durations, loudness and pixels. Only libvpx-vp9 and libopus produced different files from identical runs, differing in 16 and 40 bytes near the start while their packets and decoded output matched. Changing the thread count changed the file for x264, x265 and VP9 and never the decoded result. AAC decoded 5.3 ms long, a pure red frame read back as 254, and x264 pixel error had a maximum from 2 to 30 levels against a mean of 0.05 to 1.03. Two seconds of bars and a tone was 23 kB.

A tested section on the live stream quality guide

• Why Your Live Stream Looks Bad, and How to Fix It. New "Tested" section that scores libx264 in a live style setup at 1080p, 720p and 480p and at 1000, 2500 and 5000 kbps against a 1080p source with VMAF and PSNR, and compares VP8, VP9 and H.264 in realtime mode on one thread. 720p scored slightly higher than 1080p at 1000 kbps and lower at 2500 and 5000, and 480p came last at every bitrate. VP9 used about twice the CPU of VP8 and its quality was mixed. The sources are generated, so real footage will differ.

A tested section on the saturation and distortion guide

• Saturation and Distortion Explained. New "Tested" section using FFmpeg's asoftclip and a numpy model of a tube stage. Symmetric curves, hard clipping included, added only odd harmonics and a biased curve added a strong 2nd. Distortion rose about 200 times between a sine at a tenth of the curve's limit and one at twice it. At a fixed peak, saturation raised integrated loudness from -17.0 LUFS to as high as -5.9. Two tones through tanh made intermodulation products stronger than the 3rd harmonic, which "adds harmonic content" leaves out.

A tested section on the audio restoration guide

• Audio Restoration Beyond Noise Removal. New "Tested" section that damages a synthetic voice with hum, clicks, clipping and noise and repairs it with FFmpeg's notch, afftdn, adeclick and adeclip filters. A default afftdn pass removed none of the hum, ten narrow notches removed 33 dB of it, and given a noise profile afftdn matched a narrow notch for the same cost to the voice, so "far less collateral damage" did not hold on this voice. adeclick's default found most clicks and altered the clean voice most. De-clipping helped less as clipping got worse and made light clipping worse overall. Noise reduction last left about 4 dB less hum and noise than noise reduction first.

A tested section on the stem export guide

• Automating Stem Exports in Your DAW. New "Tested" section on the null test, with synthetic stems written through FFmpeg and summed against the mix. The residual was about -170 dBFS for 32-bit float stems, -135 for 24-bit and -97 for 16-bit. A stem 40 dB below the mix left out shows up at exactly -40 dB relative to the mix, and a 0.1 dB gain error, a one sample delay and a flipped polarity all stand clear of the floor. A stem 80 dB below the mix is invisible in 16-bit stems, and a reference mix that went through a limiter never nulls.

A tested section on the browser video playback guide

• Video Playback in the Browser: HLS, DASH, and Players. New "Tested" section that runs the page's playback claims in headless Chrome 154 and Firefox 156 against a local server that counts bytes. With the index at the end of an MP4 and a server that serves byte ranges, both browsers showed a first frame after under a tenth of the file, so "the browser must download the entire file" only holds when ranges are ignored. Firefox refused a text/plain MP4 that Chrome played. A video from another origin plays without CORS headers, and only reading its pixels or using the crossorigin attribute needs them. A wrong codec in a source element's type silently skips a playable file.

A tested section on the media metadata guide

• Media Metadata Standards: EXIF, XMP, and Sidecars. New "Tested" section using FFmpeg 8.1.2, ffprobe and Pillow on generated files. Re-encoding audio and video in FFmpeg kept container metadata, and what was lost came mostly from the destination container: MP4 dropped a custom field and WAV kept four of seven. Copying a JPEG with -map_metadata -1, the option that empties video files, left the GPS position in place. ExifTool, platforms, IPTC, XMP and sidecars were not tested.

A tested section on the photosensitivity guide, and a fix to the flash checker

• Flashing Content and Photosensitivity. New "Tested" section that scores generated frame sequences against the WCAG 2.2 definition of a flash and against the site's checker. Three flashes per second passed and four failed. A luminance step of 0.08 passed and 0.12 failed. Light grey to white passed because both states are above 0.80. Saturated red against a grey of equal luminance failed on colour alone.
• Flash & Photosensitivity Checker. Fixed a fault that stopped the tool working at all: two function declarations had been written onto the end of a comment line, so the script was a syntax error. The page section also records that the checker counts each change of direction, so it reads two flashes per second as four, that it decides on luminance alone so it misses a red only flash, and that it judges the whole frame, so a small flashing patch passes.

A tested section on the FFmpeg patterns guide

• FFmpeg Command Patterns Worth Knowing. New "Tested" section running each behaviour the page describes with FFmpeg 8.1.2 and ffprobe. Stream copy kept an identical packet hash and re-encoding did not. Default mapping dropped two audio tracks and a second video stream, but kept the subtitle track when the output was Matroska. The same -to 4 stopped at 3.0 s or 1.0 s depending on its position, and -s before -i failed. libx264 refused a 641 by 361 picture as 4:2:0 while a hardware encoder silently cropped it. Variable frame rate went unflagged by ffprobe's headline rates.

A tested section on the dithering guide

• Audio Dithering: TPDF, Noise Shaping & Bit-Depth Guide. New "Tested" section using FFmpeg 8.1.2 and numpy on computer-generated sines. FFmpeg rounds by default rather than truncating, and both leave real harmonic distortion on a quiet tone (about 20 dB below it on a three step sine, against about 53 dB with dither). A tone smaller than one step vanished when rounded and survived with TPDF dither, and a fade to -110 dBFS went to digital silence without it. FFmpeg's noise shapers cut the 2 to 4 kHz band by 16 to 23 dB but raised the A-weighted noise in all five. Two dithered passes added 3 dB.

A tested section on the de-essing guide

• De-Essing Explained. New "Tested" section on a synthetic voice, using a split-band de-esser built from FFmpeg 8.1.2's crossover and compressor because its own deesser filter has no frequency in hertz. A static EQ cut took as much off the vowels as off the s (5.9 and 6.0 dB). A de-esser spared the vowels, but only with the right settings: across sixteen combinations the vowels also lost between 14% and 83% of what the s lost. A split too low took nearly as much from the vowels as from the s, and a split above the s did nothing.

A tested section on the camera matching guide

• Matching Shots From Different Cameras. New "Tested" section, run in a numpy model of two cameras and labelled as a model, not footage. Five matching steps took a colour difference of 9.0 down to 1.4. Across all 120 orders of the same steps the result ranged from 1.0 to 5.4, so order matters, but the page's order ranked 50th and the model did not support "colour before exposure means redoing the colour". One extra stop of exposure made the colours look 7% less saturated while the vectorscope reading rose 37%.

A tested section on the proxy editing guide

• Proxy Editing: Smooth Edits on a Slow Computer. New "Tested" section: a 4K H.264 file at 100 Mbps took 871 ms to show one frame after a jump, against 46 to 68 ms for proxies, and quarter resolution proxies decoded about 30 times faster. At the same resolution a 2 Mbps H.264 proxy decoded as fast as ProRes Proxy in a fifth of the space, so the page's codec advice is qualified: the pixel count bought the speed. Measured with FFmpeg 8.1.2 on an Apple M1 Max.

A tested section on the rolling shutter guide

• Diagnosing and Fixing Rolling Shutter Artifacts. New "Tested" section, run in a numpy model of a rolling shutter and labelled as a model, not a camera. The lean of vertical lines matched pan speed times readout time, a faster shutter speed left it unchanged and only sharpened each row, and a flash lit a whole frame evenly for only 8% of flash positions at the slowest shutter with a slow readout, against 58% with a fast one. The page's advice that a faster shutter reduces skew is what this section disputes.

A tested section on the multistreaming guide

• Multistreaming: Broadcasting to Twitch, YouTube, and TikTok at Once. New "Tested" section: separate software encodes of a 1080p60 stream took 16, 54 and 73 CPU seconds for one, three and four destinations, and four could not keep up with live speed. Encoding once and copying the stream to three outputs cost the CPU of one encode but still produced 19.9 Mbps, so only a service on the far side of your connection saves upload. A hardware encoder used almost no CPU but still slowed with each output. Measured with FFmpeg 8.1.2 on an Apple M1 Max.

A tested section on the skin tones guide

• Getting Skin Tones Right. New "Tested" section: a yellow cast put five synthetic skin tones 14.7 degrees off the vectorscope skin line on average, a saturation cut left it 14.5 off, and a hue rotation centred the average but left the lightest and darkest tones 11 and 9 degrees to either side. The same rotation moved a green and a blue patch by 14 and 26 degrees, so the page's claim that it leaves everything else alone does not hold for a global rotation. Undoing the white balance error fixed all of it at once. Measured with FFmpeg 8.1.2.

A tested section on the LUT-creation guide

• Creating and Using Your Own LUTs. New "Tested" section: a LUT built with a one-stop exposure correction folded in, applied through FFmpeg 8.1.2 to footage that never needed the correction, came out 20.3% clipped to white against 0.27% for the same look built without the correction baked in. Also measured: a stronger grade needed a finer LUT grid to avoid banding that a mild grade did not, and clipped highlights a mild grade left intact.

A tested section on the noise reduction explainer

• How Noise Reduction Software Tells Noise from Signal. New "Tested" section: FFmpeg 8.1.2's afftdn denoiser measured at five reduction amounts on a recording where the speech and the hiss are known separately. The 's' sounds lost 9.8 dB at the default setting. Above 48 no more hiss went, but isolated artifacts rose from 7.9% to 28.0% of the pauses. A noise sample taken over speech removed almost nothing.

A tested section on the audio normalization explainer

• What's Actually Happening When You 'Normalize' Audio. New "Tested" section: five clips normalized by peak, by RMS and by loudness in FFmpeg 8.1.2 and measured with its EBU R128 meter. Peak normalization left a 32.6 LU spread in loudness, RMS 6.5 LU. It also records that FFmpeg's loudnorm filter changes a file's dynamics even when asked for a plain gain change, in two cases, one of them not in its documentation.

Sources for the keyframe advice in the adaptive streaming explainer

• How Livestream Bitrate Ladders and Adaptive Streaming Work. The advice on keyframe interval now says who recommends 2 seconds: YouTube (no more than 4) and Twitch. A new Sources section quotes both platforms' own encoder guidance.

Consent control restored on five DaVinci Resolve guides, and search text brought in line

• The five guides for Premiere editors moving to DaVinci Resolve were missing the consent control script, so the consent sheet could not appear on them, and advertising and analytics, which wait for consent, stayed off. It is back. The page generator had never included it, so any new guide would have shipped without it. That is fixed too.
• On four design guides (typography for screen, channel brand kits, graphics after upload, visual hierarchy), the text search engines read had not been updated when the visible text was corrected in August. It now matches what readers see.

A correction to the graphics upload guide

• Why Your Graphics Look Worse After Uploading. The answer on coloured fringes now says the worst cases are saturated red and blue, and saturated colours set against each other. It had named red alone, and our own test on the page measured blue text and red on blue as worse.

A keyframe and segment test in the adaptive streaming explainer

• How Livestream Bitrate Ladders and Adaptive Streaming Work. Now includes a test we ran ourselves. Streaming segments can only start on a keyframe, so an upload with a keyframe every 10 seconds was cut into 10-second segments even when 2 were asked for. At the same bitrate, keyframes every 4 or 10 seconds gave a better picture than every 2, and a four-rung ladder encoded faster than real time on one Mac.

A compression test in the guide to graphics that degrade on upload

• Why Your Graphics Look Worse After Uploading. Now includes a test we ran ourselves, with JPEG and H.264 standing in for a platform's encoder. Colour fringing came mostly from colour being stored at half resolution, and blue text on near-black fringed worse than red. Text that differed from its background only in hue lost up to 45 percent of its contrast at small sizes. A little noise kept gradients from banding except at the harshest setting, and oversized uploads cost 1-pixel lines up to 29 percent of their contrast, depending on the scaler.

Measurements added to the video codec and export explainers

• How Video Codecs Actually Decide What to Compress. Now includes a test we ran ourselves. At equal quality, motion prediction made a static shot about 120 times smaller and saved nothing when every frame was a new picture, and faster pans cost more because of the new picture entering the frame. A green-screen key lost 1-pixel strands at 4:2:0 and kept them at 4:4:4.
• What Happens Technically When You Hit "Export". Now includes a test we ran ourselves. An MP4 or MOV export stopped part way could not be opened at all, while fragmented MP4 and Matroska kept what had been written. With effects on, rendering took most of the time, and the hardware encoder used less processor time but barely shortened the export.

A frame rate measurement in the export settings guide

• Export Settings for Every Platform. Now includes a test we ran ourselves. We converted a moving bar between six pairs of frame rates, once by dropping or repeating frames and once by blending them. Dropping and repeating gave a regular stutter, six times a second from 24 to 30 fps. Blending gave even motion, but 75 to 88 percent of the frames held a ghosted double image. Exporting 23.976 fps footage at 24 repeated one frame every 41.7 seconds. The section also records that 8-bit and 10-bit HEVC played in every player on one Mac, and why that does not settle step 2.

A reverb measurement, and corrections to the vocal EQ and video playback guides

• Reverb and Delay Explained. Now includes a test we ran ourselves. The page's three reverb settings were measured against tempo: at 120 BPM only the drum setting had died away by the next beat. Convolution cost more processor time than a delay, but about 1 percent of one core. The example command keeps the reverb tail and the level, which FFmpeg's defaults did not. We also measured a real room, a bare 3 by 2 metre office, with a sine sweep: its decay was about 0.6 seconds in the midrange and shorter at the top.
• EQ Cheat Sheet: Best EQ Settings for Vocals. The frequency map, workflow and common moves contradicted the six steps in three places. They now agree with the steps. Step 1 no longer claims that no vocal fundamental lies below 80 Hz, and says how to set the filter on a deep voice. What You Need now names FabFilter Pro-Q 4, the version FabFilter sells now.
• Why Your Video Won't Play on Some Devices. Step 4 and the answer on the safest format now say to set the pixel format to 8-bit 4:2:0 when re-encoding to H.264. The playback test now includes Safari: it refused the same HEVC and Matroska files macOS did, and AV1, but played VP9 and ProRes.

Measurements added to the vocal EQ and video playback guides

Two more guides now include a test we ran ourselves, with the software version, the date and the method.

• EQ Cheat Sheet: Best EQ Settings for Vocals. The six moves measured alone and together in FFmpeg. The warmth shelf lifted back much of what the 80 Hz high-pass removed, leaving no cut at 100 Hz, and a bell in its place kept more of it.
• Why Your Video Won't Play on Some Devices. Ten files played through macOS, Chrome and Firefox on one Mac. The same HEVC video played in macOS only when its container tag was hvc1, and a plain re-encode of a 4:4:4 master to H.264 played everywhere except Firefox until the pixel format was set.

Measurements added to the broadcast QC and waveform guides

Two more guides now include a test we ran ourselves, with the software version, the date and the method.

• Automated QC for broadcast delivery. Six faults planted in a one-minute programme and run through FFmpeg's black, freeze and silence detectors. At their defaults they found one. With shorter minimum durations they found all six, and still flagged a title card and the closing fade that were meant to be there.
• What a waveform is actually showing you. Nine signals with waveforms of identical height measured 18.5 LU apart on a loudness meter. A 30 Hz tone read 12.1 LU quieter than a 3 kHz tone of the same height, and a clipped sine turned down 6 dB still showed flat tops.

Measurements added to the sync drift and LUFS guides

Two more guides now include a test we ran ourselves, with the software version, the date and the method.

• Fixing sync drift. How far a recorder with a fast clock drifts over 20 minutes, and two FFmpeg fixes. A clock one sample a second fast was 25.0 ms out. Correcting its sample rate removed the drift. atempo also removed it, but moved the whole track about 20 ms earlier.
• What is LUFS. A voice brought to -16 LUFS in FFmpeg. Gain alone took its true peak to +1.7 dBTP. A limiter set to -1 dB still let through -0.9 dBTP, and FFmpeg's loudnorm filter never applied a single gain.

Measurements added to the ad insertion and video verification guides

Two more guides now include a test we ran ourselves, with the software version, the date and the method.

• Podcast hosting and dynamic ad insertion. An ad joined into an episode as separate MP3 files, without decoding. Each join left 31.0 to 67.8 ms of silence, depending on which headers the files kept. Encoding once left none.
• Verifying found video. A test clip's date and location tags through four common operations. A plain stream copy dropped the date and kept the location. One command wrote a new date and place and left the video stream's checksum unchanged.

Measurements added to the drum phase and vocal compression guides

Two studio guides now include a test we ran ourselves, with the software version, the date and the method.

• Drum phase and mic placement. The guide's own 3.35 ms snare example, measured. The first notch fell at 149 Hz. Lining the close mic up in time removed it completely, and flipping polarity only moved the notches.
• Vocal compression. A consonant 6 dB above its vowel, compressed at six attack times. Attacks under 1 ms left it 2.1 to 2.9 dB above, and 5 ms left 4.7 dB.

Our own measurements added to two more guides

Two more guides now include a test we ran ourselves, with the software version, the date and the method.

• Ducking music under dialogue. A sidechain duck measured at five release times. At 250 ms the music rose 5.1 dB in each short gap between words. At 500 ms it held within 2.6 dB and still recovered fully in a one-second pause.
• Vertical video. The same 9:16 crop taken from a 1080p and a 4K master. The 1080p crop had to be enlarged 1.78 times and kept 6% of the fine detail. The 4K crop kept 93%.

Our own measurements added to three guides

Three guides now include a test we ran ourselves, with the software version, the date and the method, so you can check the figures.

• Sample rate conversion. Test tones converted from 48 kHz to 44.1 kHz three ways. At its default settings, FFmpeg's resampler left an alias 19.9 dB below a 23 kHz tone. With a longer filter the alias fell below -100 dBFS.
• What a LUT file contains. One grade saved as a 17, 33 and 65 point LUT and checked against the exact maths on 200,000 colours. Each doubling of the grid cut the average error to about a quarter.
• Log footage. The same flat gradient recorded at 8 and 10 bits, then graded. The 8-bit file broke into 34 bands, and the 10-bit file held 132 levels.

Repeated boilerplate cut from 607 pages

Several passages were written once and then placed on hundreds of pages. A reader who opened two guides saw the same paragraph twice. Those passages are gone or rewritten for each page.

• "Where This Fits". 605 guides ended with the same three sentences pointing to the main guide for their subject. Each now has one sentence of its own saying how it relates to that main guide.
• Empty "What You Need" sections. 91 history, future, case study, comparison, myth and explainer pages had a section whose only item said nothing was needed. It has been removed.
• Vocabulary. A stock opening line under the heading is gone from 99 pages. 26 entries that defined everyday words, such as "Variant" and "Foot pedal", have been removed, along with one section made up only of those.
• Nothing was added to make up the length. Some pages are now shorter, and the extra text was not what made them useful.

Broken sentences left by an old punctuation pass repaired on 148 pages

In August we removed every long dash from the site, and the pass that did it chose the same replacement every time. That left sentences a careful reader would stop on. Each one has now been read in context and repaired by hand.

• Quotation marks run into the word before them. Twelve pages had a quotation mark joined straight onto the word before it. All twelve are fixed.
• Two sentences joined by a comma. Forty-eight are now either two sentences or one with an aside in brackets.
• A description cut off as a sentence of its own. Around 140 phrases such as "The most common form." now sit back in the sentence they describe.
• The fixes are in the source files as well as the pages, so rebuilding a page keeps the repaired sentence.

245 pages taken out of search results while they are rewritten

The 182 filmmaker and pioneer profiles, and 63 short tutorials that are next in line for a rewrite, are now kept out of search engine results. They are all still on the site, linked from the same places and free to read.

