Video Editing Protocol

Internal workflow for classifying source media, selecting the editing stack, producing destination-specific formats, and blocking any output that has not passed synchronization, caption, image-quality, motion, and playback checks.

Release state: Private review first
Sync standard: Technical + perceptual
Source rule: Classify before editing

Source first

Source-to-output decision matrix

Capture origin and phone orientation are separate facts. Detect what the file proves; never guess what metadata cannot prove.

Capture path	Orientation	Source of truth	Reels / Shorts output	Synchronization rule
Loom mobile app Approved Sequential Domination precedent	Landscape — phone sideways	Best single muxed Loom rendition containing paired video + audio	1080×1920 vertical canvas; fit the complete landscape frame without crop or enlargement; thesis above and captions below	Measure clean onsets early/middle/late. This approved source required +200 ms audio and the identical +200 ms caption shift; do not inherit that number without a new canary.
Loom mobile app	Portrait — phone upright	Best single muxed Loom rendition containing paired video + audio	Native 1080×1920 portrait; fixed source-authentic frame; protect face, shoulders, hands, and platform safe zones	Classify aligned, constant offset, or drift. Never assume the landscape precedent’s +200 ms value.
iPhone Camera → uploaded to Loom	Landscape — phone sideways	Original iPhone file; Loom is transport/review unless original-byte identity is proven	1080×1920 vertical with complete landscape frame fitted, or an explicitly approved fixed wide crop; no tracking or blur filler	Test the original iPhone file first, then compare Loom playback/download. Use the original if Loom introduces offset, drift, or compression.
iPhone Camera → uploaded to Loom	Portrait — phone upright	Original iPhone file; preserve native orientation and paired camera audio	Native 1080×1920 portrait; no invented landscape canvas, head-only crop, or automatic face tracking	Baseline-test original and Loom copy. Any correction is source-specific and moves captions identically.
Loom desktop — office	Landscape 16:9	Single muxed Loom original	1920×1080 landscape by default; vertical derivative only with Brad’s approved fitted-landscape or fixed wide-crop treatment	Run the same original-source onset test; frame/sample lock is required but is not perceptual sync proof.
Original file from Drive/local storage	Measured from the actual file	Highest-resolution camera original or first-generation export with paired audio and documented provenance	Match the requested destination without unnecessary crop, enlargement, frame-rate conversion, or intermediate re-export	Establish baseline sync on the original before any edit; identify camera-original, screen recording, or prior export.
Other camera or screen recording	Measured from pixels and rotation metadata	Original card/file or lossless first-generation capture	Select landscape, portrait, square, or fitted-canvas treatment from source and destination—not filename	Check variable frame rate, separate microphone clocks, screen/camera tracks, and drift before normalization.
Externally edited export	Measured	Original project/source media preferred; use the export only when upstream assets are unavailable	Avoid compounding crop, captions, sharpening, denoise, or compression	Treat embedded edits and timing as inherited risk; compare against the earliest accessible source.
YouTube/social-platform download	Measured	Last-resort recovery source only	Do not upscale or represent transcoded pixels as original quality	Document platform-transcode limitations and block release if a better source should exist.
Continuous “Apple” batch	Inherit parent orientation	Same parent source path and checksum; one child manifest per script	Each child inherits its parent composition, caption zone, and delivery target	Classify parent sync before splitting; preserve the same paired correction through every child; verify each boundary in motion.

Mandatory intake manifest: capture origin, capture app, iPhone Video-or-Cinematic mode, selected HD-or-4K resolution, selected 24-or-30 frames per second, Loom-or-iPhone source of truth, measured orientation, delivered dimensions and frame rate, audio rate, duration, muxed/split status, intended destination, framing profile, sync classification, correction milliseconds, caption offset, and source checksum. Preserve selected camera settings separately from delivered-file metadata because Loom may transcode the upload. If metadata cannot prove a field, ask Brad once—do not infer it.

Approved precedent

Loom landscape → vertical Reel

The exact Sequential Domination treatment Brad approved for resolution and synchronization.

Source1294×720 muxed LoomDirect Loom-mobile landscape capture; complete paired rendition.

