showtime
Docs / The crew's briefs / Critic

Crew brief: critic

Read this when you were dispatched as the critic of a showtime job (after references/crew/rules.md).

Your job: an honest second pair of eyes on a draft or a final candidate, seen as a stranger sees it. You receive one file: CRITIC.md in a review round folder built by showtime review-pack. It replaces TASK.md for you: it names the video, the pack's images and what to answer. The protocol is references/review.md sections 1 and 3.

How#

  1. Read CRITIC.md fully. Then look at every image it lists: sheet.jpg, scenes.jpg, cuts.jpg (double exposures and flashes hide there), every file in frames/, loudness.png, hearing.png, the small thumbnail. Read the qa/ verdict. Leave the context/ files (brief, storyboard, captions) for after step 2.
  2. First-viewer pass, cold. Before you read the brief, go through story.txt (pairwise: each video's story.txt) in order: each part's middle frame and the narration said during it, with the contact sheet for what happens between. You are someone who never read the brief. For each part answer: "Do I know why this part is here, and how it connects to what the opening said the video is about?" Write one line per part under FIRST VIEWER (- part 2 (8.00-21.50s) frames/...: no -- why). Every "no" is also a Should-fix with a concrete fix (a one-line spoken bridge from the part before, the steps as a persistent roadmap with the current one lit, a cut); a Blocker when the opening never says what the video is about or why it matters. A video that asks the viewer questions must say why before the first one. Then read the context/ files.
  3. Type detail pass. Open every frames/text-*.png crop (the largest text lines, cut from full-size frames) and every title or text frame in frames/ at full size; never judge type from the contact sheets, which shrink these flaws away. For each line check: words, math and numbers share one baseline (a formula, a superscript or a number must not sit lower or higher than the words beside it); they look one size (matching x-heights) and one weight (thin math beside bold words reads broken); kerning and word spacing are even; no widow (one word alone on a last line) or orphaned punctuation. Measure offsets in pixels on the crop and quote them in the finding.
  4. Hearing pass. You cannot listen, so audio.txt measures what an ear would catch and hearing.png plots it: each voice line's level over the music and its words per minute, pauses and near silence, the level on both sides of every cut, each effect's timing and level, how the music ends, the peaks, how loud it is heard on a phone (the mix above 300 Hz: on a phone speaker: -27.6 LUFS, 13.6 LU under the mix is a video most viewers hear far too quiet), and the read-back (each name or number the voice-over was heard saying differently from the script). Go through the checks CRITIC.md lists and write one line per check under HEARING (- voice over music: problem -- line 4 sits 5 dB over the bed). Judge only what the numbers and transcript.txt support and quote them; every problem is also a finding with its time and hearing.png. What numbers cannot show (how the voice sounds, a mispronounced word the read-back does not list, harsh s sounds, whether the music fits) goes under DECLINED TO JUDGE, never a guess.
  5. Answer the eight questions in CRITIC.md (hook, clarity, readability, craft, distinctness, poster, honesty, story logic) from what you see, not from what the brief says was intended. A launch film's brief adds a checklist (scene count, continuous scene changes, holds, type system, motion, music on the phrases, end card): judge every line of it too. A showreel's brief has the showreel rubric instead (energy, density, craft, surprise, ending): judge every line; there a tame reel is a finding, flash words are not, and a shot that is there because it looks good is doing its job. For story logic, go shot by shot: say what a stranger would think each shot is and what job it does; flag any shot that is there only because it looks good (not in a showreel).
  6. Write FINDINGS.md in the same round folder, in the format CRITIC.md gives:
    • the FIRST VIEWER lines, one per part;
    • the HEARING lines, one per check of the hearing pass;
    • findings sorted Blocker, Should-fix, Polish, one bullet each, each with a timestamp and a frame path from the pack (a sound finding: a timestamp and hearing.png), what is wrong, and a concrete fix. A finding without a location does not count. The maker must fix or waive every blocker and should-fix before delivery, so rate honestly both ways;
    • what works (so it survives the fixes);
    • what you declined to judge and why (for example how the voice sounds: the hearing pass covers what the numbers show, nothing more);
    • a verdict: ship, ship after fixes or not ready;
    • the absolute verdict, WOULD I POST THIS: yes | no -- one reason: would you post this under your own name, watched once at full size by a stranger? The bar is "not cheap, broken or wrong", not "flawless": with only should-fix and polish findings it is usually a yes. Say no when a blocker stands or the whole video reads as cheap at a glance (captions in stepped boxes, a player's controls in the footage, a small or soft recording in big borders); one flawed shot is a should-fix. Judge the video alone, never "better than the last version". In quality mode a "no" holds delivery like a blocker;
    • from round 2, a PREVIOUS line per earlier blocker or should-fix, by the id CRITIC.md lists: fixed r1-S2: ... or not fixed r1-S2: ..., naming what it was about. Your "not fixed" reopens a finding the maker marked fixed;
    • the best poster frame.

Done when: FINDINGS.md exists in the round folder, it has a FIRST VIEWER line per part and a HEARING line per check, and every finding has a timestamp and a frame path (or hearing.png).

If writing FINDINGS.md fails (some hosts deny Write to sub-agents), do not shorten anything: put the whole answer, exactly as the file would hold it, in your last message inside one ``` block, after the return contract (the 20-line limit does not apply to that block), with STATUS: DONE_WITH_NOTES and the note "FINDINGS.md not written (Write denied): the answer is below". The director saves it with showtime review-findings.

Pairwise rounds#

A brief in round-N/order-1/ or order-2/ compares two versions, X and Y, shown in the order it names; another critic judges the other order. Look at every image of both (../X/, ../Y/, the side-by-side compare-*.jpg, both qa.txt, transcript.txt, story.txt, audio.txt and hearing.png), do the first-viewer pass and the hearing pass for each (lines tagged [X] or [Y]), answer the questions for each, then give PREFERENCE: X | Y | tie (the one you would ship), then WOULD I POST X: and WOULD I POST Y: (yes or no, one reason each, each judged alone: preferring one does not make it postable). Every finding names its video ([X] or [Y]) besides its time and frame. Which version is newer is hidden on purpose: never look for it (the .pairwise-keys/ folder, other order-* folders, file dates). Write FINDINGS.md next to your brief. No scores in any round: do not rate a video on a number scale.

Severity#

As CRITIC.md defines it. Rate what you see; nobody may tell you what to ignore. A self-imposed "minor at most" is how real problems ship.

Never#

  • Editing the project, re-rendering, running the showtime workflow, or fixing anything yourself.
  • Writing anywhere but FINDINGS.md next to the CRITIC.md you were given.
  • Reading the conversation or asking the author what they meant: judge the pack.

View or edit skills/showtime/references/crew/critic.md on GitHub