• They carry no advertising. A page kept out of search results should not carry ads.
• They are out of the sitemap and the site search. The sitemap now lists 656 pages, down from 901.
• Each page comes back once it is rewritten. The list is kept in one place, and taking a page off it restores the page exactly as it was.

No advertising on pages kept out of search

Sixteen pages that are deliberately kept out of search results still loaded the advertising script: the 404 page, search, the roadmap, and thirteen section pages that are lists of links. None of them has an article of its own, so none of them should carry ads.

• The advertising script is gone from all sixteen. Visitor statistics are unchanged.
• The pages stay that way when they are rebuilt. The scripts that build the section pages and the 404 page now leave the advertising script out of any page kept out of search.

A doubled language menu removed from five pages

The contact, terms and 404 pages and the two timecode tools showed the "Translate this page" menu twice. On three of them the first copy sent readers to a translation of a different page.

• Each page now has one menu, pointing at its own translation. The copy that linked to the wrong page was the one removed.
• The cause is fixed where the pages are built. The scripts that build these pages from another page's layout now write each page its own menu, and the script that adds menus now recognises one that is already there.

Ten thin tutorials rewritten, and a wrong Spotify figure fixed

The first batch of a longer rewrite pass. Each page gained real depth, a checklist, a worked test, a concrete range, not padding, plus a proper Vocabulary section.

• Ten tutorials roughly doubled in length with new substance. A saturation guide gained per-source starting points. A field recording guide gained how to read a location before you record. A podcast consistency guide gained a reusable session checklist. Seven more, listed in WEAK_TUTORIALS_LEDGER.MD.
• A podcast loudness guide had Spotify's figure wrong in one of three places. It listed Apple Podcasts and Spotify together under one number, -16 LUFS, which is Apple's target. Spotify's own support page states -14 LUFS. Fixed, and now cited.
• The same guide stated a YouTube loudness target as fact with no source. YouTube publishes no single official number for video content. The guide now says so, matching what this site's own loudness cheat sheet already treated more carefully.

Sources moved into the course generator, and five unread links unlinked

The seven application-course overviews carried two reference sections saying similar things. One is now built by the generator from the same data the 42 modules already use, so it survives a rebuild. Two more history guides gained a Sources section.

• Course overviews had two reference sections. A bare link list from the page generator sat beside a hand-written Sources section added afterward. The generator now builds one Sources section, with a description of what each document establishes, from the course catalogue.
• Five links pointed at pages nobody had read. Two Avid marketing pages and three Adobe help pages returned an error every time they were checked, but were still linked from 30 rendered pages, six modules each for two Pro Tools courses and eighteen for the Premiere course. They are now named as plain text instead.
• The broadcast technology and television history guides gained more sources. Three IEEE Milestone records now back the broadcast technology guide, up from one, and the television guide now cites the IEEE Milestone for the 1953 colour standard.
• The unlinked style test page is gone. It carried an old colour palette on purpose and was never linked from the site. Both files it used are removed.
• 720 pages had a stale modified date in their structured data. The tool that sets it only ever filled a page that had none, so a later change to the page's own recrawl date was never copied across. It now keeps the two in sync, and this deploy corrects the gap it found.

Sources, worked examples and a sturdier page structure

Eleven history guides and seven course overviews now list their sources, six tools gained worked examples, and the explanatory text on 30 pages moved inside the main part of the page.

• Eleven history guides now list their sources. Each source was read before it was cited, and each entry says what the record establishes. Where a record could not be reached, the section says so and names what would settle the claim. Reading the records changed two guides. The microphone guide now says the condenser microphone patent was filed in 1916 and granted in 1920. The recorded sound guide now says Berliner's 1887 patent describes a drum, and his flat-disk patent followed in 1896.
• Seven course overviews now list the vendor documents behind them. Each entry says which version of the document was read. Five pages returned an error when checked, so they are named and not linked.
• The Standards page now says what each source establishes. It grew from about 250 words to about 1,250 and includes a worked comparison of the broadcast and podcast loudness targets.
• Six tools gained worked examples. The microphone polar pattern, metronome, room mode, reverb time, compressor and loudness normalizer pages each show a table computed from the tool's own formula.
• Explanatory text on 30 pages now sits inside the main part of the page. Seventeen tools, six course pages and seven other pages had it outside, so software that picks out a page's main content saw a fraction of it. Nothing changed on screen.
• Each course module's diagram caption now lists its six stages. The old caption was the same sentence on all 42 modules and said nothing about its diagram.
• The roadmap page is no longer indexed. It is a suggestion form and a wishlist. It is out of the sitemap and the footer link stays. An unlinked style test page has been removed from the published site.

Four smaller repairs from the same review

• One guide had no advertising code at all. The dialogue editing and phase alignment guide opens straight into its first numbered section, and the script placing ad units only knew how to anchor to an opening summary, so it skipped the page silently. It now has a fallback, and both that guide and one blog post have their unit.
• The search page was asking to be indexed and refusing at the same time. It was listed in the sitemap while carrying an instruction not to index it. It has been taken out of the sitemap, and there is now a check catching any page in that state.
• A renamed guide left a dead file behind. An old copyright licensing guide was replaced by a longer media licensing guide last year and redirected, but the original file stayed in place and two pages still linked to the old address. The links now go straight to the current guide, the dead file is gone, and the redirect stays so any outside link still works.
• The privacy notice no longer says adverts are not being served. It was true when written and would quietly become false, so the sentence is gone. Everything the notice says about what advertising does, and how to switch it off, is unchanged.

Every page now says who wrote it

Only the blog posts carried an author. The 665 guides and profiles, the courses, the tools and the section pages carried none at all, in any form a reader or a search engine looks for.

That was an oversight rather than a decision. We settled in August that this site publishes as a group rather than under one person's name, and the guides simply never had that applied to them. All 903 pages now name Infinite Creation as the author in the page data, and the 666 guides and profiles also show it above the article, next to the difficulty and reading time.

Nothing about who writes the site has changed. The About page still explains who we are and how the work is checked.

Every page with an affiliate link now says so

Some guides link to products on Amazon, and a few of those links earn this site a small commission. That was disclosed on nine pages and not on the rest. An audit found 150 such links spread across 74 pages, with 91 of them on 66 pages carrying no disclosure at all.

All 66 now carry a visible notice near the top, and the wording is the same everywhere: some links on the page are Amazon affiliate links, and buying through them earns us a small commission at no extra cost to you. Nothing about which products are recommended has changed, and no page was ever paid to recommend anything.

The reason it drifted is worth recording. Our own content rules have required disclosure since August, but nothing checked it, so each new page was one more chance to forget. There is now a script that finds any page with an affiliate link and no disclosure, and it runs as part of the pre-publication checklist.

The eight discipline sections now say something the main library does not

Each discipline section, such as Capture or Post-Production, was a grid of cards and about fifty words of introduction. Every card showed the same one-line summary as the card on the main tutorial library, so between half and two thirds of each section page was text you had already read elsewhere.

Two changes. The section pages now show the fuller description for each guide while the main library keeps the short one, so the two surfaces describe a guide differently instead of identically. And each section has gained around two hundred words written for it: what the disciplines in that group have in common, why they are filed together, and where to start.

Measured the same way as before, repeated text on these pages falls from between 58 and 67 percent to between 7 and 40 percent. The main library drops from 63 to 29 percent.

Two faults turned up while doing it, both in the code generating these pages rather than in the pages. Rebuilding a section page was removing the privacy control script, and it was also putting thirteen pages back into search results after they had been deliberately removed in September for being near-duplicates. Both are now produced by the generator itself, so a future rebuild cannot undo them again.

The guide counts quoted in the new writing are calculated from the library at build time rather than typed in, so they cannot go stale as pages are added.

Thirteen guides now load their illustration in the smaller format

Most pages on this site already serve their illustration as WebP, a newer image format that looks the same and downloads far less. Thirteen guides had been missed and were still handing every visitor the full-size JPEG. On a phone those pages were sending about 412 KB for the picture at the top. They now send about 70 KB.

The pages affected cover anamorphic lenses, audio mastering, matchmoving in Blender, boom operating, C-stand safety, Resolve node trees, Dolby Atmos, film set power, filmmaking contracts, shared storage, SMPTE ST 2110, virtual production and wireless lighting control.

Separately, 427 of the original JPEGs were re-encoded at a slightly lower quality setting, which removed 11 MB across the library with no visible difference. Those files are only the fallback for browsers too old for WebP, so almost nobody downloads them, but they were the bulk of what the site stores.

A new post on short films, festivals, and the odds nobody quotes

Nobody Gets Discovered From One Short Film answers a question put to us directly: keep making shorts, or wait for the one good idea. Rather than offer encouragement, we went and read the records.

Sundance selected 54 short films from 11,480 submissions for its 2026 festival, a rate of 0.47 percent. Its feature rate was 2.1 percent, so shorts are around four and a half times harder to get in with. The post works through what a submission campaign costs, including one detail from Sundance’s own guidance worth real money: every entry is handled identically whichever deadline you pick, and the only difference is the fee. A late short film entry cost $40 more than an early one and bought nothing.

The career examples come from festival and Academy records rather than from memory. Andrea Arnold made three shorts across five years before her first feature. Lynne Ramsay won a Cannes short film Jury Prize in 1996 and again in 1998 before hers.

Four things we were unable to confirm are named on the page instead of being dropped quietly, including a widely repeated production budget whose only trace is secondary posts, and an incomplete entry in the Academy’s own 2005 record.

Fuller definitions on the Premiere to Resolve guides

The vocabulary sections on the five Premiere to Resolve guides were too short to stand on their own. Several read as notes to someone who had just finished the page, such as a definition of proxy media that only said it was separate from optimised media. Each of the twenty entries now explains its term in full, so it makes sense read alone.

Nothing new is claimed. Every definition restates what the guide itself already sets out from the DaVinci Resolve 21 Reference Manual.

The fifth Premiere to Resolve guide, and the one we did not write

The series was planned with a fifth guide on what survives when a timeline is exported from Premiere and imported into Resolve. That guide has not been written, and the reason is worth stating rather than hiding.

Answering it honestly needs either testing on real projects in named versions of both applications, or the two manufacturers’ own documentation. We had neither. The testing did not happen, and Adobe’s published pages on exporting to other applications could not be read at all: every attempt to fetch them was refused by their content network. Rather than assemble a confident list from neither source, the question is left open and said to be open.

What was written instead covers the half that is fully documented, and is arguably the half that costs people more. Not what translates, but whether Resolve can find your media once it has translated. It sets out the order Resolve searches in, the two checkboxes in the import dialogue that decide which of five strategies it uses, and the one combination that matters most: cutting against proxies and conforming back to the original camera files.

Clips arriving offline is a louder problem than a missing title and a quieter one than it looks, because a clip that relinks to the wrong take appears completely correct until somebody grades it.

Two small repairs behind the new Resolve guides

• Some pages were linking to the same place several times over. When the thirty six cinema lessons were replaced by six modules, every old link was pointed at the module that now covers it. On pages that linked to three lessons from one module, that left three links going to the same page, each still labelled with the name of a lesson that no longer exists. One page had five. They are now one link each, labelled with the page they actually go to.
• The four existing DaVinci Resolve guides now say where their facts come from. They cited no Blackmagic documentation at all. Each one now names the DaVinci Resolve 21 Reference Manual and says which part of it supports what the page describes. The claims were checked against the manual before the citation was added, rather than the other way round, and the pages are explicit that opinions in them are the author’s rather than the manufacturer’s.

Four guides on moving from Premiere Pro to DaVinci Resolve

Written for someone who already edits for a living and has decided to switch, or is deciding. Neither application is explained, because that is not the problem. The problem is knowing which of your habits still work, which quietly do not, and what will cost you a week.

• The overview: what actually changes, including the parts that get worse.
• Resolve’s seven pages: which Premiere panel maps to which room, and where the mapping stops being true. Includes the limit the manual puts on keyboard emulation presets, which is that a shortcut can only be mapped to something Resolve actually has.
• There is no project file: Resolve keeps projects in a database, which changes backup, handover and recovery. This is where people lose work.
• Where Resolve is worse: titles and motion graphics. There is no equivalent of Essential Graphics, and the more powerful of Resolve’s two title tools cannot style one word differently from the rest.

Every claim about Resolve on these pages comes from the DaVinci Resolve 21 Reference Manual, downloaded and read rather than remembered, with chapter and page numbers cited on each page. A fifth guide, on what survives when a timeline is exported from one application and imported into the other, is being tested on real projects before it is written. That one is a set of claims about behaviour, and behaviour has to be watched rather than looked up.

The newsletter signup has been broken, and we did not know

The small piece of server code that receives newsletter signups was in the wrong folder, so it was never actually published with the site. Anyone who tried to subscribe got an error. That has been corrected, along with two other things that depended on the same code and were silently doing nothing.

• If you tried to subscribe and it failed, please try again. No addresses were lost, because none ever arrived.
• The same folder held the code that works out which country a reader is in, which is why the privacy notice was being shown worldwide rather than only where it is required.
• It also held the redirect from the www address to the plain one.
• One guide, on dialogue editing and phase alignment, was published without being added to the index, so it did not appear in the guide listings or in search. It does now.

The privacy notice was appearing for everyone, everywhere. It should not have been.

The notice is meant for readers in Europe, where the law requires it. Working out where a reader is was supposed to happen on Cloudflare's servers before the page was sent. Checking the live site today, that step was never running at all, and the site was doing what it was built to do when it cannot tell: assume Europe and ask everyone.

• If you are outside Europe you should not have been asked, and now you will not be. Where the server cannot say, the site now works it out from your browser's own time zone, which never leaves your device and is not stored.
• Nothing was measured or advertised to anyone who ignored the notice, which is the safe direction to fail in, but it also meant the site could not count visits at all for most people.
• The time zone list covers the same thirty two countries as before, including the ones easy to miss: Cyprus, Iceland, the Canary Islands and the Azores.

Separately, the loudness checker tool was being served a security policy meant for the small embeddable widgets, which silently blocked some of its own code. That is corrected. The two genuine embeddable widgets no longer carry any measurement or advertising code at all, which is what should always have been true of something designed to sit inside somebody else's page.

The cinema masterclass is six written modules instead of thirty six assembled ones

Yesterday's review found that the thirty six lessons in the Visual Cinema Masterclass were not really thirty six pages. Around 603 of each lesson's 830 words were word for word identical to the other thirty five, and it was the body of the page rather than the menus. Every lesson carried the same paragraphs about the inverse-square law, aperture, colour gamut and codecs, whether the lesson was about aspect ratios or about codecs. The text that did change was a single sentence pattern with the lesson title dropped into it.

• They have been replaced by six modules that were actually written. Motion and light, the camera and optics, exposure and light, colour end to end, post production and delivery, and what comes next. Around 1,100 words each, and no two of them share so much as a sentence.
• Each module has a different shape, because a course where every page is laid out identically is how the last version went wrong. One is chronological, one works through four decisions, one through six oppositions, one follows a single frame from sensor to screen, one lists trades, one weighs claims against evidence.
• Every old lesson address still works and takes you to the module that now covers its subject. Nothing is lost, and no link anywhere on the site is broken.
• The modules carry the reasoning, not the procedure. Each one points at the site's guides for the step by step work, because a guide about which button to press goes out of date and the reasons things behave as they do do not.
• The syllabus has been rewritten to match, and it says plainly what the course does not cover: sound, directing, and what to buy.

Measured before publishing: each new module shares under 3 per cent of its phrasing with anything else on the site, against 88.6 per cent for the pages it replaces.

Forty-nine pages taken out of search, because they were saying the same thing

A review of the whole site found 58 pages that were mostly made of text already published elsewhere on it. The writing itself was not the problem: measured without JavaScript, the guides run to a median of 852 words, the biographies 863, and the blog 1,341, with almost nothing overlapping between them. The problem was pages built by machine from a template.

• The 36 cinema masterclass lessons were the worst of it. Around 603 of each lesson's 830 words were word for word identical to the other 35. Every lesson carried the same paragraphs about the inverse-square law, aperture, colour gamut and codecs, whether the lesson was about aspect ratios or about codecs. They have been removed from search while they are either rewritten properly or withdrawn.
• Thirteen category pages were near-copies of other category pages. The site had grown three overlapping ways of listing the same guides, so "History" was 91% the same page as "Media history". One way of listing them is enough. The pages still work and still link where they always did. They are simply no longer offered to search engines as separate reading.
• Nothing was deleted and no link was broken.

Advertising code is back on the site after being switched off yesterday, and it now goes through the same consent panel as everything else: nothing is requested from Google's advertising servers unless you have allowed it. No advert is being served either way at the moment, because the advertising account is not approved.

The cookie banner is ours now, and it asks a straight question

Until today the consent notice was run by an outside company. It loaded three files from three of their servers before the page could settle, it reported back to them, and we could not read a line of what it did. It has been replaced by something written for this site, served from this site, about a tenth the size.

• Reject and Accept look exactly the same. Same size, same colour, same weight. Neither one is the easy button, because a choice where refusing is made to feel harder is not really a choice.
• Three switches instead of one decision: measurement, advertising, and the embedded video player. You can allow one and refuse the others.
• Refusing now genuinely sends nothing. We tested this rather than assuming it. Simply telling Google not to store anything still let it send a message on every page, so the tracking code is no longer loaded at all until you agree.
• Changing your mind actually undoes it. Turning measurement back off deletes the identifiers already in your browser, instead of just quietly stopping new ones.
• A "Privacy choices" link is now in the footer of every page. There was no way to revisit this before. If you are outside Europe you were never asked at all, and everything was on by default, so this is the first time that has been yours to change.
• The video player waits for you. On guides with a companion video, nothing is requested from YouTube until you allow it, and there is a link to watch on YouTube instead if you would rather not.
• One notice at a time. The translation offer and the privacy notice used to want the same corner of the screen. The translation offer now waits its turn.

Two side effects worth naming. Every page footer is now identical, which fixed 109 pages that linked to no policy page at all. And the site's last remaining accessibility complaint came from the old notice itself, so it went with it.

The privacy policy now matches what the site actually does

Four things in it had drifted out of date, and one had never been written down at all.

• The embedded video was never mentioned. Forty-eight guides load a YouTube player, and the policy did not say so anywhere. It does now, including the part people care about: loading the player tells Google you are on the page, even if you never press play.
• Advertising cookies. The policy described adverts setting cookies on you. Advertising is switched off, so the honest answer is none, and this section will change before any of that comes back.
• YouTube was missing from the list of outside companies involved in running this site, despite serving video on 48 pages.
• A stored item that no longer exists was still listed. It was left behind when the old privacy notice was removed.

The tutorial videos no longer come with a tracker attached

Forty-eight guides carry a companion video. The player was loaded in a way that let Google set tracking cookies on you as soon as the video came into view, whether or not you ever pressed play.

• The player now uses YouTube’s privacy-enhanced mode. Nothing persistent is stored on your device unless you actually start watching.
• The video still appears in the same place and plays the same way. Nothing to click that you did not have to click before.
• This is a reduction, not a cure. Loading the player still tells Google you are on the page. Putting that behind a proper choice is the next piece of work.

Advertising is switched off for now

The advertising code has been taken off every page. It was loading a script from Google on any page you touched or scrolled, and because the advertising account is not approved, that script has never actually shown you an advert. So it was a request made about you, to a company you did not choose, in exchange for nothing.

• No page now contacts Google’s advertising servers at all, whatever you do on it.
• Nothing looks different. The advert spaces were already empty, so no layout has moved.
• This is deliberately temporary and reversible. If advertising is approved and switched back on, it will be said here first.

The site’s fonts no longer tell Google you were here

Every page on this site fetched its two typefaces from Google, and it started that connection before it had finished reading its own instructions. That handed your IP address and browser details to Google on every page you opened, whether or not you ever agreed to anything, and there was no way to decline it. The privacy policy argued this was allowed. A better answer was to stop doing it.