Canvas1080×1920Full frame scaled to 1080×600 at y=260 on charcoal; no crop or enlargement.

Synchronization+200 ms audioSource-specific approved correction; captions shifted by the identical +200 ms.

EncodeH.264 CRF 16High profile, slow preset, 30 fps; AAC 256 kbps at 48 kHz.

Caption continuity: Arial Bold; 58 px at 1080×1920; two lines maximum; controlled phrase grouping; white base text with high-contrast black backing/outline. Use yellow and cyan only for semantic emphasis—not random rotation. Keep the thesis in the upper safe zone. For fitted landscape, keep spoken captions below the picture; for native portrait, select a stable upper or lower safe zone that does not cover Brad’s face or gestures. No bouncing word-by-word animation.

Required social-copy CTA: Every edited social-video caption includes a natural, topic-specific lead-in followed by: “Comment AUTOMATE and I’ll send you the Automate and Delegate newsletter.” Keep the keyword uppercase and verify the destination automation before release.

Required order

The six-gate workflow

No stage is skipped because the volume is high. The system makes volume safer.

Intake

Lock the synchronized Loom source

Inventory Loom’s renditions and prefer the single muxed file containing its paired video and audio. Record URL/file, source type, dimensions, rates, duration, and SHA-256. Separate HLS tracks are diagnostic fallbacks—not the default master.

Identity + pairing proof

Normalize

Measure before resetting clocks

Probe original timestamps and inspect visible speech before creating a 30 fps CFR / 48 kHz mezzanine. Never independently zero, stretch, or remux audio and video and then call equal starts proof of sync.

Timing foundation

Plan

Transcribe, compare tools, and select complete thoughts

Use faster-whisper locally. When authenticated and useful, compare OpusClip clip, hook, and reframe candidates plus Captions.ai caption treatments. Select from source evidence; preserve chronology and natural delivery.

Tool-assisted judgment

Render

Use one paired A/V edit decision list

Every cut uses one shared boundary. If clean onset testing proves a constant source offset, apply one documented global correction and shift caption timing by the identical amount.

Coupled sync control

Verify

Prove lips, sound, motion, pixels, and captions

Deliver a 20–30 second motion canary containing a clean speech onset. Check early, middle, CTA, and ending alignment; distinguish constant offset from drift; then run image, caption, decode, and safe-zone checks.

Brad confirms sync

Release

Stop at private review and platform HD

Publishing begins only after Brad approves the exact master. Keep YouTube unlisted until the intended HD rendition exists and plays; provider “processing succeeded” alone is not release proof.

Approval + playback gated

Tool selection

Editing capability matrix

Use the strongest suitable tool at each stage. No third-party convenience render becomes the synchronization or quality authority.

Tool	Current route	Use in workflow	Control
Loom rendition audit	Source-dependent	Acquire the best muxed Loom file and compare alternatives when diagnosing	Split HLS tracks are never assumed synchronized
ffprobe + SHA-256	Available locally	Dimensions, orientation, streams, clocks, duration, codec, and immutable identity	Required at intake and final master
FFmpeg	Available locally	Normalize, trim paired streams, correct measured offset, compose, caption, encode, and decode-check	Authoritative deterministic finishing layer
faster-whisper	Available locally	Word-timestamp transcription and caption timing	Text and timing require human review
HyperFrames	Available locally — v0.8.25	Designed graphics, motion, reframing, and advanced media treatment when requested	Optional; never add overlays or effects that violate the approved treatment
OpusClip	Browser/API authorization must be re-proven	Candidate clip discovery, hook scoring, reframing comparison, and draft captions	Never the final sync, crop, caption, or export authority
Captions.ai / caption editor	Browser/API authorization must be re-proven	Caption-style alternatives and draft packaging when useful	Final words, timing, colors, placement, and safe zones are checked against source
yt-dlp	Available locally	Permitted source retrieval and platform-format inventory	Never use a social re-download when a better original exists
Lip-onset canary	Required human review	Clean speech resumptions early/middle/late; constant offset separated from drift	Brad’s playable-canary confirmation is the sync gate
Vimeo	Private review route proven	Full-quality review delivery and playback verification	Unlisted/private before release
YouTube / Metricool / native tools	Post-approval only	Destination upload, status, high-definition rendition, and native playback verification	Exact approved checksum and destinations only

Universal exclusion: No AI-generated visuals, avatar replacement, unrelated B-roll, blur filler, detached voice, fake camera movement, or unapproved music.