• Inter and Roboto are now served from this site’s own servers. Opening a page makes no request to Google Fonts at all, so nothing about you is sent there for this purpose.
• Nothing looks different. The same two typefaces, the same weights, and the same coverage for other alphabets: the language list in the footer still renders Russian and Vietnamese names properly, and names carrying Polish or other accents are unaffected.
• Pages also make two fewer connections to another company before they can draw anything, which makes them start a little sooner.
• The privacy policy has been corrected. It previously described this transfer to Google and defended it. That section now says what actually happens, and Google Fonts has been removed from the list of third parties this site uses.

Five guides now let you try the thing instead of reading about it

Measured across the guides, the only thing a reader could do was scroll. The loudness guide is seven screens long and had no control on it anywhere, while the loudness tool one click away is full of them.

• The podcast loudness guide, the normalisation explainer and the “louder is better” myth page now carry a small live panel where the text discusses targets: pick a delivery format, see the loudness and true peak figures change. The two caption guides carry a readability panel: type a caption, set how long it is on screen, and watch the reading speed and line length update against the limits the page just described.
• These use the same two widgets already offered for embedding on other people’s sites, so the numbers have one home and cannot drift apart from the article around them.
• This covers five guides, not all of them. Only two of the site’s tools have an embeddable form; the rest would each need one built.

The biographies stopped pretending to be instruction manuals

All 182 profiles used the layout built for a step-by-step guide. A life story was headed “What You Need” above the subject’s birth year, that fact list sat below the in-article advert several screens down, and the page offered “~17 min” as a reading estimate while competing with an encyclopaedia entry.

• Profiles now open with “At a Glance”: the summary paragraph and the facts together, at the top. Someone arriving from a search for a birth year or a notable work sees the answer first rather than after scrolling past an advert.
• The remaining sections are named for what they hold: “Career” for the life, “Known For” for the work. The reading estimate is gone; the page just says “Profile”.
• Not a word was rewritten. The facts were already on every page, in the wrong place under the wrong heading. The page builder was changed first, so newly built profiles come out this way rather than reverting.

111 pages offered related reading and then showed none

Every pioneer profile, and the broadcast technology guide they sit under, ended with the heading “Related Tools & Guides” and nothing beneath it. A reader who finished Alan Blumlein or Alice Guy-Blaché was shown an empty section and then the sources box.

• The links had been written. They were held in the site’s own registry, but the page builder only ever read a second, separate field, so nothing authored in the registry reached the page. That is now fixed in the builder, which is why it will not happen again to a new page.
• Each profile now ends with four places to go next: the history guide for their subject, the two figures closest to them in time working on the same thing, and the guide that frames the whole chain. Blumlein leads to the history of the microphone, Harvey Fletcher and Georg Neumann. Wendy Carlos leads to the history of the synthesizer, Don Buchla and Robert Moog.
• No new pages and nothing newly written. Every link points somewhere that already existed.

The tool that does the job now sits next to the step that describes it

Forty-seven guides explain a task the site already has a browser tool for, but the only link to that tool sat in the related list at the very bottom of the page. The link now appears inline, directly under the step it belongs to.

• A guide to AI tools offers the Background Noise Remover under the step about denoising, not eleven screens later. The LUT guide offers the LUT Previewer under the step explaining why one LUT looks wrong on different footage. Placement is chosen per page from what each step actually discusses.
• Two rounds of review removed the matches that did not earn their place, including a sun position calculator on a film editor’s biography and a contrast checker under a section about traffic sources.
• No new pages, and no new wording in the guides themselves. Every link points to a tool that was already published.

Every page reserved about a thousand pixels it never used

The small link cards at the foot of a page shared a rule written for the much taller tutorial cards, so each one held open a 320 pixel gap until you scrolled to it, then collapsed to 56.

• Across 724 pages carrying 3,058 of these cards, that was roughly a thousand pixels of phantom height per page, and the page shifted under the reader as it closed up.
• The cards now reserve their real height. Measured at desktop and phone widths, the space held open and the space used are the same, so nothing moves.

Searching for a short word was hiding most of the answers

Reported by a reader: searching “time” returned three results, while typing one more letter, “timec”, returned eleven. A shorter, more general word should never find less.

• The site runs two searches. One understands meaning, the other matches words. The first was allowed to answer alone whenever it found anything at all, and it is deliberately strict, so a vague word like “time” passed its test only three times. Meanwhile the word-matching search had seventy-two pages ready, because “time” sits inside timecode, time-lapse and real-time.
• The two are now combined. The meaning-based results still come first, and everything the word match found is listed behind them. Searching “time” now returns 73 results instead of 3, with the timecode tools at the top.
• Fixed alongside it: typing quickly could show results for part of what you had typed, because replies arrived out of order and the older one landed last. Replies are now numbered and a stale one is discarded.

A tool that burns timecode into the picture

The site could already generate timecode as audio. It could not generate it as something you can see, which is what most people mean when they look for a timecode generator. The new Burnt-In Timecode Generator makes a test clip with the number running in the corner of the frame, or a single frame as a transparent image.

• Drop frame is handled properly, including the frame labels it skips. The tool refuses a start timecode that cannot exist, and says which one is the first valid frame of that minute.
• Nothing is uploaded. The clip is drawn and recorded inside your own browser.
• The page is honest about a limit rather than quiet about it. A browser records in real time, so frames can be dropped and the finished file's frame rate is a request rather than a promise. Our own test asked for 25 and measured 24.96. The numbers on each frame are always right; the spacing between them is approximate, so the clip is for checking and demonstrating, not for conforming against. The single frame has no such problem.

European visitors can now actually choose, and the choice is honoured

Until today a reader in Europe had no way to agree to anything, so nothing was measured about their visit and no advertising could be personalised. A certified consent tool now runs on every page, first thing, before any other code.

• The banner offers Reject, Accept, and a customise option that lets each purpose be set separately. Checked from an actual European connection: choosing Accept sets the analytics cookies and records the agreement, choosing Reject sets nothing at all and sends no measurement, and the choice can be changed later.
• The old notice offered only a Dismiss button, which is not a choice. It was removed earlier this week and this replaces it properly.
• Readers outside Europe are unaffected. No banner applies there and measurement continues as before.

Three pages gave three different answers to how many tools there are

The tools page said 91, the homepage said 108, and there were 109. None of the three was right, and one tool was listed twice.

• The tools page counter was never wrong by accident. It is calculated by the page, but the calculation only ran when somebody typed in the search box, so the number written into the page by hand survived until then. It now runs on load, which means it cannot go stale again whatever the markup says.
• The caption checker had two entries, one describing the tool as it was before it learned to read subtitle files. The out-of-date one is gone.
• All three pages now take their figures from one count of what actually exists, so they cannot drift apart again.

One consent notice instead of two, and the old one is gone

A certified consent tool is now running on the site, which is what advertising in Europe requires. The old notice, the one with a single Dismiss button, was still appearing alongside it. European readers were being asked twice, once by something that offered no actual choice.

• The old notice has been removed from every page and its script deleted. Europe and the UK get the certified tool, which offers a real choice and lets it be changed later. Everywhere else, the privacy policy in the footer is the right place for that disclosure and always was.
• Checked from an actual European connection: nothing is stored or personalised before consent is given, and the site now sends that signal on every page before any Google code runs.
• Caught while removing it: the offline cache listed the deleted script among the files it must fetch, and that list fails as a whole if any single file is missing. Left alone, offline support would have quietly stopped working for everyone.

European visitors were being advertised to without being asked

Roughly one in six people who find this site through search is in Europe or the UK. Advertising was loading for them without consent, and the notice at the bottom of the page offered only a Dismiss button, which is not a choice.

• Every page now tells Google, before any of its code runs, that nothing may be stored or personalised for visitors in the European Economic Area, the United Kingdom or Switzerland until they have actually agreed. Outside those places nothing changes, because no consent banner applies there.
• The Dismiss button was deliberately left unwired. Dismissing a notice is not agreement, and treating it as agreement would invent a permission nobody gave.
• This is the groundwork rather than the whole fix. Google requires a certified consent tool for European advertising, and that is a setting on the advertising account rather than something that lives in the site. The site is now ready for it.

Seventy-one pages had a structural fault that browsers were quietly repairing

Seventy pages opened the element that marks the main article and never closed it. Three more opened it twice. Browsers patch both faults silently, so the pages looked correct to anyone reading them.

• Screen readers were the ones paying for it. On those pages the element ran to the bottom of the document, so the footer, the site navigation and the translation picker were all announced as part of the article.
• It also blinded some of our own quality checks. Anything that looks for the article by its opening and closing markers found nothing on those pages and moved on, which is how the fault survived this long.
• The check that should have caught it did not exist. It does now, and it was tested against pages carrying each fault before being trusted.
• The image credits page was building its own article wrapper and then being handed to a template that adds one too. The generator was fixed rather than the page, so it cannot come back on the next rebuild.

Every wrong address on this site was quietly answering with the homepage

Typing an address that does not exist returned the front page, with a success code rather than an error. Not a redirect and not a message, just the homepage pretending to be whatever you asked for. The cause is a default: Cloudflare only serves a proper error page if the site provides one, and this site never had.

• There is now a page for addresses that do not exist. It searches the library and offers the usual starting points, rather than apologising.
• Google had already indexed twelve addresses that were never real pages, and was reading the homepage under each of them. Those now redirect to the course they belong to.
• The side effect that matters most: a wrong link on this site could never be detected, because every wrong link appeared to work. That is no longer true.

A timecode converter, because the calculator could not convert between frame rates

The Timecode Calculator adds and subtracts timecodes at one frame rate. It cannot answer the question people ask most often, which is what a timecode becomes at a different rate. The new Timecode Frame Rate Converter does that.

• It gives two answers, because there are two and they are not interchangeable. Keeping the same frame count relabels the material and changes its duration, which is what you want when interpreting a 23.976 file as 24. Keeping the same instant preserves the moment on a clock and recounts the frames, which is what you want when syncing to audio or a second camera. Choosing the wrong one is how a conform drifts.
• Drop frame is handled properly, including the labels it skips. The arithmetic was checked against published values before the page shipped: one hour of drop frame timecode is 107,892 frames, not 108,000, and the frame after 00:00:59;29 is labelled 00:01:00;02.
• A table shows the same instant at all ten common rates at once, which makes the drift visible: one real hour at 29.97 non-drop reads 00:59:56:12.

Two tools rebuilt around what people are actually searching for, and a correction to the safe-area figures

Search Console data showed that the safe-zone and lower-third tools sit in the site's best-converting clusters, and that the searches ask for numbers and files rather than for reading. Both tools now answer that.

• The Safe Zone Overlay Generator was named a generator and produced nothing. It now exports a transparent PNG overlay at the full resolution you choose, ready to drop onto a timeline as a guide layer, at any size from a phone clip to 16K.
• Correction. Its safe-area figures were wrong and disagreed with its own diagram. The page said action safe 90% and title safe 80%. The standard, EBU R 95, sets action safe at a 3.5% inset and graphics safe at 5%, which is 93% and 90%. The 80% figure is the old 4:3 television convention. The standard is now read directly and cited on the page, and the arithmetic is checked against its own published erratum.
• The Lower Third Template Builder gave its specifications as percentages of the frame. It now converts them to pixels for whichever resolution you pick, and exports a placement guide showing the band, the text baselines, and the safe area it has to sit inside. It also says, per style and per resolution, whether the design crosses that line.

Advertising was missing from 151 pages, and one course never counted a short visit

Two faults found by rendering the site the way an ad network sees it rather than by reading the code.

• 151 written pages carried no advertising slot at all, including every guide published this week and the most recent blog post. The script that places the slot is a retrofit, and the page generator never produced one, so everything built since that script last ran shipped without it. Both halves are fixed, so it cannot drift again.
• The 36 lessons of the Visual Cinema Masterclass and its overview page were still deferring analytics along with advertising. A reader who arrived, found their answer and left within a couple of seconds was never counted. Measured on a visit with no interaction: nothing was recorded before, and the pageview now registers in about 30 milliseconds. The numbers for those pages have been an undercount, and will step up.
• The interactive tools remain ad-free, as the privacy notice says.

The tutorials page was eighty-five screens long

Every one of the 658 guides was listed on one page, which made it 76,300 pixels tall. It is the second item in the main menu, so it is the page most first-time visitors open, and what they found was eighty-five screens of cards above four hundred words of writing.

• The list now starts at 48 guides, with buttons for the next 48 or all of them. Eight screens instead of eighty-five.
• Filtering and search still work across all 658. The limit applies to whatever the filters match and resets when the filters change, so narrowing after expanding does not leave a stale count.
• The measurement that decided the approach: the page is 534 KB of markup but only 74 KB once compressed, because repetitive cards compress well. Loading the list separately would have sent more data, not less, so the cards stay in the page and only the height changed.
• Fifteen guides published this week were reachable only from that one page. They now appear on their subject hubs as well.
• The homepage advertised 570 guides and 92 tools while the site held 658 and 108. Those numbers are now read from the catalogue.

The site search had not learned about the newest guides, and four other things behind it

Yesterday's fifteen guides went live and the site's own semantic search could not find them. Keyword search picked them up because that index ships with the pages. The semantic index lives elsewhere and nothing updated it, so searching a new guide by its exact title returned something else entirely. Fixed, along with four faults found while checking it.

• Search now finds every page. The step that refreshes the semantic index was missing from the release checklist, and the command recorded for it pointed at an address that accepts the request and indexes nothing.
• Search was also being taught the wrong text. It learned each guide from the one-line summary on its card rather than from the full description, so it knew a fraction of what each page is about. A search for music theory returned a mixing guide even after the new page existed. It now returns the music theory guide.
• The tutorials page said 692 guides. There are 658. Two scripts were both writing that page and one of them had been adding cards below the end of the grid, where the filters and the search box cannot reach them, for weeks. Thirty-four duplicates removed, and one card pointing at a page that has redirected elsewhere since it was merged.
• Thirteen guides listed related reading that never appeared on the page. The links were written and the page never rendered them, so finishing one of those guides was a dead end.
• The RSS feed had stopped on 7 August, seven posts and ninety-one updates behind. It is now built from this changelog and the blog index rather than kept by hand.
• Security: the address that writes the search index accepted writes from anyone. It now requires a key, and refuses every write if that key is not configured rather than falling back to open. Search itself is unchanged and stays open.
• The search index can now have entries removed. It could only ever be added to, so every renamed or deleted page had been leaving a permanent search result behind pointing at an address that no longer exists.

What 142 game audio job postings actually asked for, and why there is only one page like it

A plan existed for a whole strand of pages answering what work exists in media and what employers ask for. The first step was a deliberate go or no go: find out whether counted job-postings data exists across enough disciplines to fill a series. It does not, and the honest outcome is one page rather than a shell promising a series that cannot be filled.

• Game Audio Jobs in 2026 reads a systematic analysis of 142 postings collected between April and July 2026 across roughly 90 studios in 26 countries. Entry level was 9 postings out of 142, and all but one was outside the United States. That is the finding, and it is not softened.
• Every figure carries its denominator. Two are qualified on the page because the source qualifies them: tool percentages count a mention anywhere in a posting including a nice to have, and the salary figures come almost entirely from two US states whose law requires a posted range.
• A whole section says what is not covered and why. Video, post-production, broadcast, podcasting and photography have no comparable public counted analysis, and the two sources most likely to hold one both refuse automated requests, which is a statement about access rather than about existence.
• The blog post that named this gap in its closing paragraph now points at the page. One sentence was added and none of the author's own were changed.
• Twenty existing guides gained links to the fifteen pages written this week, because a new page otherwise ships with two inbound links and nothing that already covers the subject pointing at it.

Six crafts had the pages but no route through them

The eight learning paths are a real curriculum. Nothing routed a beginner through photography, journalism, motion graphics, design, music production or the business of freelancing, although photography had thirteen guides and journalism fourteen. The material existed. The order to read it in did not.

• Six new routes on Learning Paths, six guides each, sequenced by what the work depends on rather than by difficulty. They are routes and not courses, and they say so: no capstone is claimed, and the reading time is added up from the guides themselves rather than estimated.
• Freelancing had a rate calculator and a contracts guide and nothing between them. Three new guides close it: proposals and scoping, invoicing and getting paid, and finding and keeping clients.
• Two subjects had twelve guides each and no beginner entry point, which is a reachability problem rather than a depth one. Your First Media Feature and How a Broadcast Operation Differs each start from someone who has not done it before.
• None of the freelancing guides quotes a tax rate, a statutory interest rate or a market rate. All three differ by territory and by year, and a figure nobody re-checks is worse than none. Where a number decides something, the page names the authority that publishes it.

Music production is now a subject, not a shelf of composer biographies

The library covered everything that happens to a piece of music once it exists: mixing, mastering, synthesis, quantisation, pitch correction, drum recording. It covered almost nothing about where the music comes from. A reader searching for music theory was answered with a mixing guide, because a search that always answers hides the gap it is covering for.

• Nine new guides covering the writing half of the work: music theory without notation, songwriting structure, chord progressions, arrangement, drum programming, sampling technique, bass and low end, and releasing music independently.
• Making a Record is the route through them, in the order the work actually happens. It exists because problems created while writing cannot be solved while mixing, and knowing which stage a problem belongs to is most of the skill.
• These guides are filed under a new Music Production subject rather than under Music & Scoring. Scoring is writing music to picture, and filing a songwriting guide under it pointed readers at a page about spotting and stem delivery.
• The release guide describes the ISRC format from the standard’s own documentation, because several widely repeated summaries of it describe the code incorrectly.
• Fixed underneath: the script that rebuilds the tutorial catalogue was copying editorial review dates over the site map’s change dates, which moved 589 pages backwards in one run and told search engines the site had un-changed. It now only ever moves a date forward.

Fifty-four guides were live but missing from the catalogue, so the site could not find them

A review of the whole library found that the recent run of expansions had outgrown the machinery behind it. Fifty-four tutorial pages, forty thousand words, were published and reachable but absent from every subject hub, because those hubs are built from a catalogue the pages had never been added to.

• All fifty-four are now catalogued and appear under their subject: Nuendo game audio, Unreal virtual production, live sound mixing, Dolby Atmos, Steadicam operating, Illustrator, Affinity Designer, Cavalry, Photoshop, Pro Tools, drum recording, vocal compression and more.
• Thirty-three links pointed at pages that were never published, mostly a guide linking to the address its author expected rather than the one that shipped. Each was repointed by reading the subject rather than by matching the text, because automatic matching proposed sending readers from camera movement to audio mastering.
• Fifteen calculators explained too little to be useful on their own. Each now carries a worked-example table covering every combination the tool offers, with the figures calculated by the same arithmetic the tool itself uses.
• Fifteen guides explained a technique and stopped before the part readers need most, which is what goes wrong and how to recognise it. Each now names its real failure modes.
• A fault in the tooling surfaced during the repair and is fixed at source: the script that adds each guide's link to its parent subject only recognised one of the two heading styles in use, so it had been adding a second copy to pages that already had one, and quietly leaving eight pointing at the wrong parent.

The opening paragraph on thirty-eight pages was falling back to plain body text

The cinema masterclass pages and one blog post use the same markup the blog does, but they were loading only the main stylesheet and not the one that styles it. The header paragraph on every one of them lost its width limit, its softened colour and its line spacing, so it ran the full width of the screen in ordinary body type.

• Fixed in the generator that writes thirty-seven of them, so a rebuild keeps the fix. The blog post has no generator that owns its head and was corrected in place.
• Three further pages use the same paragraph style: the pioneers A to Z, the technology timeline and the image credits page. Those deliberately carry their stylesheet inside the page rather than linking to it, because this site has already shipped a bug where the markup arrived before its styles did. They get the same three rules added to the block they already carry.
• Rebuilding exposed two things the generators had been quietly dropping. The masterclass overview page emitted its course data but not its breadcrumb trail, and the image credits page passed none at all, so both lost navigation data that a later pass had added by hand. Both generators now emit it.

Fifty-one links carried a name the page they point to no longer uses

Checking one guide's title against the rest of the site turned up the same problem elsewhere. When a guide is renamed, the links to it from other guides keep the old name, so a reader clicks a title and lands on an article that appears to be something else.

• Six guides had been renamed and left stale links behind, forty-seven in total. The worst was a guide on equalisation, where eleven links still advertised a title it stopped using.
• One guide had four different names in circulation across sixteen links, two of them differing only in the British and American spelling of colour.
• One link had a corrupted name: a Polish film school rendered as three nonsense words by a character-encoding fault. The other three links to that page were correct, so it was a single bad write rather than something systemic. Nothing else on the site shows the same corruption.
• Three links used a curly apostrophe where the site uses the straight one everywhere else, including in the titles they were quoting.
• Shortened link names are deliberate and were left alone. A card is one line on a phone, so "Reading a Delivery Spec" pointing at "Reading and Meeting a Delivery Specification" is the convention working. Only names that said something the page does not were changed.

One guide had two names, depending on where you found it

The synthesizer history guide called itself one thing on the page and another everywhere else. Every link to it from other guides, the history index, the search results and the changelog said "A History of the Synthesizer". The page itself said "History of 20th Century Synthesizers: Electro-Mechanical to MIDI". A reader following a link arrived somewhere that appeared to be a different article.

• The page now uses the shorter name, which matches the thirty other history guides and every link pointing at it.
• The longer descriptive form is not lost. It has moved to the field this site already uses for the version that appears in a search result, so the search listing still says what period the guide covers. That form was also two characters over the length a search result displays, so it was being cut off anyway.
• Its structured data carried yet a third variant, longer still. All three now agree.
• The one-shot script that wrote the old title is kept for the record and marked as not to be re-run, since it would also reintroduce a card layout the grid stopped using.

A check that had been failing quietly, and twenty pages that were missing a link home

Every guide on this site is meant to carry a short section pointing at the deeper guide for its subject, so a reader landing on one specific page can find the one that frames the whole discipline. Twenty pages did not have it, including eight published this week.

• The check for this exists and reports a failure, but it was not in the pre-publication list, so nothing was running it. Nothing else would have caught this: a page missing its link home is valid in every other respect and reads perfectly well.
• All twenty now carry the section, and the check is in the release list and in the repository's own instructions.
• The twelve newest pioneer profiles were also missing their review date, so they would never have appeared in the report that says which pages are due for re-checking. Fixed in the record and in the script that writes it, so the next batch carries one.

Two prices were still in pounds where the site says dollars

A conversion in August moved every price on the site to US dollars, per the site's own style rule, but it reached the page bodies and not the catalogue entries or the description tags. So two guides quoted dollars in their text and pounds in the description search engines show.

• The voice over booth guide and the budget lighting guide are now consistent. The figures were taken from the converted page bodies rather than converted a second time, so the two agree.
• Nineteen more cards on the tutorials grid disagreed with the catalogue in wording, mostly punctuation from the recent sweep. Those are now in step. Two remain, on the synthesizer history guide, which uses one title on the page and a different one everywhere else on the site, and which is an editorial call rather than a mechanical fix.

The REAPER custom actions guide now tells you how, not only why

The page explained what custom actions are and which sequences are worth chaining, and stopped short of the setup. A reader arriving from another DAW and wanting to build one had to go elsewhere for the mechanics. It has gone from 781 words to 2,326, and everything added was read in REAPER's own user guide before it was written.

• The worked example the manual itself uses, three commands chained so one key splits whatever sits under the pointer, with the exact command names.
• Consolidate undo points, which is the single setting people miss. Without it one keystroke leaves ten entries in the undo history and one undo reverses a fraction of the work.
• The action list sections. A key assigned in Main does nothing in the MIDI editor, which is the first thing to look wrong to anyone arriving from a DAW with one global key set, and the global scope setting that resolves it.
• A macOS section. Which substitutions to make when reading Windows-first documentation, why a binding sometimes does nothing at all, and where the configuration lives.
• Backing the set up, putting a macro on a toolbar, arming it so it acts where you click rather than where the button is, looping a macro with the two meta actions, boolean search in the filter box, and where scripting begins.
• A Sources section, which the page did not have. It names the guide, its version and date, and the specific sections each claim came from.

Three faults found while extending one page

Rebuilding a page means running its generator, and running this one showed it had been quietly losing work for a while.

• The generator for this seven-page series emitted the old advertising loader, no breadcrumb data, and none of the long-form section every page in the series carries. All three had been added to the rendered pages afterwards, so every rebuild would have removed them. The generator owns all three now, and the existing wording was moved into it unchanged rather than rewritten.
• The series had a house-style gate whose own test counted HTML entities as semicolons, because an escaped apostrophe ends in one. It was reporting violations on pages with none.
• The script that keeps the tutorials grid in step with the catalogue was only ever checking twenty of 590 cards. It matched one spelling of the separator in a card footer and 569 cards use the other. It now checks every card, and reports a card whose title or description disagrees with the catalogue rather than overwriting it, because the card is sometimes the corrected copy: one card reads six dollars where the catalogue still says five pounds.

The two illustrations that came back with invented lettering have been replaced

Both were generated in August and both failed the same way: too much lettering asked for in one frame. The synthesiser illustration carried four blocks of caption text that was not language. The film dealmaking illustration had a misspelled poster, mirrored text on the documents, and a signature belonging to a person who does not exist.

• Both are regenerated with a stricter limit of two pieces of lettering in the frame, an explicit instruction that no other surface carries writing, and a ban on a signature. Everything else in both pictures now shows ruled lines, colour bars and blank columns.
• The dealmaking frame was generated twice. The first clean version came back as three men, because the prompt asked only for a mix of ethnicities. The prompt names the gender mix now as well, and so does the script, so the next person to run it gets the same result.
• The alt text on both pages described the old pictures and has been rewritten to describe what is actually in the frame, which is what this site's own guidance on alt text asks for.
• A fault in the image tooling surfaced while shipping these. Replacing an illustration rewrote the JPEG and left the WebP versions untouched, and the WebP is what a browser actually loads, so the pages would have kept serving the old pictures behind a corrected fallback. The optimiser now treats a WebP older than its JPEG as stale rather than as done.

The rest of the damaged prose, recovered from the modules the pages are built from

Clearing the em dash out of the structured data fixed every passage whose intact wording had survived there. Only the steps, the description and the FAQ reach structured data, so everything in an overview, a tip or a Knowledge Base section was outside that reach. For those, the generator module was the only intact copy, and it still held 1,951 em dashes.

• Every one of them is now decided per occurrence by the same rule as before, and written to the module and, where the August substitution had damaged the published sentence, to the page as well. What is left in the modules is 46 marks, all of them in a module's own description of itself or in the single-character placeholder the tools use for an empty value.
• The page prose was checked passage by passage against the deployed site rather than against a local copy. Every change is either a colon or comma replacing a full stop that had split a clause in half, or a capital restored to a sentence that had lost one.
• Two faults in the repair itself were caught and undone before they shipped. A first pass edited every module rather than only those carrying page prose, and rewrote a character class inside the site's own style checker, which quietly disabled its em dash test. A second attempt at capitalising sentence starts was written as a pattern and capitalised a lower-case brand name. Both were reverted against the deployed pages and redone precisely.
• The August script that caused all of this is kept, marked superseded, with its faulty rule and the damage it did recorded at the top of the file. Its scope notes still stand: empty-value markers, layout dashes in tables, and accessible names are deliberately not prose and are not touched.

The last sweep of a banned punctuation mark broke the sentences it touched, and the broken versions were live

House style bans the em dash. A pass in August replaced every one in the visible text with a full stop and never reached the structured data, so 116 pages ended up holding two versions of the same sentence: an intact one in the JSON-LD and a damaged one on the page. "Louis Le Prince shot moving-picture sequences in Leeds in 1888, the earliest that survive, and then vanished from a train in 1890" had become three sentences, one of them "And then vanished from a train in 1890." with no subject.

• The structured data turned out to be the only surviving copy of the original wording, so it was used to repair the prose rather than swept to match it.
• Punctuation is now chosen per occurrence instead of by blanket substitution. A pair of dashes bounding an aside becomes a pair of commas. A dash before a conjunction or a participle becomes a comma. A dash before an expansion or a list becomes a colon. Only a clause with its own subject and verb is allowed to become a new sentence, which is the check the last pass did not make.
• 407 occurrences across 116 pages, in the structured data, in the visible prose, in the description tags, and in the generators and data files those pages are built from. That last part is the difference between this and the previous three attempts, all of which were undone by the next rebuild.
• A further 16 catalogue descriptions carried one, which appear on the tutorials grid and on every discipline hub but in no structured data, so nothing had flagged them. Six timeline headings and ten page descriptions were repaired the same way.
• What is deliberately left: the dash used as an empty-value marker in the tools, which is a symbol rather than punctuation, and the dashes inside interface labels. Both are interface, not prose.

The blog index was not in date order

Three of the fourteen posts sat out of sequence, because new posts had been inserted next to a neighbour rather than at the top. The most recent post was third.

• The index now sorts on the date each card displays, newest first, and the script that does it is safe to re-run after every new post.
• The dates on the cards and the publication dates recorded in each post's structured data already agreed everywhere both exist, so nothing else needed reconciling.

Three subjects the site kept referring to and never explained

A review in August found six content gaps. Three were written earlier this month. These are the last three, and each sits next to material the site already had: delivery specifications, a release-form tool, and thirteen journalism guides.

• Sustainable production and albert certification. What the four phases actually ask for, why it behaves like a delivery requirement rather than a policy, and why the reductions that matter are decided by the schedule in pre-production rather than bought later. albert's own certification page and calculator were read, along with one broadcaster's producer guidance stating the requirement in its own words.
• Data protection for media producers. Why a signed release form is not a lawful basis, which article covers a documentary, why the exemption for companies under 250 people almost never applies to anyone who films people for a living, and why the transcription service somebody signed up for mid-shoot is the real exposure. Every article quoted was read in the official text at EUR-Lex.
• A verification method. The site had verification techniques and a page on Content Credentials, and nothing on running a verification as a recorded procedure. This follows the Berkeley Protocol's split into source, technical and content analysis, compared against each other rather than blurred into one impression, and is explicit that a cryptographic provenance check is a statement about a file rather than about the world.
• Each page says what could not be checked. The albert star thresholds are not quoted, because the criteria paper carrying them no longer resolves at either albert domain. The national journalism derogations under GDPR article 85 are not enumerated, because they are 27 separate national laws and none were read.

Two guides had placeholder text on them, and site search had no tutorials in it

Building the three guides above through the proper generator surfaced four faults in what was already published. All four came from the same habit: writing a page or a sweep directly against the rendered HTML instead of against the thing that generates it.

• The accessibility and podcast guides published earlier this month were rendered outside the standard build, and shipped with the generator's own section placeholder visible on the page as a stray letter over a stray letter. Both now carry the sections that were meant to be there, and both are built the same way as every other guide.
• Neither of those two pages had a catalogue record, so they were absent from their hubs and from search. Comparing the catalogue against the pages on disk found two more in the same state, both history guides live since August. All four are registered, and the script that adds a card to the tutorials grid no longer takes a hard-coded list of slugs: it works from the catalogue, and reports any card whose filter attributes have drifted from it. Seventeen had.
• Site search was returning no tutorials at all. The script that builds the search catalogue matched a card attribute that stopped being the last one when the topic filters were added, so every tutorial silently dropped out. The catalogue went from 258 entries to 851. Blog posts are now read from the blog page rather than from a list maintained by hand, which had missed three.
• Rebuilding the hub pages used to delete things. Breadcrumbs added in an earlier pass, the editorial sections on two hubs, and a punctuation sweep were all being reverted every time the hubs were regenerated, because they had been applied to the output rather than to the generator. All of it now lives in the generator, and fourteen hub pages that never had a breadcrumb trail have one.

A new post reads the employment data on creative work, and says plainly which parts of it could not be checked

Two of the most common questions we get are whether a course is worth taking and whether to give up freelancing for a salary. Both have a confident answer going around and neither answer is supported by what the statistical agencies publish. The new post goes and reads them.

• The central finding is that the aggregate and the specific disagree. Gallup's index of occupational exposure to generative AI finds earnings for highly exposed artistic occupations slightly positive and not statistically distinguishable from zero, while specific freelance categories report roughly a third of their volume gone. Both can be true, and the post explains why.
• It separates two things usually merged. Gallup measures whether artists earn less. The CREAATIF study, run with Equity, BECTU, the Musicians' Union and the Society of Authors, measures whether the work is becoming less secure. Income can hold while conditions get worse.
• Three sources were opened and read. The United States figures were not, because the Bureau of Labor Statistics refused automated access and the trade reporting is paywalled, so they are labelled as repeated rather than checked, in the text and again in the source list.
• A section is given over to what the article cannot tell you: every figure comes from the United Kingdom or the United States, the sector definitions bundle a camera operator with a museum curator, games and music are barely represented, and the most recent complete figures end in March 2024.
• The claims about where hiring is growing are flagged as the weakest evidence in the piece, because most of it originates with training providers and forecast vendors who benefit from being believed.

The pioneer corpus stopped around 1990. Twelve profiles now carry it to 2021

A review found seven of 110 profiles anchored after 1990, on a site whose About page claims the same attention to where media goes next as to where it came from. The corpus said otherwise. It now holds 122 profiles, 19 of them anchored in the last thirty-five years.

• The new cohort is the layer contemporary media actually runs on: the Opus codec and its two halves, H.264 and HEVC, VP9 and AV1, WebRTC and the negotiation underneath it, adaptive streaming, high dynamic range capture, and the file transfer protocol broadcast delivery depends on.
• Every source was fetched and read before it was cited. Three specifications from the RFC editor, five journal and conference records through Crossref, and one patent. Nothing rests on a search result.
• None of these pages carries an illustration, because the site's own rule forbids generated portraits of real people. That rule matters more for the living than the dead.
• The cohort was also chosen to widen a corpus concentrated in the United States and Western Europe, and includes work done in Canada, the Netherlands, Norway, Iran and India.
• Two faults surfaced while building it. The A to Z page was counting one woman where there are seventeen, because the new entries recorded a field differently. And semicolons the house style bans came back into the hubs, because a sweep in August fixed the published pages but not the data they are generated from. Both are fixed at source.

Transcripts and chapters belong in the feed, not just on the page

A podcast feed is an RSS file, and RSS was built for text articles. Everything podcasting needs beyond a headline and an audio file has been added since. The podcast namespace is the open attempt at doing that properly, and two of its tags change the most for the least work.

• The new guide gives the exact attributes for the transcript and chapters tags, taken from the specification rather than from a summary of it, including which transcript formats are accepted and what the captions flag does.
• It explains why chapters kept in a separate file can be corrected after publication and chapters baked into the audio cannot, which is the practical reason to prefer the tag.
• It is honest that app support is uneven, and explains why that is not a reason to wait: an application that does not recognise a namespaced tag skips it, so nothing breaks.

A guide to the European Accessibility Act, which this site had mentioned once

The Act has applied since June 2025 and reaches anyone selling media or media services to people in the European Union, wherever they are based. Accessibility is the deepest subject on this site, with nineteen guides and a set of caption and contrast tools, and it had one passing mention of the law that now governs the work.

• The new guide covers who the Act reaches, why conformity is measured against a standard the law does not name, and what the work consists of: captions that can actually be read, audio description where the pictures carry the meaning, a player that works from a keyboard, and a checkout that does too.
• It says plainly which claims rest on documents that were read and which do not. The Commission's own page and the WCAG 2.2 specification were both read directly. The application date of 28 June 2025 was not, because the consolidated legal text would not load, so the page says so and points at the article that settles it.
• One blog post had no publication date and now carries the one printed in its own byline. A second still has none, because there is no date anywhere on it and inventing one would be worse than leaving it blank.

Forty course pages said the same thing forty times, so they are gone

Five courses shipped as an overview plus seven to nine modules each. Measuring them settled what to do: 31 sentences appeared word for word in all forty modules, and the median module was two thirds sentences that recurred across five or more of its siblings. They were not short. They were repetitive.

• The sequence each course promised is now written once, properly, in the overview that introduces it, with a vocabulary section for the terms it uses. Forty-five pages became five, and each says its own subject's particular thing.
• The modules had been hidden from search in August, which stopped them being found but left five courses hollow and five indexed overviews linking to pages nobody could reach.
• All 170 biographical profiles now carry a Sources section. The 110 pioneers were done in August. The 60 profiles of cinematographers, composers, directors, editors, producers and screenwriters now say what has not been checked and which record would settle it.
• Those 60 are notes rather than citations, and the reason is worth stating. Every institutional record refused automated access: the Academy's awards database, BAFTA, the American Society of Cinematographers and the AFI catalogue all returned errors, and the BFI's search was down. The one database that did answer records an Academy Award nomination against the wrong film, so citing it would have added a mistake to a page that did not have one.
• 109 pages showed no breadcrumb while telling search engines what their breadcrumb was. They now show it, built from the same data, so the two cannot drift apart.

Three tutorials were listed nowhere, and 52 titles were being cut off in search

A review of the whole site turned up a set of small structural faults. None of them broke a page, and all of them made the site slightly harder to find or to navigate. They are fixed.

• Three tutorials appeared on no listing page at all and could only be reached by a link from a sibling article. The tutorials index also counted 570 when it held 573. Both corrected.
• 52 titles ran past the length a search result will show, the worst at 143 characters because two subtitles had been joined with colons. They now keep as many whole clauses as fit rather than being cut mid-phrase.
• 158 breadcrumb trails carried a link to Learning Paths wedged between two real steps, with no separator, so it rendered as one run-on word. A learning path is not a parent of a tutorial, and the machine-readable trail never included it, so it has been removed from the trail rather than patched.
• Four pages showed a breadcrumb with no category in it while telling search engines there was one. Six more used a name for their category that nothing else on the site used.
• 125 pages had no machine-readable breadcrumb, and 709 had no modification date. Both are now present. The date is copied from the sitemap rather than invented, so the two can never disagree, and it says when a page changed rather than claiming anybody reviewed it.

Visits are counted again when somebody reads one page and leaves

Analytics and advertising were both held back until the first scroll or tap, or for about three seconds. That is a reasonable trade for advertising and a poor one for counting: anybody who arrived, read what they came for and left inside that window was never recorded at all. On a reference site those are the visits most worth knowing about.

• Analytics now runs when the page loads, so a visit counts whether or not the reader touches anything. Advertising stays held back until the first interaction, which is where nearly all of the loading cost was anyway.
• Seven pages were found loading both at once, counting every visit to them twice. They are a different shape from the twenty six found on 20 August, which is why the earlier sweep missed them. All 878 pages now carry exactly one of each.
• Checked in a real browser rather than assumed: loading a page and touching nothing sends the visit within about a second, and the advertising script still does not appear until later.
• Figures from 26 August onward will look higher than the days before, and the earlier figures were the undercount.

The microphone we recommended is out of production

Two pages recommended the Audio-Technica ATR2100x-USB. It has been discontinued, and its replacement is not a replacement: Audio-Technica split it into two models, one USB only and one XLR only, at twice the price. Neither fits a list of dual-output microphones under a hundred dollars.

• Both pages now recommend the Audio-Technica AT2005USB, which is still made, still has both USB and XLR outputs, and still costs around eighty dollars. That is what those entries were describing all along.
• Fifty prices inside the machine-readable copy of these pages were still in pounds and euro while the visible text said dollars. Search engines read both. They agree now.

Prices are approximate now, and all in dollars

This site quoted prices to the exact dollar across 51 pages in three currencies. A figure like $99 is wrong the moment the manufacturer changes it, and checking that many by hand is not something anyone keeps up with. Two of them turned out to be subscription tiers that had never existed. The choice was to drop prices entirely or to stop pretending they are precise, and the second is more useful to a reader.