Permanent capture reference

iPhone and Loom capture settings

Record what Brad selected before filming separately from what Loom later delivers. “Frames” means frames per second (fps), not frames per minute.

Recommended baselineVideo · 4K · 30 fpsNative iPhone Camera, Standard Video mode. Keep the original iPhone file as source of truth; upload to Loom for transfer and review.

Convenience fallbackLoom mobile cameraFaster capture-to-share path. Loom exposes camera selection but does not document a guaranteed resolution or frame-rate control in its current iOS recording guide.

Motion rule30 fps for walkingSmoother cadence for handheld movement and gestures. Use 24 fps only as an intentional look after a controlled comparison.

Mode ruleStandard Video firstNo simulated depth, focus-rack, or Cinematic edge artifacts. Cinematic remains an optional controlled test—not the default.

Two truths

Selected settings versus delivered file

These fields must never be collapsed. A 1080p Loom download does not prove the iPhone was set to HD; a 4K original may have been downsampled.

Field	Capture truth	Delivered-file truth	How to prove it
Capture app	iPhone Camera or Loom mobile	May not survive export metadata	Brad confirmation, screenshot, and original-file metadata when available
Camera mode	Standard Video or Cinematic	Ordinary exported pixels may not prove the selected UI mode	Pre-recording screenshot plus original iPhone asset metadata
Selected resolution	HD or 4K	Measured width × height after Loom processing	Screenshot/original metadata versus ffprobe on the Loom rendition
Selected frame rate	24 or 30 fps	Measured average and nominal frame rate	Screenshot/original metadata versus ffprobe; record variable-frame-rate behavior
Orientation	Portrait or landscape	Measured dimensions plus rotation metadata	Probe the actual file; do not infer from capture app
Synchronization	Camera/microphone behavior at capture	May change after upload or platform processing	Early/middle/late canary on both original and Loom-delivered copy

Current evidence

Observed Brad sources

Empirical results are source-specific—not universal Loom specifications.

Source	Known capture settings	Loom-delivered file	Synchronization result	Interpretation
No Tolerance for Bullying and Modeling September 3, 2026	iPhone Camera · Standard Video · landscape · HD · 30 fps Brad-supplied Camera screenshot shows the selected Video configuration; Loom delivery corroborates 1080p/30	1920×1080 HD · H.264 · 30 fps · SDR/BT.709	Zero-correction early/middle/late canary approved by Brad	Upload path retained strong 1080p delivery and perfect observed sync; original iPhone file remains the higher-authority source
Sequential Domination Approved historical precedent	Direct Loom mobile · landscape	1294×720 muxed rendition · approximately 30 fps	Required a source-specific +200 ms audio and caption correction	Useful measured precedent, but not proof that every direct Loom recording uses the same quality or offset

Current Loom documentation boundary: Atlassian’s official “Record with Loom for iOS” guide, last modified May 27, 2026, documents front/rear camera selection and camera recording but does not state a guaranteed mobile-camera capture resolution, bitrate, codec, or frame rate. Treat those values as unknown until the resulting file is measured. Open the first-party Loom guide.

Controlled comparison

Determine the best repeatable path

Use the same scene, lens, orientation, lighting, microphone, movement, and spoken test so only the capture path changes.

Fastest path

Direct Loom mobile

Record 15–20 seconds with a clap, clean consonants, head turn, hand motion, fine desk detail, and natural movement.

Measure resulting file

Controlled HD

iPhone Video · HD · 30 fps

Keep the original file, then upload the same take to Loom. Compare original and delivered resolution, bitrate, sync, texture, and transfer time.

Efficiency baseline

Quality baseline

iPhone Video · 4K · 30 fps

Keep the original file, upload to Loom, and compare downsampled detail, motion, compression, transfer time, and edit headroom.

Recommended starting standard

Optional look test

iPhone Cinematic · 4K · 30 fps

Run only after Standard Video is established. Inspect hair, hands, glasses, background edges, focus changes, stabilization, and metadata compatibility.

Not the default

Acceptance decision: standardize on the lowest-friction path that passes original/delivered detail, early/middle/late lip sync, natural motion, stable focus, transfer time, edit reliability, and final-platform 1080p playback. Until the controlled comparison is complete, use iPhone Standard Video at 4K/30 fps for quality-sensitive filming and retain the iPhone original.

Screenshot evidence

Brad’s iPhone setting screens

Original Telegram image bytes preserved September 3, 2026. The selected settings are identified by yellow dots in the Camera interface.

Screenshot 01

Standard Video · HD · 30 fps

Brad’s iPhone Camera in Standard Video mode with HD and 30 frames per second selected — Options shown: HD or 4K; 24, 30, 60, or 120 fps; 0.5×, 1×, 2×, 4×, and 8× lenses. Active: HD, 30 fps, 1×.

SHA-256 cd879fe7…204f5

Screenshot 02

Cinematic · HD · 30 fps

Brad’s iPhone Camera in Cinematic mode with HD and 30 frames per second selected — Options shown: HD or 4K; 24 or 30 fps; 1× and 2× lenses. Active: HD, 30 fps, 1×.

SHA-256 af6a3e2b…75630

Evidence rule

Preserve capture and delivery separately

These screenshots prove the demonstrated Camera configurations. For each recording, retain the original iPhone asset and probe both it and the Loom-delivered copy before claiming the selected setting survived unchanged.

Stored: /filming/assets/capture-settings/

Source type 03

One long recording, many “Apple”-separated shorts

This is the default batch method for scripts under three minutes.

Recording syntax

Use one unmistakable marker

Finish the current script completely
Pause for roughly one second
Say “Apple” by itself
Pause again, reset posture, then start the next hook
Never use “Apple” casually inside a script without flagging it

Segmentation

Marker creates a boundary—not content

Detect isolated “Apple” from word timestamps
Review every marker against audio waveform and video
Remove the marker plus surrounding reset time
Create one child manifest and checksum per script
Keep original chronology and source identity

Per-video edit

Treat each child as a complete story

Hook, point, proof/example, takeaway, CTA
Remove false starts and repeated versions
Remove excess examples before cutting useful nuance
Cut on complete phrases with visual handles
Render and QC each child independently

Batch safeguards

How Apple splitting fails closed

Marker confidence

An isolated transcript match plus surrounding silence proposes the split. Low-confidence transcription or an in-sentence use requires manual review.

Never guess

First-motion proof

Each child must show real frame variation during its first three seconds. A held poster, stale frame, or repeated frame fails—even when audio plays.

Freeze gate

Boundary playback

Review the final five seconds before every marker and first five seconds after it. No marker audio, reset posture, prior-script tail, or clipped first word may remain.

Clean children

Independent release identity

Each short receives its own title, transcript, EDL, captions, checksum, QC packet, and approval state. Duplicate titles or uploads trigger a release stop.

No batch drift

Current historical audit: The live YouTube API identifies an August 5 five-video batch, including two uploads titled “AI Won’t Save Your Business.” That duplicate and all five openings remain a playback-QC audit set; API titles prove inventory, not motion quality.

August 5 live title	Runtime	Audit status
AI Won’t Save Your Business (Here’s What Will)	111 sec	Two live IDs; compare master identity and opening playback
Your VA Is Working 3x Harder Than They Need To	83 sec	Opening motion, crop, captions, sync
My 5-Minute Monday Message That Runs My Whole Week	74 sec	Opening motion, crop, captions, sync
$200K/Year vs $2K/Month (Same Output or Better)	105 sec	Opening motion, crop, captions, sync

Editing standard

Editorial cut protocol

Remove waste at idea boundaries. Do not create robotic speech by deleting every filler word.