• 111 figures changed across 24 pages. Everything is in US dollars, and a price beside a named product now reads as an approximation: about $100 rather than $99.
• Prices in pounds and euro were converted at the rate on 24 August, after dividing out the 20% sales tax those prices include and US prices do not. Checked against six products whose US price was confirmed directly: four landed within one percent, the worst was eleven percent out.
• Band edges keep their shape and stay round. "Under $100 dynamic" is a category, and the number is chosen for being round rather than for being exact.
• Five pages that carry a genuine check date were left alone. Somebody verified those figures against a vendor, and rounding them off would throw away the only precise pricing on the site.
• The voice-over rate calculator has a currency selector and simply opened on pounds. It opens on dollars now.

Six affiliate links in a dropdown had never worked, and 39 pages failed a target size rule

The accessibility check that runs in a real browser, rather than by reading the source, is now a script instead of something done by hand. Its first full run found 43 pages with problems. All of them are fixed.

• The noise floor tool listed microphones in a dropdown, and each entry had a shop link wrapped around the product name. A link inside a dropdown option does nothing in any browser, so those six had never been clickable by anyone. They are plain text now, which is what they always were in practice.
• The DAW picker put links inside its answer buttons. A link inside a button is ambiguous for anyone using a keyboard or a screen reader, because it is not clear what pressing it will do. The product names stay, the links are gone.
• Three comparison tables scroll sideways on narrow screens and could not be scrolled without a mouse. They are reachable from the keyboard now and announce themselves properly.
• 39 pages had citation lists where each link sat on its own 19 pixel line with no gap above or below. The links are wide but the space around them was too tight to hit reliably. One line of stylesheet fixes every one of them.

A correction to yesterday's correction about the affiliate links

The note about the non-existent ATR2020 said the links behind it were pointing at the wrong product. Resolving the shortened URLs shows that was not true, and the record should say so.

• Both links pointed at real Audio-Technica microphones. One resolves to the AT2020USB-X and the other to the AT2020. Only the words wrapped around them were wrong, naming a model the company does not make.
• The AT2020USB-X link is restored on the starting a podcast guide, now labelled with the product it actually goes to.
• Three mentions of the ATR2100x-USB carry no link at all for now. That is the correct microphone for those entries, a dynamic with USB and XLR outputs, and it needs a link of its own rather than being pointed at a condenser.

Illustrations for the tutorials that had none, and the image folder halved

443 of the 573 tutorials carried no illustration. 37 of those are pioneer profiles, where a missing portrait is a deliberate state rather than a gap, so the real backlog is 406. Ten are now done as a first batch.

• The prompts are built from each page rather than from its title. The scene comes from the page's own summary, the equipment in the frame comes from counting which tools that page actually talks about, and the people come from the discipline. An earlier attempt at this derived prompts from the file name and produced pictures that could have illustrated anything.
• Alt text describes the illustration rather than repeating the heading beside it, which is the argument this site's own alt text guide makes. A screen reader user who has already heard the heading learns something new from the image description.
• The JPEG fallbacks were the heaviest thing the site shipped: 208 MB across 196 files, a median of 1.2 MB each, while the WebP versions almost everyone actually downloads are 67 KB. Re-encoding at a quality where a side-by-side comparison shows no difference took that to 104 MB.
• That matters for what comes next. Finishing the remaining 396 illustrations would have pushed the image folder close to a gigabyte. At the new size it lands near 430 MB.

A microphone that does not exist, and two subscription prices that were never offered

A price check turned into a correction of the corrections. Two pages were still recommending an "Audio-Technica ATR2020". No such product has ever existed. The changelog for 4 August records that error as fixed, and the fix never reached either page.

• The real product is the ATR2100x-USB, a cardioid dynamic with both USB-C and XLR outputs, which is also what the roughly eighty dollars already on the page matches. The earlier correction resolved it to the AT2020, which would have been wrong in one of the two places: that entry sits in a list of dynamic microphones and the AT2020 is a condenser.
• The affiliate links on those two items have been removed rather than repointed. One of them was aimed at a different microphone entirely. A commission link pointing at the wrong product is worse than no link.
• Riverside was listed at $19 a month. There is no such plan. It is $29 a month, or $24 billed yearly, with a free tier below that. SquadCast was listed at $10 a month, which is also not a rate they offer: the cheapest paid tier is $24 per person a month, or $16 billed yearly.
• Buzzsprout, Podbean and Captivate all had figures that are only available if you pay a year up front. The pages now say so. Captivate is priced in euro and was shown in dollars.
• Two pages have had their check date removed. A date that survived an error the size of a non-existent product is not evidence of anything, and the podcast hosting page still quotes figures for Libsyn, RSS.com and Simplecast that could not be verified.

The About page now shows the working instead of describing it

It said that calculations disclose their method and that corrections are recorded. Both were true and neither was demonstrated, so a reader had no way to check. The page has gone from 415 words to just over a thousand, and the new material is all things you can go and verify.

• How the calculator tables are built, and the depth of field error that building them this way exposed. That tool was live and wrong, overstating hyperfocal distance by half.
• Where the sources are, with the split stated plainly: 89 of the 110 pioneer profiles link a document that was read before it was cited, and 21 explain why no document could be.
• Three pages that were corrected because the patent named more inventors than the page did, and two widely repeated claims that were checked and then not published because the records did not support them.
• How prices are dated, and why roughly ninety of the money figures on this site carry no check date and never will. They are category estimates rather than product prices, and stamping them as verified would be the easiest way to look more authoritative than the evidence allows.

137 pages were carrying no advertising code, and 26 were counting every visit twice

Every page on the site now loads analytics and advertising through the same deferred loader, which waits for the first scroll or tap rather than running on page load.

• All 110 pioneer profiles, the pioneers hub, the technology timeline and every tutorials category page were on an older head that loaded analytics immediately and carried no advertising code at all. That is the most heavily sourced writing on the site and none of it could serve an ad.
• Checking the swap turned up a separate fault. 26 pages carried both the new loader and a leftover copy of the old analytics tag, so Google Analytics loaded twice and every visit to those pages was counted twice. The leftover has been removed.
• Six page generators were emitting the old head, so the fix would have reverted the next time any of them ran. Three of them also emitted an advertising tag that ran on page load. All six now emit the deferred loader.
• Verified across all 878 pages: one loader each, one advertising reference each, one analytics reference each, and no stale tags anywhere.

All 110 pioneer profiles now show their working

The last 26 are done, which finishes the sourcing pass: 89 profiles cite a document that was fetched and read, and 21 carry a Sources section explaining why no document could be cited.

• Marey's own 1882 article on the chronophotographic gun was found transcribed in full. He gives the rate as twelve exposures a second at one seven hundred and twentieth of a second each, which is exactly what this site had claimed.
• Scott de Martinville's 1857 manuscript is cited from the First Sounds facsimile, whose title page records the original at the Academie des sciences as number 1639.
• The Lumiere US patent turned out to be the way to verify the French one. Its foreign priority claim records French patent 245,032 of 13 February 1895, and it names Auguste alongside Louis.
• Twenty one profiles get a different kind of section. Charles Cros is the model: the secondary accounts contradict each other on his deposit date, some saying 30 April 1877 and some 18 October, so the page names the contradiction and says the Comptes rendus volume on Gallica would settle it. Bebe Barron's note explains that the missing document is the story, since Forbidden Planet credits the Barrons for electronic tonalities rather than music. Delia Derbyshire's says the broadcast credit names Ron Grainer and not her, which is the whole difficulty.
• A page that makes dated claims and shows no working is worse than one that shows the working and says where it stops.

Kalmus is not on the Technicolor camera patent

Nineteen more pioneer profiles now carry a visible Sources section, taking the total to eighty four of a hundred and ten.

• The three-strip beam splitter camera, US 2,072,091, names Joseph A. Ball and Gerald F. Rackett. Herbert Kalmus is not on it. His page already said the camera was a team product, and the patent is now cited as the specific evidence for that.
• Natalie Kalmus's "Color Consciousness" is now cited from the Journal of the SMPTE, August 1935, under her own name. Whatever is disputed about her on-screen credits, that paper is hers.
• Kerns Powers has two of his own SMPTE papers cited, and the note says plainly that neither is the 16:9 derivation itself. No published derivation of the ratio could be found, so the 1984 date on that page rests on secondary accounts. Saying so is more useful than citing something weaker.
• HTTP Live Streaming is cited from both the 2009 Internet-Draft and RFC 8216, with a note that the RFC is an Independent Submission with informational status rather than an IETF standard.
• Six profiles were researched and left uncited because no primary record could be reached: Georg Neumann, Lou Ottens, Isaac Shoenberg, Jack Mullin, William Friese-Greene and Louis Lumiere. Those need a patent office archive rather than a web search.

Two more profiles were crediting one person where the patent named two

Twenty one more pioneer profiles now carry a visible Sources section, taking the total to sixty five. Reading the actual records rather than the received account changed two of the pages.

• The digital camera patent, US 4,131,919, names Gareth A. Lloyd first and Steven Sasson second. This site credited Sasson alone. Both pages, the visible text and its structured data, now name Lloyd.
• Erna Hoover's switching patent, US 3,623,007, names Barry J. Eckhart alongside her. The same correction has been made there.
• Two widely repeated claims were checked and dropped rather than written up. US 5,086,428 is described all over the web as Radia Perlman's spanning tree patent and is actually a wide area broadcast patent with two co-inventors. Her spanning tree work is cited from her 1985 paper instead. Separately, Bernard Loughlin's US 2,943,142 is offered by several accounts as the prior art that stopped Walter Bruch patenting PAL, and reading it shows it alternates phase by field rather than by line, which does not support the claim.
• Papers are now cited by DOI with the volume, issue, pages and date taken from the publisher record: Huffman 1952, Nyquist 1928, Fletcher and Munson 1933, Ahmed, Natarajan and Rao 1974, Ziv and Lempel 1977, Kao and Hockham 1966, Braun 1897, Kirsch and colleagues 1957, Perlman 1985.

The videotape patent connects four of these profiles

Reading the Ampex quadruplex patent turned up something the separate profiles had not made obvious. It names four inventors, and one of them is Ray Dolby, seventeen years before he founded the company known for noise reduction.

• The patent is now linked from the Charles Ginsburg, Fred Pfost and Ray Dolby pages. Ginsburg's page already described the recorder as the work of a six person team, and the record supports that: Ginsburg, Henderson, Dolby and Anderson are all named on it.
• Fred Pfost is not on the inventor list, which the source note says plainly. He built the rotating heads and their ferrite cores, and component work of that kind frequently sits outside the inventor list on a system patent.
• Eadweard Muybridge's 1881 motion photography patent is linked, describing the bank of cameras with electrically synchronised shutters.
• Peter Goldmark's phonograph record patent is linked, with a note that it was filed in 1956 and therefore documents the format rather than being the original 1948 filing.

Sources for the wireless dispute and early cinema

• The Marconi page now links two patents rather than one. His own from 1896, and Oliver Lodge's tuning patent filed in February 1898, which covers the synchronised tuned circuits that stop stations interfering with each other. Lodge's page links the same record. Marconi's company later acquired that patent, and having both dates visible lets a reader judge the priority question rather than take a side's word for it.
• Jagadish Chandra Bose's detector patent is linked, filed in 1901 under the transliteration Jagadis Chunder Bose. It describes a contact that recovers on its own, which is a semiconducting detector rather than a coherer needing to be reset by hand.
• Louis Le Prince's 1886 moving picture patent, Léon Theremin's 1925 instrument patent, Homer Dudley's vocoder patent and Fritz Pfleumer's magnetic tape patent are all now linked from their pages.
• Several of these were filed under names or transliterations different from the ones the people are known by, which is noted each time, because it is the reason a patent search under the familiar name comes back empty.

Thirteen more pioneer profiles now cite their patents

A larger batch, covering the founding patents of telephony, recorded sound, radio and stereo. Each one was read before it was cited.

• Bell's 1876 telephone patent, Edison's phonograph patent filed on Christmas Eve 1877, Eastman's roll film camera, Tesla and Marconi's era of wireless patents, Fleming's valve and de Forest's grid, Armstrong's superheterodyne, Blumlein's stereo patent, Granville Woods's induction telegraph, and Ray Dolby's companding noise reduction.
• Hedy Lamarr's frequency-hopping patent is filed under Hedy Kiesler Markey, her married name at the time, which is why searching a patent index for the name she is known by finds nothing. The source note says so.
• The electret microphone patent names Gerhard Sessler and James West together, and is linked from both their pages.
• A note on method: several patents are cited on more than one page, which is deliberate. Where two people disputed the same ground, both pages link the same record so a reader can check the claim from either side.

Correction: two inventors were missing from the image sensor page

The Eric Fossum profile credited him alone for the CMOS active pixel sensor. The patent names three people. Sunetra Mendis and Sabrina Kemeny are listed alongside him on US 5,471,515, and this page left them out until today. That is exactly the kind of omission these profiles exist to correct, so it is worth saying plainly rather than quietly fixing. The page now names all three and links the patent.

• Michael Tompsett's imaging patent is linked, naming him as sole inventor and filed in 1972. He holds the patent for applying charge coupled devices to imaging, and the 2009 Nobel for the CCD went to two other people.
• Claude Shannon's founding paper is linked as scans of the two 1948 journal issues it was split across, rather than as a summary of it.

Sources for the colour film chemists

• Kodachrome's patent is now linked from both the Mannes and Godowsky pages. It names the two of them jointly, was filed in February 1935, and its specification confirms the point the pages make: the dye is formed by coupling during development, with no colouring compounds sitting in the emulsion beforehand.
• Rudolf Fischer's 1912 patent is linked, describing the opposite approach that became the basis of nearly all colour film: couplers in the emulsion, forming dye where the silver develops.
• One thing the patent record surfaced. Fischer's US patent names him alone, while contemporary accounts of the work also credit the chemist Hans Siegrist. The source note says so, and the question of how that credit should be divided is one we want to research properly before rewriting the page.

Sources for the recorded sound pioneers

• Emile Berliner's gramophone patent is linked, granted 8 November 1887. The record also shows it as a division of an application filed back in 1883, which is a detail most accounts skip.
• Valdemar Poulsen's Telegraphone patent is linked, filed 8 July 1899 and granted 13 November 1900. Its specification describes recording onto a steel wire magnetised along its length, which is exactly what the page claims for it.
• Marvin Camras's AC bias patent is linked, filed 22 December 1941. The specification describes adding a high frequency current alongside the audio to straighten the magnetisation curve, which is the technique that lifted magnetic tape from dictation to broadcast quality.
• Charles Cros has deliberately not been given sources yet. His claim rests on a sealed envelope rather than a patent, and the secondary accounts available disagree with each other about the date. The page will get a citation when the entry in the Academy's own 1877 proceedings has been located, and not before.

Sources added to the rest of the television cluster

Three more profiles now carry their sources, completing the group of pages that argue about who invented television.

• Kálmán Tihanyi's 1926 Radioskop application is held in UNESCO's Memory of the World register, inscribed in 2001. The entry records the charge storage principle and the 1928 French and British filings, and it is now linked from his page.
• Kenjiro Takayanagi's demonstration is confirmed by his own university: 25 December 1926, at Hamamatsu Higher Technical College. The institutional account also confirms the detail that matters, which is that the image was scanned mechanically and displayed electronically.
• John Logie Baird's January 1926 demonstration is sourced to the Royal Institution, whose members walked to Frith Street to watch it, alongside the Science Museum record for the surviving apparatus.

Pioneer profiles start carrying their sources

The profiles make dated claims about named people and carried no citations. The first eight now show their sources on the page, starting with the ones where the popular version of events is most often wrong.

• The television priority dispute is now documented from the patents themselves. Farnsworth filed on 7 January 1927 and was granted in 1930. Zworykin filed on 29 December 1923 and was not granted until 1938. That fifteen year gap is the whole argument, and both records are linked.
• The Yagi-Uda antenna page links the US patent, which names Hidetsugu Yagi alone. Shintaro Uda, who did the principal experimental work, does not appear on it, which is the documentary root of the single-name credit still used in the West.
• Otis Boykin is often said to have invented the pacemaker. Both of his resistor patents are now linked, filed in 1954 and 1956, and neither is for a pacemaker.
• Interlaced scanning is credited to Schröter in German accounts and Ballard in American ones. Both patents are linked, showing Schröter filed on 27 September 1930 and Ballard on 19 July 1932.
• Every citation was checked against the record before it was published. Where a claim could not be verified, no source was invented for it.

Punctuation brought in line across the site, and some stale prices corrected

• Em dashes and semicolons have been removed from the writing across 281 pages, following the site's own style guide. Code samples, embed snippets and the placeholder dashes tools use to show an empty value were all left alone.
• Price corrections. The voice-over home setup guide recommended a USB microphone that is discontinued, at roughly half what its current replacement costs, and quoted the Rode NT1 at $180 when it is now $214. Both are fixed.
• The starting a podcast guide named an Audio-Technica ATR2020, which is not a product Audio-Technica makes, and recommended Anchor for hosting. Anchor became Spotify for Podcasters in 2023 and Spotify for Creators in 2024. Both corrected.
• Prices carry a checked date only where they have genuinely been verified against a manufacturer or retailer listing. Pages where only some figures were confirmed say so and keep no date, because a date on an unchecked number is worse than no date.

Every guide on the site now defines the terms it uses

The last 50 tutorials are done, which completes a pass across the whole site. Nearly two hundred pages now carry a vocabulary section written for readers meeting the subject for the first time.

• Eight guides that were running short gained a full section as well: when the Logic groove track is the right fix and when quantising is, why matching exposures across a shoot is wrong when the light genuinely changed, why log and HDR are a recording choice and a delivery choice rather than the same thing, and why exposing a property upward beats duplicating a composition.
• The remaining guides gained definitions covering everything from rolling shutter and egress charges to pre-delay, formants, sidechain inputs, analytics prefixes, and the difference between master and sync rights.
• 739 distinct terms are now defined across the site. Where two pages need the same term, each is written for its own context rather than copied, and no definition sentence appears twice anywhere.

Vocabulary sections across 53 profiles

The profile pages assume a working knowledge of the field. Each one now defines the technical ideas it refers to, written for readers meeting them for the first time.

• 43 broadcast pioneer profiles gained definitions of the concepts behind the work: critical bands and masking, the sampling theorem and aliasing, prefix-free codes, chromogenic development, total internal reflection, parasitic antenna elements, AC bias and coercivity.
• 10 profiles of working cinematographers, producers, screenwriters and editors gained the craft terms they use, covering anamorphic lenses, available light, co-production and pre-sales, spec scripts, assembly cuts and cutting on action.
• Every definition is written for the page it sits on. No term is defined on two pages, and no definition repeats another's wording anywhere on the site.

Every course module and orientation page now explains where the work goes wrong

The course material listed what to do and rarely what happens when it goes wrong, which is the part a numbered workflow cannot carry.

• All 42 remaining lesson modules across Ableton, Avid, Logic, Premiere and Pro Tools gained a section written for that module specifically: why quantising everything kills a groove, why syncing by timecode fails at the tail of a roll rather than the head, why the ADR guide track is a reference rather than something to imitate, and why movement in an immersive mix works best when it is rare.
• The two orientation courses, which were the thinnest pages on the site at around 215 words, now cover why readiness comes before the camera and why the workflow habits exist. Both are written for students with no prior production experience.
• Five hub pages and one blog post gained closing sections, including how to read a caption check result and what a batching workflow costs when it goes wrong.

Every tool page now explains its subject

The last twelve pages are done, which completes a pass across all 93 tools. Each one now carries the working knowledge behind it and a vocabulary section defining its terms for readers new to the subject.

• The metadata stripper covers what a photograph carries that you did not intend: camera serial number, timestamps to the second, and on most phones the coordinates of the place it was taken, accurate to a few metres. Platforms strip this inconsistently, so assuming they will is unwise.
• The time-lapse calculator explains why shutter speed matters as much as interval, since frames frozen with a fast shutter strobe instead of flowing.
• The storage calculator makes the point that codec matters far more than resolution, and that footage exists on one card until it exists in two places.
• The colour temperature page covers mixed lighting, where no single white balance can neutralise two sources and the fix belongs at the light rather than in the grade.
• The four caption pages now take deliberately different angles: what the rules are, how the file format works, what to do when a check fails, and when to burn captions in permanently.

Twelve more tools explained, and an affiliate disclosure fixed

Captions, type, microphones, compression, streaming and pricing.

• Disclosure correction. The Plugin Chooser carried its affiliate disclosure inside the tool, where it appeared only after finishing the quiz. It is now stated in plain text at the top of the page, visible whether or not you use the tool: we may earn a commission on purchases made through those links, at no extra cost to you. The recommendations are unchanged, and the page now opens by saying most readers should buy nothing.
• The Caption QA Studio page covers what caption quality actually means: reading speed rather than spelling, two lines as the working maximum, a minimum duration below which a cue registers as a flash, and gaps between cues so a reader can see the text changed.
• The compression page explains what each control does, and why the gain reduction meter is the thing to watch rather than the output.
• The microphone pattern page covers choosing a pattern for the room you have, which in an untreated space matters more than the microphone's price.
• Font pairing, caption editing, banner safe areas, speaking pace, streaming bitrate, voice-over pricing and caption burn-in each gained similar guidance and a vocabulary section.

Ten measurement and delivery tools now explain what they measure

Loudness, noise, stereo width, room acoustics and ear training. These tools produce numbers, and a number is only useful once you know what it does and does not tell you.

• The loudness pages explain why delivering louder than a platform's target gains nothing, since every major service turns your file down to its own figure and a hard-compressed master arrives quieter and flatter than one delivered at target.
• The stereo width checker covers why mono compatibility still matters, given how much listening happens on phone speakers and smart speakers that sum both channels, and why cancellation removes content rather than merely quietening it.
• The room mode calculator explains why bass is uneven in small rooms, and why acoustic foam does effectively nothing at those frequencies because the wavelengths involved are metres long.
• The noise floor page makes the point no equipment purchase can beat: halving the distance to the microphone improves the ratio by about six decibels and costs nothing.
• All ten gained vocabulary sections covering LUFS, true peak, phase correlation, RT60, standing waves and the rest.

Ten production and workflow tools now carry the practice behind them

Call sheets, shot logs, storyboards, shot planning, interview prep, releases, file naming, prompting, voice-over and the metronome.

• The call sheet page covers why crew call, talent call and first shot are three different times, and why a sheet issued the night before catches problems a sheet issued on the morning does not.
• The file renamer explains the two habits doing most of the work in an archive: dates written year first so they sort chronologically, and numbers padded with leading zeros so take 9 stops sorting after take 10.
• The shot planner covers grouping by setup rather than by scene, since moving the camera and its lighting is the slowest thing that happens on a shooting day.
• The voice-over recorder explains why the room matters more than the microphone, and why recording as loud as possible is a habit inherited from tape that actively harms a digital recording.
• The release form page explains what a release is for and what makes one hard to rely on, and says plainly that it is not legal advice and that a lawyer where you are is the only reliable answer.

Ten podcast and publishing tools now explain the decisions behind them

The podcasting set gets the same treatment as the audio and picture tools.

• The RSS validator explains the mistake that causes the most damage: changing an episode GUID during a host migration, which makes every app treat a months-old episode as a new release and notify every subscriber.
• The artwork validator covers the requirement people miss, which is colour mode. Artwork exported from print software carries CMYK colour, looks correct on screen, and is rejected outright.
• The chapter generator lists the three conditions YouTube requires before a timestamp list becomes chapters at all, and notes that chapters change how a retention graph should be read.
• The CPM calculator explains why an estimate overstates: lifetime downloads instead of a thirty day window, every slot assumed sold, and no allowance for the network's share.
• Show notes, guest outreach, episode outlines, titles and intro scripts each gained similar guidance, plus vocabulary sections defining every term.

Ten picture and colour tools now explain the craft behind them

The same treatment the audio tools received. Each page now carries the working knowledge that decides whether the tool gives you something usable.

• The green screen checker explains why keys really fail: uneven backdrop lighting, spill from standing too close, and chroma subsampling, which stores colour at a quarter of the picture detail and limits how clean any key can be.
• The vectorscope page covers the skin tone line, which sits at the same angle for every skin colour because blood beneath the surface dominates the hue.
• The LUT previewer separates technical LUTs from creative ones, and explains why applying a look directly to log footage produces the flat, washed result people so often get.
• The contrast checker explains why yellow on white fails despite the colours looking different, and the colour blindness simulator covers why brightness difference rescues almost any design.
• False colour, safe zones, lower thirds, the background remover and the palette extractor each gained similar guidance, plus vocabulary sections defining every term used.

Eight audio tools now explain how to use them well

These tools process a file you supply, so there was no table to add. What they were missing was the knowledge that makes the difference between a usable result and a wasted hour.

• The noise remover now covers why pushing reduction too far leaves a watery warble that no later processing removes, and when moving the microphone closer beats any amount of repair.
• The converter explains what sample rate and bit depth actually control, and why converting up from 16 bit to 24 adds nothing back.
• The trimmer covers why cuts click, and why ten milliseconds of fade fixes it inaudibly.
• The silence remover, sibilance finder, pitch analyzer, spectrum analyzer and audiogram generator each gained similar guidance, including the frequency band where sibilance sits and why an audiogram without captions communicates nothing.
• All eight now carry a vocabulary section defining every term they use, written for readers new to audio.

Worked examples for the timecode and aspect ratio tools

The last two calculators with a formula worth tabulating.

• Timecode against real elapsed time, at eight frame rates. One hour of 29.97 non-drop timecode takes 1:00:03.600 of clock time, and drop-frame pulls that back to within four milliseconds. Seeing the drift written down is the clearest answer to why drop-frame exists at all.
• What reframing costs from a 16:9 source. Cutting to 9:16 for vertical discards 68 percent of the width, which is why a straight crop so rarely works and why deciding on vertical before the shoot saves an entire reframing pass.
• Both pages gained vocabulary sections covering timecode, drop-frame, pull-down, letterboxing and centre-safe framing.

Worked examples for four more calculators

The same treatment as the five tools covered earlier. Each page now shows its common cases already worked out, computed from the same formula the calculator runs.

• Field of view for six lenses across four sensor sizes, with the full frame equivalent alongside. A 24mm on Micro Four Thirds and a 50mm on full frame both see 39.6 degrees, which is what the crop factor is telling you.
• Every frame rate conversion, with the speed change and how far the audio moves in semitones. Halving the frame rate drops pitch by a full octave.
• Musical timings in frames at 24, 25 and 30, so animation cuts land on the beat rather than near it.
• Day rates from target income across four income levels and three assumptions about billable days, which is the figure most people overestimate.

Correction: the depth of field calculator was wrong, and worked examples for five tools

The depth of field calculator has been reporting incorrect figures. It overstated hyperfocal distance by half again on every sensor and lens, and understated depth of field by about a third. A 50mm lens at f/2.8 focused three metres away was reported as holding 0.39 metres of depth when the real figure is 0.59 metres. If you set focus marks using this tool, they were tighter than they needed to be. The cause was a unit conversion applied twice, and it is fixed.

• Five calculators now show their common cases already worked out: bitrate by resolution and frame rate with file sizes, delay times at six tempos, levels converted across dBu, dBV, volts and gain, depth of field for four focal lengths on two sensor sizes, and safe shutter speeds for both mains frequencies.
• Every table is computed from the same formula the calculator runs, so the two always agree. Working through those formulas line by line is what surfaced the depth of field error.
• Each of the five pages also gained a vocabulary section defining the terms it uses, written for readers coming to the subject new.

Reference tables for five tools, and real introductions on five hubs

Several pages were carrying their content inside the tool rather than on the page, so you had to work the widget to find out what was in it.

• Five tools now print their full data as a table underneath: every platform specification side by side, all eleven podcast hosts with the figures the picker uses, every DAW it considers, the frequency bands in full, and common light sources by colour temperature. The tools return one answer. The tables show you the whole set behind it.
• Filmmaker Profiles, Primary Sources, Embeddable Tools, Standards and the Roadmap each gained a real introduction covering what the collection holds, how the pages are researched, and how to read them alongside the guides.
• Filmmaker Profiles was claiming sixty profiles in one place and a hundred and seventy in another. Sixty is the filmmaker count, ten each across six roles. The other hundred and ten are broadcast pioneers, which are a separate collection.

Hero illustrations for all 24 pillar guides

The pillar guides, one canonical page per discipline, were the last major pages running without an illustration. All 24 now have one, drawn in the same 1960s editorial style as the rest of the site.

• Each illustration is built from its own page rather than from the topic in general. The audio recording guide shows distance, room, pattern and gain as four labelled zones with a polar pattern diagram and a headroom meter, because those are the four physical facts the page is about.
• The mixing guide reads left to right through rough levels, pan and spread, EQ carving, compression, send effects and bus glue, matching the order the guide works in.
• The broadcast history page uses generic apparatus and anonymous silhouettes throughout. Illustrating real people or specific historical machines would mean inventing them, so that page uses shapes instead.
• All 24 ship as WebP at three widths with a JPEG fallback. A phone now receives 66 KB per illustration rather than 964 KB.

Contact and Terms pages, and one footer across the whole site

Two pages that should have existed a long time ago, and a footer that finally says the same thing everywhere.

• New Contact page: what we can and cannot help with, and how to report a correction, a tool bug or an accessibility failure usefully.
• New Terms of Use: how the browser-based tools may be used, that files never leave your device, and the limits of the guidance published here.
• The footer had quietly drifted into 13 variants across 838 pages. On 271 of them, nearly every app page, there was no way to reach the privacy policy or anything else from the bottom of the page. Every page now carries the same footer: About, Contact, Privacy, Terms and Changelog.
• Generators now import that footer from one module instead of each carrying its own copy, so the drift cannot come back on the next content wave.
• The About page now says who writes this site: a group of media professionals who worked through the move from film to digital acquisition, tape to file-based storage, and scheduled transmission to streams.

New blog post: How to Post Social Without Burning Out

A first-person workflow piece on posting without burning out. The argument: filming was never the bottleneck, the thirty minutes of trimming and captions after every video was, which is why the avatar tools solve the cheap half.

• Filming in blocks on a rhythm the posting schedule never touches, and batching the caption and trim pass by task rather than by video.
• Why a two-week queue turns publishing from a deadline into a stock level, and how to build one without extra hours.
• What AI avatars, clip finders, transcript-based editing and auto-captioning each remove, and which of those steps still needs a human.
• An order of operations for creators already locked into a cadence they cannot hold.

Pioneer images: corrections, attribution and an accessibility fix

The broadcast pioneers corpus now carries 81 images and 29 cards, with full provenance on the image credits page. Several images were replaced or corrected: a photograph filed under Bryce Bayer was identified as Steven Sasson holding the 1975 Kodak digital camera prototype and moved to his profile, and Bayer now appears against his own colour filter array.

Eadweard Muybridge’s page shows his 1887 galloping-horse sequence as an animation. It loads as a still frame with a play control rather than looping on its own, so it does not become motion nobody can stop.

Where an image illustrates a format rather than a person’s own equipment, the page says so. The reel-to-reel deck on Jack Mullin’s profile is a 1978 Akai, three decades after the machines he actually rebuilt.

Historical images added to the broadcast pioneers corpus

81 of the 110 broadcast technology pioneer profiles now carry a real historical image. A photograph, an apparatus shot, or a patent drawing. Under a verified free licence: public domain, CC0, CC BY or CC BY-SA. Full provenance for every one is on the new image credits page, which records the literal licence template rather than a summary.

We never use generated or hand-drawn depictions of real people. Where no freely licensed image is known to exist. 29 entries, including Bebe Barron, Daphne Oram and Fritz Pfleumer. The page shows a plain card saying so rather than inventing something. Roughly a quarter of the images are archival scans smaller than our usual minimum. Those render at their original size in a distinct frame and are never upscaled.

New section: Media History, and the first broadcast technology pioneers

The site now has a dedicated Media History section. The 28 existing history guides have moved into it, joined by a new 2,450-word pillar, A History of Broadcast Technology, which follows the whole chain from capturing motion and sound through transmission, television, colour, tape and compression to streaming.

It launches with the first ten profiles of the people who built that technology: Paul Nipkow, Karl Ferdinand Braun, A. A. Campbell-Swinton, Boris Rosing, John Logie Baird, Kenjiro Takayanagi, Kálmán Tihanyi, Philo Farnsworth, Vladimir Zworykin and David Sarnoff.

All twelve waves have now shipped: 110 of 110 profiles are live. The corpus covers the whole chain. Capturing motion and sound, waves in the air, television’s invention and standardisation, colour, tape and the studio, electronic sound, the digital chain, delivery, and the electronic image. From Muybridge in 1878 to HTTP Live Streaming in 2009.

It includes 68 figures in the overlooked strand: Kenjiro Takayanagi, who displayed the first electronic television image months before Farnsworth, Kálmán Tihanyi, whose charge-storage patent RCA bought. Delia Derbyshire, whom the BBC forbade from taking a credit Ron Grainer tried to give her, Alice Guy-Blaché, whose sound films predate The Jazz Singer by twenty years, and T. Natarajan, co-author of the paper introducing the transform inside every image you will see today, of whom no biographical record appears to survive anywhere.

Two new browse surfaces sit over the corpus: an A-Z of broadcast technology pioneers and a timeline of broadcast technology running from 1860 to 2009. Both cover all 110 figures in the planned corpus, so entries whose profiles are still being written appear as timeline events now and become links as each wave ships. Both can be filtered by strand. Canon or overlooked, and by gender, which researchers asked for.

Every contested claim in these pages is stated carefully rather than confidently. The 1943 Supreme Court decision did not rule that Tesla invented radio. Farnsworth won his patent interference against RCA rather than losing it. Takayanagi displayed the first electronic television image on 25 December 1926, months before Farnsworth, and is almost absent from English-language accounts. Where historians genuinely disagree, both readings are given.

Thirteen new guides on writing and editing

Story and editing was the thinnest area on the site once you looked past the profiles. Thirty profiles of directors, editors and screenwriters were supported by six guides, and there was nothing at all on screenwriting. Ten screenwriter profiles pointed readers at a guide to YouTube retention hooks, because it was the only story page available to point at.

Screenwriting now has six guides: format and the reasons behind every rule, want and need, dialogue, the rewrite, loglines and coverage, and a myth-bust on whether you have to write in three acts.

Editing as a craft, rather than as software operation, has six more: when to cut, rhythm and pacing, choosing takes, assembly to fine cut, montage, and what to do when a scene refuses to cut together. There is also a screenwriting software comparison with prices taken from the official pages rather than from other people’s articles.

All forty director, editor, screenwriter and producer profiles now link to craft guides matching what the person does.

New blog post: why is everything so loud?

Why Is Everything So Loud? works through cinemas, trailers, television adverts, streaming and restaurants, and finds a different mechanism behind each one. The surprise is the cinema: most run their faders well below reference, which is why a film sounds too quiet and too loud in the same sitting. You lose the dialogue and keep the explosions.

It draws on re-recording mixers Tony Lamberti, Michael Babcock, Mark Mangini and Brent Kiser, plus Christopher Nolan and his supervising sound editor Richard King on deliberately burying dialogue. Every figure and quotation is sourced at the foot of the article.

The whole site now follows its own writing style

Our style guide has banned em dashes and semicolons for a while, and new guides have been checked against it automatically. Older pages predated the rule, so most of the site did not follow it. That is now fixed: roughly 7,500 dashes and 855 semicolons across 560 pages have been rewritten into ordinary commas, full stops, brackets and colons.

The tools were left alone on purpose. A dash in a results panel means "no value yet", and a comma there would be nonsense, so anything inside an app interface, a table cell or a label keeps its punctuation.

Corrected two gear errors and a stale interface price

The microphone guide listed the Rode Wireless GO II under shotgun microphones. It is a clip-on wireless system, which is a different tool for a different job, and the entry now says so. The podcast setup guide referred to an "Audio-Technica ATR2020", which is not a product Audio-Technica makes. It is the AT2020.

Prices on both pages were re-checked at the same time. The Focusrite Scarlett 2i2 had drifted furthest, listed at around $180 against a real price nearer $225. Three pages now carry a verified date.

Fixed share previews on 70 pages

When a link to this site was shared, 70 pages showed a title that no longer matched the page, left behind when titles were shortened for search results. One was garbled outright. All of them now match, and a new check stops them drifting apart again.

Eight new guides, aimed at the gaps rather than the popular topics

Four are troubleshooting guides for areas that had none: why your grade looks different on every other screen, why a render takes far longer than the preview suggested, why your photos are not sharp and why your CPM dropped. Symptom-first pages are the format people actually search with, and the site had them in five subject areas out of twenty-three.

Four more fill the two thinnest areas on the site. Music scoring and production management both looked well covered and were mostly composer and producer profiles, so the practical work was missing: tempo maps and hit points, orchestral mockups that do not sound like samples, scheduling a shoot day and insurance, permits and releases.

Illustrations now load about 13 times smaller on a phone

The article illustrations were 1.3 MB JPEGs, sent at full size to every device. They are now WebP at three widths, so a phone receives roughly 93 KB instead of 1.2 MB and a desktop 143 KB. The original JPEGs stay as a fallback, and the images keep their priority loading, because they sit at the top of the article where loading them late would slow the page down rather than speed it up.

Prices now carry the date they were checked

42 pages quote prices and four of them said when the figure was taken. Any page quoting a price now records when someone last verified it, and says so on the page once that check has happened.

The podcast hosting comparison is the first one done. Two of its six platform prices had drifted: Buzzsprout and Podbean were both understated. All six now match the official pricing pages as of August 2026. The rest of the price-carrying pages are queued rather than quietly stamped, because dating a figure nobody checked would be worse than leaving it undated.

Fixed 71 broken links, 151 mangled characters and 245 cut-off search descriptions

71 internal links pointed at pages that do not exist, mostly in the Related Guides blocks, and most of them were near-miss names for a page that does exist. A new check now fails the build on a link with no destination.

Separately, 151 characters across 26 pages had been mangled by an encoding fault, including the search icon in the navigation. And 245 pages had search descriptions long enough that Google cut them mid-phrase, so each now ends where we chose rather than where the cut landed.

New guide: EQ basics

EQ Basics covers what each control does rather than listing settings to copy: the four filter shapes, why a high-pass filter is named for what it keeps, what Q changes, and the sweep that finds a problem frequency in under a minute. Twelve pages across the site had been linking to this guide before it existed.

New guide: why your live stream looks bad

Why Your Live Stream Looks Bad, and How to Fix It covers browser-based studios, where the broadcast runs in a tab. Blocky, soft and freezing pictures each point at a different cause, so the guide starts by reading the damage on screen and then works through bitrate, resolution, codec choice and load in order.

Two pieces of common advice get corrected. A widely repeated floor of 2500 kbps for high quality sits below YouTube's own published minimum for 720p. And splitting your stream and your conference across two machines fixes an overloaded processor rather than a slow connection, because two machines on one connection still share one uplink.

Repaired structured data, share previews and images on 125 pages

A script linking product mentions to Amazon ran over whole pages rather than page bodies, so it wrote links into places no link belongs. The damage was invisible on screen and serious underneath: 108 pages had structured data a search engine could no longer parse, share previews were truncated mid-sentence, several canonical URLs had a link spliced into the middle, and 23 image paths and data attributes pointed at a URL fragment instead of a file.