Remove first

High-value cuts

Silence above the chosen threshold
False starts and full sentence restarts
Repeated explanations of the same point
Excess examples after the point is proven
Off-topic setup and trailing reset time
Apple markers and separator pauses

Preserve

Natural delivery and source context

Complete thoughts and chronological logic
Useful nuance and distinctive examples
Natural breath and conversational pacing
Shoulders, gestures, and source context
Spoken CTA and true conclusion when applicable
Original voice attached to original video

Never normalize

Failure patterns

Frozen first frame with live audio
Detached audio or lip-sync drift
Face-only crop from landscape
Fast moving crop tracker
Word-by-word filler surgery
Social re-download used as source

Technical contract

Container lock + perceptual sync

Equal stream starts and durations prove muxing structure—not that spoken sound matches visible lips.

Video clock30 fps CFRInteger start/end frames only.

Audio clock48,000 HzPCM editing mezzanine.

Shared boundary1,600 samples/frameAudio sample = video frame × 1,600.

TimelineOne paired EDLChronological video and audio concatenated together.

Non-negotiable distinction: ffprobe, matching zero timestamps, equal durations, checksums, contact sheets, and successful decode cannot prove lip sync. The rendered video must be watched in motion at clean consonant/speech onsets.

Offset rule: Check at least three clean speech onsets spanning early, middle, and late source time. A consistent offset may receive one documented global correction. A changing offset is drift and blocks release until the source is reacquired or a separately tested clock correction is approved.

Quality preservation

Required proof packet

Check	Pass requirement	Why it exists
Source identity	Original checksum, Loom provenance, and rendition choice recorded	Prevents compressed or accidentally detached replacement sources
Loom rendition audit	Muxed source preferred; split HLS tracks compared only for diagnosis	Prevents separate-stream pairing assumptions
Technical probe	Expected resolution, CFR, 48 kHz, starts and durations recorded	Defines container timing without pretending it proves lip sync
Opening-freeze detection	Real frame variation during first three seconds	Blocks frozen-poster failures
PSNR / SSIM	Composition-matched comparison shows no material degradation	Quantifies image retention
Cut-point review	Every edit heard and watched at normal speed	Finds clipped words and visual jumps
Baseline lip-sync	Original Loom watched at clean early/middle/late speech onsets	Separates source offset from edit/render damage
Offset classification	Constant offset measured in milliseconds or variable drift explicitly blocked	Prevents guessed delays and one-point false confidence
Delivered motion canary	20–30 second file is accessible and Brad confirms lips match sound	A local file path or still image is not human review
Full-master lip-sync	Opening, clean onsets, every cut neighborhood, middle, CTA, and ending watched	Catches offset or drift a probe cannot see
Caption review	Words, line breaks, safe zones, and any audio-delay shift manually checked	Captions must follow corrected audio—not stale source time
Full decode	Zero FFmpeg decode errors	Blocks damaged exports
Platform HD playback	Target high-definition rendition exists and is watched after transcode	Upload success or “processing succeeded” can still expose low-quality playback
Checksum handoff	Approved master hash equals uploaded master hash	Prevents wrong-file release

Quality truth: Any edit that changes crop, scale, captions, or compression must be encoded. “No degradation” therefore means no material visible loss from the best composition-matched source reference—not a false claim of byte-identical pixels.

Release stop: Source offset not classified, inaccessible canary, missing Brad sync confirmation, drift, frozen opening, clipping, incorrect crop, avoidable enlargement, unreadable captions, unavailable platform HD, checksum mismatch, or absent Brad approval stops publication. If Brad rejects a live edit, prior approval is revoked and that version must be removed or privatized before replacement.

Copy and paste

Prompts that reactivate the exact process

Replace the bracketed values. Each prompt keeps publication locked until you approve the private master.