All of it is repaired, and the linker now only ever touches the body of a page. Five pages that were live but missing from the tutorial index have been registered, so they finally appear in the hubs, the grid and search.

Seven more entries in Features You May Not Know About

The series now covers Pro Tools, After Effects, REAPER, Lightroom, Logic Pro, Media Composer and Blender. Each one is a capability shipping in software you already own, and each removes work rather than saving a click.

A sample: Pro Tools clip gain sits before your inserts and volume automation sits after them, so only clip gain changes what your compressor hears. Media Composer has two commands for trimming to the playhead in one keystroke, unmapped by default, which is why most editors never find them. Blender applies a value change to every selected object if you hold Alt, and has for years.

Search results now show the whole title

Google displays roughly 60 characters of a page title before cutting it off. Nearly half the site was over that line, with the longest running to 115 characters, so the most useful part of many titles never reached anyone browsing search results. Every title now fits.

Page headings are unchanged. A profile still opens with its full subtitle, and the cards throughout the site still carry the longer descriptive version. Only the search result and the browser tab are shorter, which is the one place where the length was costing something.

Fixed 26 pages with corrupted titles

The guides in the Future of Media series had broken characters in their titles, showing a stray letter where a dash belonged. Anyone finding those pages in a search result or looking at a browser tab saw the corruption. The cause was a text encoding fault affecting 416 characters across the 26 pages, in the body text as well as the titles. All of it is repaired.

The repair tool had been reporting no problems for months. Its list of broken characters to look for had been written from what the corruption looked like on screen rather than from what it actually was, so the list never matched anything. It does now.

New guide: OBJ, FBX and STL for a character you want to animate

The new comparison covers what each 3D format stores and what it throws away. OBJ and STL have no way to represent a skeleton, so exporting a rigged character to either does not degrade the rig. It deletes it. The export succeeds, the file opens, the model looks correct, and you find out on the day you try to move it.

FBX is the answer of the three, with one caveat worth reading. FBX moves scenes between programs by baking your setup into a form both ends understand, which loses procedural materials, constraints and the control rig you built to make animating bearable. Save your native project file as the master and treat FBX as output you regenerate. The guide also covers glTF, axis and scale traps, and why an FBX import arrives with no bones.

Page titles no longer use a dash

Every page title now reads "Page Name | Infinite Creation" instead of using a long dash. The change covers browser tabs, search results and social cards across 528 pages, and the generators build it the new way so it stays consistent.

New post: advice for first-time directors

Know What You Need, Know What You Want, and Know the Difference collects advice from working directors and checks it against how celebrated directors actually operated. Bong Joon-ho storyboards every shot of every film, and says himself the practice is not necessary for good films. Robert Altman put ninety percent of his creative work in casting. Clint Eastwood shoots first takes and finishes ahead of schedule. The mechanical sharks built for Jaws were designed for fresh water, corroded in the sea, and forced Steven Spielberg to invent the film's tension by shooting around them.

The piece also covers the one claim with hard evidence behind it. Two 2018 experiments presented identical talks in good and bad audio. Listeners rated the low quality version worse on the talk, the speaker's intelligence, the speaker's likeability, and the importance of the research. Same words, worse microphone.

Quantization explained, plus deeper microphone and green-screen guides

A new guide to quantization covers the most-used and most-misused edit in music production. Snapping notes to a grid at the default setting (nearest sixteenth, full strength) is the setting least likely to be right, and it produces exactly the lifeless quality people then try to fix with plugins. The piece explains the four controls that actually matter, why straight sixteenths sound mechanical, what quantizing audio costs that quantizing MIDI does not, and when the right amount is none.

Two existing guides got the depth they were missing rather than a competing second page. The microphone guide now treats ribbon, boundary and MEMS as types in their own right, and separates polar pattern from transducer, the distinction behind most bad microphone purchases. The history of chroma key now names who invented what: Frank Williams’s 1918 travelling matte, Petro Vlahos’s colour-difference process, and Disney’s sodium vapour system, which was better than colour keying by most measures and lost anyway because it needed a special camera. It also explains why green replaced blue once cameras went digital, which comes down to sensors carrying twice as many green photosites as red or blue.

Read the site in sixteen languages

Every page now has a Translate this page control in the footer offering machine translation into sixteen languages, and if your browser is set to one of them the site will offer it once rather than making you go looking. It offers. It never redirects you, because a link you shared should open the page you shared.

Worth being straight about what this is. The translations come from Google Translate and are rendered on Google's servers, not ours. We do not check them and they will get technical terms wrong, so the English page remains the authoritative one. Following a link sends that page's address to Google. Nothing is sent unless you pick a language, and your browser's own translation feature does the same job without involving anyone else. Code samples, filenames and timecodes are now marked so no translator mangles them. The privacy policy has been updated to cover all of it.

New post: how many layers does a door need?

An SFX editor asked people cutting mid- and big-budget features a question most sound editors have quietly wondered about, is that door on screen one well-chosen sample, or eight synced elements? The new post works through the answers from an SFX editor, a supervising sound editor, a re-recording mixer, and an editor with a decade of features behind them.

There is no layer count, and the two things that actually decide it are more useful than one: how much narrative weight the moment carries, and how much time the mixer has on the stage. The most transferable idea in it has nothing to do with sound, when something is not working, the urge to add another layer is usually information that the central choice is wrong. Swap, do not stack.

The monetization guides go deeper

The revenue framework and its four companion guides were published as outlines. They now carry the detail that makes them usable: which of the four music royalty streams is collected by whom and what you have to register with to get paid at all. The difference between your PRO and SoundExchange being the one that costs independent artists the most. What the four video distribution models actually pay and why exclusivity is priced by what it blocks. How the Universal Category System works and why metadata, not audio, is what makes a sound library sell.

The framework page now covers choosing which revenue pillars fit your position rather than listing all six, and the concentration risk that ends more independent businesses than small income does. The guide to emerging models has been made honest about where each one currently stands, including why on-chain resale royalties largely failed to pay out, with the specific conditions that would make them matter.

Twelve guides that were published but unfindable

A batch of recent guides. The monetization series, two Foley pieces, the REAPER-on-Linux setup, the podcast rules, and the After Effects series, were live and being indexed by Google, but were missing from the registry that drives the tutorials grid, the hubs, the filter pills, and site search. If you did not arrive from a search engine or a direct link, there was no way to find them. They are now filed properly and appear everywhere they should.

The older Copyright Licenses for Media Creators page has been folded into the fuller How Media Licensing Actually Works, and its address redirects there, so any existing link still lands somewhere useful.

New guide: how media licensing actually works

Most licensing trouble is not theft. It is someone who bought a licence, believed it covered what they were doing, and was wrong about which right they had bought. The new guide explains the structure underneath music, footage, and everything else you clear: why every recorded song contains two separate copyrights owned by two different parties, why putting music to picture needs two separate purchases, and why synchronisation is the one right in the whole system that nobody can compel. There is no blanket licence and no statutory rate, so a publisher can simply say no.

It also covers the traps that catch people who are being careful: footage marked "editorial use only" sitting in the same search results as commercial footage, subscription libraries whose licences may not survive cancellation, fonts licensed by use rather than by file, and the fact that a Content ID claim is a private matching system rather than a legal finding. Plus what a distributor or an insurer will actually ask you to produce.

A new section: Primary Sources

Everything on this site up to now has been a survey, written after the fact, from secondary sources. Primary Sources is for the other kind of material: first-hand accounts from the people who built the field, read closely and checked against the documentary record rather than simply repeated.

The first document is Dave Smith's 2016 Heyser Memorial Lecture, delivered to the Audio Engineering Society two weeks after Don Buchla died and six years before Smith himself did. It covers the Prophet-5, the manufacturers' meeting that nearly killed MIDI before it existed, the two decades he called the "digital dark ages", and why the designer of one of the first commercial software synthesizers went back to building hardware. Where his recollection and the record differ (a date here, a production figure there) the page says so instead of quietly picking one.

Ads on the written guides, and none on the tools

The written guides (tutorials, blog posts, and the comparison articles) now carry a single Google AdSense unit each, placed after the opening summary and before the instructions begin, so it never interrupts a procedure you are following. The interactive tools, course modules, kits, and everything else on the site stay ad-free, which is the part that mattered most in deciding where to draw the line.

The privacy policy and the notice at the bottom of the page have been updated to match. They previously said the site ran no advertising, and leaving that in place while serving ads would have been untrue. The policy now sets out what Google sets, what we receive (nothing), and how to turn off ad personalisation.

An accessibility pass over every page

We write a lot here about making media accessible, so the site itself ought to hold up. All 658 pages were checked against WCAG 2.2 twice: once for structure, and once in a real browser, because contrast is something you can only measure after the page has actually been drawn. We also tabbed through the pages to check that every stop is visible, reachable and never hidden behind the header. That is not the same as a full conformance certificate, no tool can judge whether alt text is actually a good description, or whether the focus order matches how a page reads, but everything a machine can catch has been fixed.

The browser pass was the one that mattered. Our teal had been failing the contrast minimum for normal text since the light redesign. It read at about 3:1 where the standard asks for 4.5:1, and because it was both our link colour and the fill behind white button text, it was failing in around three hundred places at once. It is now a darker teal from the same palette. A handful of leftover colours from the old dark theme went the same way, including status labels that were close to invisible on a pale background.

The other significant fix was form labels. On many of the tools the label above a field looked right but was not actually attached to it, so a screen reader announced the control with no name at all, 325 of them across the app pages. They are properly connected now, including the fields that get built as you add rows. Smaller things too: tap targets that fell under the 24-pixel minimum, links in body text that relied on colour alone and now carry an underline, and a fix so that tabbing through a page can no longer park the focused element underneath the fixed header.

Two genuine bugs turned up along the way. The spam-trap field on the roadmap suggestion form had its label pointing at the real message box, so anyone using a screen reader was told to leave it blank. Two tools had features that could only be reached with a mouse: reordering chapters in the Chapter Generator and setting the trim points in the Audio Trimmer were both drag-only, so with a keyboard they could not be done at all. The first now has move up and down buttons, the second a pair of sliders.

New tool: Flash & Photosensitivity Checker

Flashing imagery can trigger seizures, and until now there was no quick way to check for it. The new checker reads your video frame by frame, measures how much the picture brightens and darkens, and counts flashes in every one-second window against the three-per-second threshold, showing you a timeline of the result and the exact moments that go over. It runs entirely in your browser, so unreleased footage never leaves your machine. It is a screening aid rather than a broadcast certification, and the page is explicit about where its limits are.

The contrast checker is no longer hidden behind its own name

The tool called "Thumbnail Contrast Checker" was really a general WCAG contrast checker (you can type in any two colours), but the name meant people working on captions, lower thirds, or slides never found it. It is now the Contrast Checker (WCAG), and it is linked from the accessibility guides that needed it most. The web address is unchanged, so existing links still work.

Course and lesson diagrams are readable again

Labels in the workflow diagrams were running straight through each other and off the edge of the image, because the text had no line breaks and the boxes were far too narrow to hold it. The live broadcast overview was the worst, with all six step labels overlapping into an unreadable smear. Every diagram on the site has been checked and 167 labels across 87 files now wrap properly inside their boxes, with keyboard shortcuts breaking sensibly after the slash or plus rather than being shrunk. No wording changed anywhere.

New guide: using Pro Tools and Media Composer speech-to-text together

Avid put speech-to-text into both halves of its post pipeline. PhraseFind AI and ScriptSync AI in Media Composer, and a Speech-to-Text engine in Pro Tools, and this guide covers how to actually work across the two. The key thing nobody says up front: they do not share transcripts in either direction, so the AAF handoff carries no text and you transcribe twice. Also covers which versions and licences you need, why both engines running locally matters for confidential material, and why keeping dialogue as WAV or AIFF decides whether your transcript survives being handed to another editor.

New guide: what NDI is and how it is actually used

A full explainer on sending video and audio over an ordinary network instead of dedicated video cabling, how devices find each other, what separates the full-bandwidth and HX variants, how much of a gigabit link a stream really consumes, and why most NDI problems are network problems in disguise. It covers where NDI has become standard in live production and, just as usefully, the honest reasons it is not a colour-critical path in finishing. Includes comparison tables against SDI, SMPTE ST 2110, and SRT.

The site icon now shows on every page

Ten pages were missing the favicon markup, including the blog index and every blog post, so those tabs showed a blank icon. All 654 pages now carry it, and favicon.ico has been added at the site root, which is the address browsers try automatically and which was previously returning a not-found for every visit.

A global directory of guilds and unions for media professionals

A companion reference to our essay on what guilds and unions actually do: a directory of 80 organisations covering film, television, music, and media work across North America, the UK and Ireland, Europe, Asia-Pacific, Latin America, and Africa: grouped by region, with what each one actually covers and a link. It also sets out the difference between a union, a guild, an honorary society like the ASC, and a collecting society like Sacem, because reading them as equivalent is the most common way people join the wrong thing. Every link was checked before publication.

Search now finds names with accents

Searching for "Varda""Cuaron", or "Gudnadottir" returned nothing, because the search index was splitting accented words apart at each accent"Agnès" was being stored as "agn" and "s". Around half the indexed pages contain an accented character somewhere, so this affected most of the filmmaker profiles and anything mentioning cinéma vérité. Accents are now folded on both sides, so typing a name with or without them finds the same result.

New blog post: The AI Vérité Workflow

A look at an emerging hybrid documentary practice that anchors generative AI to real, unscripted footage rather than working prompt-first. The post walks through the four stages, run-and-gun capture, mining the transcript for the images hidden in how people actually talk, generating only for what no camera recorded, and style-matching the result so it belongs to the same film, and is honest about the problems: the seam between recorded and generated is the whole ethical question, consent extends past the recording, and a house style is already hardening into cliché.

Guides now point you at the free tool that does the job

The site has 91 browser tools, and a lot of guides described exactly what one of them does without ever mentioning it. The teleprompter comparison used the word "teleprompter" 37 times and never linked the free browser teleprompter. That is fixed: 147 tool links were added across 114 guides, so the guide on de-essing now points to the sibilance finder, the LUFS explainer to the LUFS checker, and the LUT guide to the LUT previewer. Eighty-three of the 91 tools are now reachable from a guide that explains the underlying task.

Browsing tutorials by subject actually works now

The filters on Tutorials & Guides had been stuck on an old set of eleven broad categories, two of which, Video and Audio, between them covered nearly two thirds of the site. Choosing one gave you a list rather than a useful shortlist. Filtering is now two levels and matches how the site is actually organised: pick one of seven areas, then narrow to a specific subject within it, with a count on every option so you can see what is there. Search and filtering are also noticeably quicker on phones, and links from a guide's breadcrumb still work exactly as before.

Thirty new guides across journalism, sports, colour, captions, development, and broadcast

The six remaining thin sections of the site have been filled out. Journalism gains filming protests safely, going live from the field, interviewing people after trauma, cutting a news package, and protecting a source on camera. Sports gains indoor venues and flicker, filming youth sport, esports production, slow motion and replay, and social clips from a match. Colour gains matching cameras, skin tones, making your own LUTs, monitor setup, and HDR delivery. Captions and access gains sign language interpretation, publishing transcripts, live captioning, localising on-screen text, and caption style guides. Development gains browser playback, WebCodecs, FFmpeg patterns, Web Audio, and preview pipelines. Broadcast gains delivery specs, loudness compliance, playout, compliance recording, and corporate live events.

Ten new guides for media teams and for teachers

Media infrastructure gains the operational side its asset-management guide had been pointing at: storage tiers and archive strategy (including why RAID is not a backup), file naming that survives, EXIF, XMP and sidecars, monitoring a transcode pipeline, and migrating an archive without losing anything. For teachers: teaching audio recording, running a student podcast that lasts, copyright and fair use for student projects, rubrics for assessing creative work, and teaching accessible media as craft.

Ten new guides on accessibility and design

Accessibility and design were the two thinnest sections on the site at five pages each. Both are now ten. New on accessibility: writing alt text that actually helps, what a video player must do to be operable without a mouse, flashing content and seizure risk, accessible social media posts, and running your first screen reader test. New on design: typography for screen, building a channel brand kit, why graphics look worse after uploading, visual hierarchy for slides and explainers, and social graphic sizes and safe areas.

Ten new photography guides, written for people who already shoot video

Photography was the thinnest section on the site: a single overview and a history piece. It now has ten practical guides covering the jobs media creators actually have: portrait lighting with one light, headshots that still read at avatar size, tabletop product shots for reviews, shooting in low light, covering an event, culling a large shoot quickly, a consistent raw developing workflow, what focal length actually changes, behind-the-scenes stills on your own shoot, and why more megapixels does not mean better photos.

Sound and colour are now separate sections, and 64 guides moved to where they belong

Sound and colour had been filed together as one subject, which meant colour-grading guides were pointing readers at a music-mixing guide for background. They are now separate sections with their own starting guides. Color Grading in DaVinci Resolve has been expanded into a full foundational treatment covering correction versus look, scopes, node structure, shot matching, LUTs and log, and delivery. Sixty-four other guides moved to a better-fitting section: lighting, lens, and camera guides now sit under Cinematography, AI and provenance material is together, and monetisation guides are all under Creator Business. Twenty filmmaker profiles that had been unreachable from any section listing are now included. Everything is easier to browse as a result.

Every discipline now has a proper starting guide, and every tutorial points to one

Fourteen existing guides have been substantially expanded into full foundational treatments of their subject, captions, colour theory, mixing, story structure, OBS, mobile journalism, classroom video, broadcast ingest, media asset management, Blender, sports multicam, accessible publishing, AI-assisted editing, and podcast setup. Between them they gained 65 new sections and 53 new questions covering the ground readers were previously sent elsewhere for. Every one of the site's 341 tutorials now also carries a short "Where This Fits" note pointing to the foundational guide for its subject, so it is always clear what a specific how-to is a part of and where to start if you are new to the area.

Eight new in-depth guides: including cinematography, scoring, and production management from scratch

Eight disciplines had plenty of specific how-tos but nothing that framed the subject as a whole, so there was no sensible place to start. Three of them (cinematography, film scoring, and production management) had no craft guidance at all, only composer and producer biographies. All eight now have a full-length foundational guide, each roughly three times the length of a standard tutorial: Video Production Fundamentals, Audio Recording Fundamentals, Cinematography Fundamentals, Still Photography Fundamentals, The Creator Business, End to End, Building Media Applications, Scoring for Picture, and Production Management. Each one covers the whole arc of its subject and links out to the detailed guides underneath it.

Five content-type hubs now actually list their guides, plus cleaner labels across 334 tutorials

The Case Studies, Myth-Busting, Troubleshooting, Behind the Tools, and Comparisons hubs were shipped as landing pages but had been sitting empty. The 50 guides that belong on them were only reachable from the main tutorials grid. All five now list their full set, and the Future of Media hub is rebuilt from the same source so it stays correct as new forecasts land. Tutorial cards and page headers also read more usefully: history, future, and profile pieces now show what they are instead of a skill level that never applied to them, and a new Professional tier separates delivery-spec and broadcast-infrastructure guides from general advanced material. Fixed some garbled punctuation on 26 pages along the way.

Synthetic Sound Generator: real synthesis engine, presets, and seamless long beds

The Synthetic Sound Generator got a full engine upgrade. All eight sound types (bass, lead, pad, kick, snare, hi-hat, ambient, and riser) are now built on the browser's native Web Audio synthesis, with real resonant filters, band-limited oscillators, and true envelopes, so they sound noticeably cleaner and richer. New in this release: four one-click presets per sound, seamless looping for minutes-long pad and ambient beds, clip-safe WAV export, and validated playback and rendering on mobile. What you hear in the live preview is exactly what downloads.

EQ Cheat Sheet now covers spoken voice for content creators

The EQ Cheat Sheet: Best EQ Settings for Vocals gained a dedicated section on tuning the settings for podcasts, YouTube, and voiceover, why presence beats the air band on phone speakers, how close-miking changes the mud cut, dynamic mic vs. condenser differences, and where loudness fits in. Two creator-focused FAQ entries were added as well.

Synthetic Sound Generator audio is now explicitly public domain (CC0)

The Synthetic Sound Generator now states its licensing clearly: every sound you generate is released into the public domain under CC0 1.0: free for any commercial or personal use, with no attribution and no royalties. A license note sits right by the download button, with the full terms in the Knowledge Base.

The Future of Media hub now lists all 26 forecasts

The Future of Media hub previously showed only a single overview guide. It now lists all 26 Future tutorials as browsable cards. Each of those tutorials also gained a link back to the hub in its Related Tools & Guides section, so you can move between any single forecast and the full collection in either direction.

60 new Filmmaker Profiles: directors, cinematographers, editors, composers, producers, and screenwriters

Sixty deep-dive profiles of celebrated filmmakers, ten each across six roles, tracing each one from student or amateur beginnings through the work that made them an auteur of cinema. At least half the roster is women, spanning nationalities across every continent with a major film industry, with a close look at each person's signature creative and technical contributions.

• New hub: Filmmaker Profiles, with six role-based sub-hubs: Directors, Cinematographers, Editors, Film Composers, Producers, and Screenwriters.
• The site now has 334 tutorials total.

New blog post: Guilds and Unions in Film and TV

A plain-language look at what the DGA, WGA, SAG-AFTRA, IATSE, and their international counterparts actually do: the difference (or lack of one) between a "guild" and a union, what a collective bargaining agreement actually gets a member, and the real tradeoffs of entry barriers, dues, and non-union competition.

• New post: Guilds and Unions in Film and TV: What They Actually Do, covers the DGA, WGA, SAG-AFTRA, and IATSE's founding and purpose, what union scale and residuals actually mean, the 2023 WGA/SAG-AFTRA strike outcomes on AI and streaming, and a practical framework for when joining is worth it.

Five new content types: Case Studies, Myth-Busting, Troubleshooting, Comparisons, and Behind the Tools, 50 new tutorials

The tutorial taxonomy expands with five new content types alongside History and Future, each with its own hub page and 10 launch tutorials: real productions deconstructed, persistent bad advice corrected, symptom-first diagnostic guides, head-to-head buying comparisons, and under-the-hood technical explainers. The site now has 274 tutorials total.

• Case Studies (10 new): how MrBeast's editing style works, NPR's Tiny Desk sound, documentary access, a viral video essay, local breaking news coverage, horror sound design, an indie game studio's audio pipeline, a podcast's real first-year numbers, a brand ad shoot, and a livestream failure post-mortem.
• Myth-Busting (10 new): the $3,000 camera myth, loudness war myths, the rule of thirds, auto settings, microphone placement, megapixels, "just add reverb" the perfect take, daily posting, and LUT magic.
• Troubleshooting & Diagnostics (10 new): muffled audio, sync drift, dropped stream frames, rolling shutter, degraded exports, green screen fringing, inconsistent podcast sound, wireless mic dropouts, videos that won't play, and clipped audio.
• Comparisons & Buying Guides (10 new): Zoom H6 vs. H4n Pro, budget wireless lavs, LED vs. softbox lighting, Blackmagic vs. Sony budget cinema cameras, stream deck hardware, cloud storage for editors, stock music libraries, podcast hosting platforms, teleprompters, and noise-cancelling headphones for editing.
• Behind the Tools (10 new): how video codecs decide what to compress, what audio normalization actually does, how autofocus tracking works, what's inside a LUT file, how noise reduction works, what happens on export, adaptive bitrate streaming, OBS scene files, auto-captioning speech recognition, and what a waveform actually shows.

25 new Future-of-media tutorials: AI video generation, virtual production, spatial audio, and more

The Future content type gets a major expansion: 25 new grounded, hedged forecasts covering where generative AI, capture technology, distribution, and production practice are realistically headed over the next several years, each separating what's already real from what's still genuinely uncertain. The Future of Media hub now covers 26 topics, and the site now has 224 tutorials total.

• New tutorial: The Future of AI Video Generation
• New tutorial: The Future of Virtual Production
• New tutorial: The Future of Spatial and Immersive Audio
• New tutorial: The Future of Synthetic Voice and Voice Cloning
• New tutorial: The Future of AR Glasses as a Content Platform
• New tutorial: The Future of Volumetric Video Capture
• New tutorial: The Future of Light Field and Glasses-Free 3D Displays
• New tutorial: The Future of Neural Video Codecs
• New tutorial: The Future of On-Device AI in Media Creation
• New tutorial: The Future of Generative and AI-Assisted Music Production
• New tutorial: The Future of Virtual Influencers and Synthetic Performers
• New tutorial: The Future of Real-Time Translation and Universal Dubbing
• New tutorial: The Future of Content Provenance and Authenticity Infrastructure
• New tutorial: The Future of Personalized and Adaptive Storytelling
• New tutorial: The Future of Live Virtual Events and Digital Concerts
• New tutorial: The Future of Autonomous Camera Robotics and Drones
• New tutorial: The Future of Digital Twins in Media Production
• New tutorial: The Future of Decentralized Content Publishing
• New tutorial: The Future of Brain-Computer Interfaces in Media Creation
• New tutorial: The Future of AI-Native Editing Workflows
• New tutorial: The Future of Procedural and Generative Game Worlds
• New tutorial: The Future of AI-Driven Accessibility by Default
• New tutorial: The Future of Synthetic Training Data for Computer Vision
• New tutorial: The Future of Satellite and Remote Production
• New tutorial: The Future of Sustainable and Climate-Aware Media Production

25 new History tutorials: cinema, television, radio, photography, animation, and more

The History content type gets a major expansion: 25 new tutorials tracing where today's tools and conventions came from, spanning film, broadcast, audio, games, and journalism. The History hub now covers 26 topics, and the site now has 199 tutorials total.

• New tutorial: A History of Cinema: From Peepshows to Streaming
• New tutorial: A History of Television Broadcasting
• New tutorial: A History of Radio Broadcasting
• New tutorial: A History of Photography
• New tutorial: A History of Animation
• New tutorial: A History of Video Game Sound and Music
• New tutorial: A History of the Podcast Medium
• New tutorial: A History of Video Codecs and Compression
• New tutorial: A History of Closed Captioning and Subtitling
• New tutorial: A History of the Music Video
• New tutorial: A History of Visual Effects in Film
• New tutorial: A History of the Microphone
• New tutorial: A History of Video Editing Technology
• New tutorial: A History of Color in Film
• New tutorial: A History of the Home Video Camera
• New tutorial: A History of Streaming Media
• New tutorial: A History of the Synthesizer
• New tutorial: A History of Sampling and Hip-Hop Production
• New tutorial: A History of Live Television and Broadcast Sports
• New tutorial: A History of Online Video: From Camcorders to YouTube
• New tutorial: A History of Film Ratings and Censorship
• New tutorial: A History of Broadcast Journalism and ENG
• New tutorial: A History of Motion Capture Technology
• New tutorial: A History of the Green Screen and Chroma Key
• New tutorial: A History of Esports Broadcasting

Ten new tutorials: drone, photography, Blender, documentary, radio, corporate video, multistreaming, color theory, plus the site's first history and future pieces

The biggest single-day breadth expansion yet: new sub-disciplines (aerial cinematography, still photography, Blender motion graphics, radio production, corporate/explainer video, multistreaming), a color theory foundations guide, and, new for the site, a dedicated history tutorial and a dedicated future-facing tutorial, opening up those as ongoing content types alongside how-to guides.

• New tutorial: Drone Cinematography for Video Creators: Gear, Rules, and Shot Craft, registration, airspace rules, camera settings, and the five aerial shot types that read as intentional.
• New tutorial: Still Photography Fundamentals: Composition, Exposure, and Light. The exposure triangle, composition, and light for creators who need photos, not just video.
• New tutorial: Blender for Motion Graphics: A Free After Effects Alternative: the interface, 2D Grease Pencil animation, and a first 3D title render.
• New tutorial: Documentary Storytelling Structure: Character, Arc, and Truth, verité vs. expository structure, and where editorial craft becomes distortion.
• New tutorial: Radio Production Fundamentals: Board Ops to On-Air Delivery, board operation, format clocks, and live delivery for podcasters crossing into broadcast.
• New tutorial: Corporate & Explainer Video: A Client-Ready Production Workflow, scoping, scripting for clarity, and managing revisions before they become scope creep.
• New tutorial: Multistreaming: Broadcasting to Twitch, YouTube, and TikTok at Once, relay services vs. local multi-output, and the bitrate math that decides which one your connection can support.
• New tutorial: Color Theory for Visual Storytellers. The color wheel, contrast relationships, and the psychology claims worth being skeptical of.
• New tutorial: A History of Recorded Sound: From Wax Cylinders to Streaming, why today's LUFS targets, album lengths, and studio conventions are the residue of a century of format history.
• New tutorial: The Next Decade of Media Creation: a grounded look at generative AI's realistic near-term role, spatial/volumetric formats, and content provenance infrastructure.

New tutorial: How to Light a Green Screen

A practical, diagram-led guide to lighting a green screen for a clean chroma key, with three original top-down and side-view setup diagrams.

• New tutorial: How to Light a Green Screen: separating the subject from the background, three-point subject lighting, matched background lights, shadow and spill control, and exposing for a flat, even key. Includes custom SVG diagrams for the full setup, shadow angles, and spill vs. distance.

A foundational cinema-sound reading list

A new reference tutorial for anyone building real theory foundations in cinema audio. An annotated bibliography spanning aesthetics, craft, sound-system engineering, and the SMPTE theatrical standards, with a working link for every title.

• New tutorial: Cinema Sound Theory: Essential Books, Essays & Standards. Chion, Weis & Belton, Altman, Lastra, Murch, Sonnenschein, Holman, Ballou, and more, plus free SMPTE standards and reports (ST 202, ST 2095-1, RP 2096, ER 0994) linked to their official downloads.

A 4:3 blog essay and a plugin download directory

A new blog post on the return of boxier framing, plus a reference tutorial that lists every capable DAW stock plugin and links the best free third-party plugins to their official download pages.

• New post: Is 4:3 Making a Comeback?: why open-gate capture, taller compositions, phone-to-TV portability, and nostalgia all point the same direction, and why it is both real craft logic and a differentiator.
• New tutorial: DAW Stock Plugins & Free Plugin Download Directory. The stock EQ/compressor/limiter sets in Logic, Pro Tools, Ableton, and Reaper, plus a linked directory of free bundles (Analog Obsession, TDR, Kilohearts, Valhalla, Melda, UAD) and 25+ standalone plugins, each pointing to its official download.

Two more DAW tutorials, a client-review guide, and a second desktop app

Closing out a batch of gaps flagged in the July 10 content audit: dedicated Logic Pro and Ableton Live workflows, a first Enterprise-pillar guide on client review, and the site's second native desktop app.

• New tutorials: Logic Pro Workflow: From Recording to Mixdown and Ableton Live: From Session View to Finished Arrangement. The two major DAWs that only had passing mentions until now.
• New Enterprise tutorial: Client Review and Approval Workflows for Video Editors: timecoded feedback, scoped revision rounds, and version handoff for paid edit work.
• New desktop app: Metadata Stripper, scans a folder of JPEG/PNG/WebP images for EXIF (including GPS location), IPTC, and XMP metadata, then writes cleaned copies, originals untouched. Free, open-source, macOS/Windows/Linux.

Creator business, spatial audio, creator safety, plus two new tools and an eighth kit

Ten new tutorials across three clusters, two new browser tools, an Accessible Publishing starter kit, a full refresh of the AI tools guide, and a sitewide structured-data fix.

• Creator-economy cluster: UGC Content Creation, Selling Digital Products, An Email Newsletter for Creators, Community and Memberships.
• Spatial & immersive cluster: Spatial and Binaural Audio, Mixing for Spatial Delivery, 360 Video Basics.
• Creator safety cluster: Account Security for Creators, Handling Harassment, Copyright Strikes and Takedowns.
• New tools: the Caption Readability Checker (per-cue reading speed, line length, timing, and overlap checks for SRT/VTT files) and the UGC Rate Calculator (usage rights, exclusivity, revisions, and rush priced into a defensible quote).
• The eighth starter kit: the Accessible Publishing Kit, an ordered path from raw auto-captions to a genuinely accessible publish.
• AI Tools for Creators guide refreshed for mid-2026, and tutorial category breadcrumbs are now real, deep-linkable filter URLs (e.g. tutorials?cat=audio), also fixing a Search Console structured-data error.

The modern-media wave: AI, Accessibility, and Localization

15 new tutorials covering how media creation actually changed in 2025 to 26: two new pillars (AI & Synthetic Media, Accessibility) and a localization series, all written with the site's usual honesty about what the technology can and can't do.

• New AI pillar: Content Credentials (C2PA), Text-Based Video Editing, AI Voice Cloning: Consent and Contracts, When to Label Content as AI-Made, A Practical AI-Assisted Edit, Protecting Your Work in the AI Era.
• New Accessibility pillar: Audio Description for Video, Beyond Auto-Captions, Publishing Accessible Video, Accessible Podcasting, Accessible Thumbnails and Graphics.
• New localization series: Multi-Language Audio and Auto-Dubbing on YouTube, Translating Subtitles, Dubbing Workflows, Running a Global Channel.

Three new audience pillars: Education, Journalism, and Sports

A new-verticals wave aimed at audiences the site had zero coverage of before today: teachers and students, reporters, and local sports broadcasters. 17 new tutorials across three brand-new pillars, each launched at a full five-or-more tutorials.

• New Education pillar: Teaching Video Production, Running a Student Film Project, Setting Up a School Media Lab on a Budget, Teaching Media Literacy Through Production, Recording Lectures Students Actually Watch, Building a Student Media Portfolio.
• New Journalism pillar: Mobile Journalism, Interviewing for Journalism, Press Conference and Scrum Audio, Verifying Found and User-Submitted Video, A Small Newsroom Video Workflow, Recording in Public: Ground Rules.
• New Sports pillar: Livestreaming a Local Sports Game, Multi-Camera Coverage for Live Sports, Commentary Audio, Score Overlays and Graphics, Cutting Sports Highlights Fast.

A second desktop app, full FAQ coverage, and two tutorial pillars doubled

The Desktop Tools line grows to two native apps, consolidated into one monorepo. Every tutorial on the site now has an FAQ section, and the Enterprise and Developer pillars each grew to five tutorials.

• New desktop app: Bulk Renamer: pattern-based batch file renaming (find/replace, sequence numbers, case conversion) with a preview-before-apply safety check, free and open-source for macOS, Windows, and Linux.
• The Desktop Tools line now lives in one consolidated monorepo, alongside the Batch Timecode Calculator.
• Every one of the site's 111 tutorials now has an FAQ section with matching structured data, closing a gap the content audit flagged across 39 pages.
• New Enterprise tutorials: Automated QC for Broadcast Delivery, CDN Architecture for Video at Scale, bringing the pillar to five.
• New Developer tutorials: Containerizing Media Processing Tools, API Design for Async Media Processing Jobs, Handling Large File Uploads in Web Applications, bringing the pillar to five.
• The three apps mislabeled with a "Live" badge (Podcast Hosting Guide, Remote Recording Apps, Microphone Buyer's Guide) now carry an honest "Guide" badge instead.

A Voice-Over Artist Kit, a Final Cut Pro comparison, and five new tutorials

A content-audit-driven wave focused on filling gaps: a seventh starter kit, a new head-to-head comparison, and tutorials covering monetization, backup discipline, short-form growth, audio restoration, and color management.

• New starter kit: Voice-Over Artist Kit, seven ordered steps from a home booth to a first paid booking, bringing the total to seven.
• New comparison: Final Cut Pro vs. Premiere Pro: cost model, platform support, and timeline paradigm.
• New tutorials: How YouTube Monetization Actually Works, A Backup and Archival Workflow That Actually Protects Your Footage, Getting Discovered on Instagram Reels and TikTok, Audio Restoration Beyond Noise Removal, and ACES and Color Management.

A video editor, a native desktop app, three more starter kits, and nine advanced tutorials

An SEO-driven wave: real in-browser video editing, the site's first native desktop app, and content aimed at professional and enterprise-adjacent audiences for the first time.

• New app: Video Clip Editor. Upload multiple clips, trim and reorder them on a zoomable timeline, add titles, then export one MP4 with real FFmpeg encoding, entirely in the browser.
• New desktop app: Batch Timecode Calculator, the site's first native app (macOS, Windows, Linux), free and open-source, with a signed cross-platform release pipeline.
• New starter kits: Audio Post-Production Kit, Livestream Starter Kit, Short-Form Video Kit, bringing the total to six.
• New advanced tutorials: Removing Wind Noise in iZotope RX, Navigating DAW License Structures, Building a Pro Tools Session Template, Automating Stem Exports
• Two new tutorial pillars: Enterprise (Broadcast Ingest, Transcoding at Scale, Media Asset Management) and Developer (Automated Testing for Media Applications, CI/CD for Cross-Platform Builds).
• The homepage's Methodology section is now a real, linkable anchor with its own structured data, and the Tool Stack table links out to the relevant tutorials.

Camera scopes, business tools, and curated starter kits

Five new camera and colour apps, three new publishing/business tools, three curated starter kits, a 140+ term glossary, and four head-to-head comparison pages.

• New apps: False Color Exposure Checker, Vectorscope & RGB Parade, LUT Previewer, Flicker-Free Shutter Calculator, Time-Lapse Calculator, Call Sheet Generator, Podcast CPM Calculator, Channel Banner Safe-Area Checker
• New starter kits: Podcast Starter Kit, YouTube Creator Kit, Home Studio Kit
• New reference pages: Media Production Glossary, Riverside vs. Zencastr, REAPER vs. Audacity, DJI Mic vs. Rode Wireless GO, H.264 vs. H.265
• Every page on the site now has a "Related Tools & Guides" block for easier cross-discovery.

Sitewide search, embeddable widgets, and offline support

Four infrastructure features that make the existing catalogue more useful, not more numerous.

• Sitewide search, one box across every page, reachable from the nav on every page.
• Suggest a Tool + public roadmap. A direct line to request the next tool or tutorial.
• Embeddable widgets, 21 calculator tools now offer a copy-paste iframe snippet for embedding elsewhere.
• Offline support. The site is now installable as a PWA, and any tool you've already opened keeps working with no signal.

20 new tutorials and 7 real-time audio apps

A full pass on music production, deeper audio engineering, podcasting/camera craft, and business tutorials, plus real-time Web Audio tools that reuse the site's existing FFT and DSP engines.

• New tutorials spanning mixing, mastering, synthesis, sound design, field recording, saturation, limiters, mid-side processing, interview technique, video podcast setup, camera exposure and focus, and freelance/licensing topics.
• New apps: Live Spectrum Analyzer, Audio Trimmer & Fade Tool, Stereo Width Checker, Voice Pitch & Range Analyzer, Speaking Pace Analyzer, Metronome & Tap Tempo, dB Converter
• Sitewide SEO fixes: social share images on every page, FAQ rich-result markup, breadcrumbs, and a performance pass on shared fonts/scripts.

50 new pages: livestreaming, captions, and a full business layer

The catalogue's first major expansion: 25 new tutorials and 25 new apps covering livestreaming setup, captions and accessibility, short-form platform strategy, editing craft, and freelance business tools.

• New pillars added: Livestreaming, Publishing & Platform, and Business.
• Real Web Audio/canvas processing shipped for the heaviest apps: BS.1770 loudness measurement, FFT-based sibilance detection, and true drop-frame timecode arithmetic.

Infinite Creation launches

The original catalogue goes live: free tutorials and interactive browser tools for filmmaking, podcasting, audio engineering, voice over, and motion graphics, no login, no paywall.