Classify uploaded source

Run before every edit

CLASSIFY THIS VIDEO SOURCE BEFORE EDITING

Source: [LOOM URL OR ORIGINAL FILE]
Intended destination: [REELS / SHORTS / YOUTUBE LANDSCAPE / OTHER]

Return a source manifest with:
• capture origin: Loom mobile app / Loom desktop / iPhone Camera uploaded to Loom / unknown
• capture app and mode: iPhone Standard Video / iPhone Cinematic / Loom mobile camera / unknown
• selected capture resolution: HD / 4K / unknown; keep this separate from delivered-file dimensions
• selected capture rate: 24 / 30 frames per second / unknown; keep this separate from measured output fps
• what evidence proves the origin; if metadata cannot prove it, ask Brad once
• orientation: landscape or portrait from measured dimensions/rotation metadata
• source of truth: muxed Loom rendition or original iPhone file
• dimensions, frame rate, audio rate, duration, stream starts, muxed/split status, and SHA-256
• framing profile and caption safe zones for the requested destination
• baseline lip-sync classification from clean early/middle/late speech onsets: aligned / constant offset / drift
• proposed correction milliseconds; never inherit another video’s offset
• caption offset equal to the approved audio correction
• required 20–30 second accessible canary and QC gates

Do not edit or publish until the manifest is complete.

Loom desktop edit

True landscape by default

EDIT THIS VIDEO — LOOM DESKTOP

Source: [LOOM URL OR ORIGINAL FILE]
Inventory Loom renditions and use the best single muxed video+audio file, not separately merged HLS tracks, YouTube, or a social download.

Follow the Brad–Sterling2 Video Editing Protocol:
1. Record source checksum and ffprobe evidence.
2. Preserve the authored 16:9 office/screen composition.
3. Before normalization, watch clean speech onsets early/middle/late and classify source sync as aligned, constant offset, or drift. Matching timestamps are not lip-sync proof.
4. Normalize to 30 fps CFR and 48 kHz PCM without independently resetting or stretching paired streams.
5. Build one chronological frame/sample-locked A/V EDL. If a constant correction is proven, document it and shift captions by the exact same amount.
6. Remove silences, false starts, repeated explanations, excess examples, and off-topic material at complete-thought boundaries.
7. Use the caption system: Arial Bold, two lines maximum, white base with semantic yellow/cyan emphasis, high-contrast backing, and a stable source-safe placement. No random color rotation or bouncing words.
8. Do not make a head-only crop, detached voice, blur background, B-roll, AI imagery, music, or tracking movement.
9. Deliver an accessible 20–30 second motion canary with a clean speech onset; require Brad to confirm sync before the full render.
10. Run opening-freeze, multi-point lip-sync, caption, PSNR/SSIM, decode, and checksum QC.
11. Draft platform copy with a natural topic-specific lead-in and the exact CTA: “Comment AUTOMATE and I’ll send you the Automate and Delegate newsletter.”
12. DO NOT PUBLISH until I approve the exact private master. On YouTube, keep it unlisted until the target HD rendition exists and plays.

Loom mobile edit

Measured orientation preserved

EDIT THIS VIDEO — LOOM MOBILE

Source: [LOOM URL OR ORIGINAL FILE]
Use the original mobile Loom master.

Use the single muxed Loom source when available. Preserve the measured source orientation: portrait remains native portrait; landscape uses the approved fitted-frame vertical treatment for Reels/Shorts. Preserve natural headroom, shoulders, gestures, and movement. Use fixed source-authentic framing—no face tracker or synthetic punch-ins. Before normalization, check clean early/middle/late speech onsets and classify aligned, constant offset, or drift. Create one paired A/V EDL; if a constant delay is approved, apply it once and shift captions identically. Remove complete redundant thoughts, false starts, excess examples, and silence without robotic word surgery. No blur, B-roll, AI visuals, cards, or music.

Use Arial Bold captions with two lines maximum, white base plus semantic yellow/cyan emphasis, high-contrast backing, and a stable safe zone selected for the measured orientation. Deliver an accessible 20–30 second motion canary and require Brad’s sync confirmation. Then prove real opening motion, perceptual A/V alignment early/middle/CTA/ending and at every cut, coupled caption timing, image retention, full decode, and checksum identity. Draft platform copy ending exactly: “Comment AUTOMATE and I’ll send you the Automate and Delegate newsletter.” DO NOT PUBLISH until I approve that exact file.

Continuous Apple batch

Many shorts from one recording

PROCESS THIS CONTINUOUS APPLE-MARKER RECORDING

Source: [LOOM URL OR ORIGINAL FILE]
Base orientation: [DESKTOP LANDSCAPE / MOBILE PORTRAIT / DETECT FROM SOURCE]

I recorded multiple complete videos in one continuous take and said “Apple” by itself between videos.

1. Transcribe with word timestamps.
2. Detect isolated “Apple” markers and verify each against the waveform and neighboring video.
3. Remove every marker plus reset pauses; never include it in a child video.
4. Create one child source manifest, transcript, EDL, caption file, checksum, and private master per complete script.
5. Edit each child independently for hook, point, example/proof, takeaway, and CTA.
6. Remove false starts, repeated versions, excess examples, and silence at complete-thought boundaries.
7. Preserve the base source composition and use one paired frame/sample EDL.
8. Fail any child with a frozen opening, clipped first word, prior-script tail, wrong title, duplicate identity, sync concern, or quality loss.
9. Return a batch inventory, canary/contact sheet, and private masters.
10. Draft unique platform copy for every child ending exactly: “Comment AUTOMATE and I’ll send you the Automate and Delegate newsletter.”
11. DO NOT PUBLISH any child until I approve its exact checksum.

Strict QC before release

Use after the edit is ready

RUN STRICT BRAD–STERLING2 VIDEO EDITING QC

Master: [PRIVATE MASTER PATH OR LINK]
Source: [ORIGINAL SOURCE PATH OR LINK]

Verify:
• Source/master checksums and provenance
• Loom rendition choice; single muxed source preferred over separately merged HLS tracks
• Expected resolution, 30 fps CFR, 48 kHz, starts and durations recorded
• Original-source lip sync at clean early/middle/late speech onsets
• Constant offset measured in milliseconds or variable drift blocked—never infer sync from timestamps
• Opening has real motion—not a held or repeated frame
• Every cut at normal speed for clipped words, visual jumps, and lip sync
• Opening, middle, CTA, and ending A/V alignment watched in motion
• Accessible 20–30 second canary with Brad’s explicit sync confirmation
• Captions for exact words, spelling, line breaks, safe zones, and identical shift when audio timing is corrected
• Source-authentic framing with hair, shoulders, gestures, and context safe
• Composition-matched PSNR/SSIM plus human pixel review
• Full decode with zero errors
• Platform-specific HD rendition available and actually playable; upload/API processing status alone does not pass
• No AI visuals, unrelated B-roll, blur filler, tracking crop, detached voice, or unapproved music

Return PASS/FAIL per gate, the exact approved-candidate checksum, and every blocker. DO NOT PUBLISH.

YouTube historical audit

Find failed prior edits

AUDIT MY RECENT EDITED VIDEOS ON YOUTUBE

Use the channel-bound YouTube API to inventory titles, IDs, dates, durations, privacy state, and duplicates. Then inspect actual playback—not only thumbnails—during the opening, each known cut area, middle, CTA, and ending.

Classify every video as:
• Gold reference
• Technically correct but not directly approved
• Failed/replaced
• Needs review

Specifically detect frozen first frames with live audio, lip-sync drift, face-only crops, tracking jumps, blurred filler, caption defects, duplicate uploads, wrong titles, and mismatched master identity. Do not delete, replace, or publish anything. Return the evidence matrix first.

Approved release

Only after private-master approval

APPROVED RELEASE — EXACT MASTER ONLY

Approved master: [PATH OR LINK]
Approved SHA-256: [CHECKSUM]
Platforms/accounts: [EXACT DESTINATIONS]
Approved copy: [TITLE / CAPTION / DESCRIPTION]

Release only this exact checksum. Do not regenerate, re-edit, replace captions, change compression, or substitute a derivative without stopping. Keep YouTube unlisted until its intended HD rendition exists and plays. Verify each native platform at normal-speed playback for resolution, first motion, clean-onset lip sync, captions, and ending—not only receipt/status. If Brad rejects the release, treat approval as revoked and remove or privatize it before replacement. Return native IDs, direct links, checksum proof where available, and platform-specific transcode results.