# Lessons from client notes

Each note below changed a rule in `SKILL.md`; the section it led to is in brackets. The films are the teaser (a 60 s teaser for an executive education program), the opener (a company's internal event opener, six versions), the reel (this project's own 60 s homepage film) and the lesson (a film for a skill that teaches a way of judging work). The notes are paraphrased and shortened. Read them when a new note arrives. Diagnose with measurement, fix the cause, and report before and after numbers.

## From the teaser

- **Flat on "you".** The take ran into the word and dropped it 6 dB. Re-directed and measured, it went from 0.08 s, +0.8 st, -6.0 dB to 0.38 s, +5.3 st, +4.2 dB. [5.3]
- **It repeats the people at the end.** The closing montage was cut, and its five seconds became the payoff the opening had set up. [3.2, 8.1]

## From the opener

- **The voice should sound like the audience.** Recast in the audience's own accent. [3.4]
- **The experts are not highlighted enough.** A 2.3 s name ripple became a spotlight tour, two beats a person. [3.7]
- **The joke does not need to be said out loud.** The joke now plays in type, in silence. [2]
- **The words flash so quickly they are missed.** One moment per screen at 78 to 84 px, spoken by the narrator and held while she speaks. [3.2]
- **Use icons to show people coming together.** The template's people icons gather into one crowd, then one ring; only the named people kept their photos. [3.7]
- **Open on leadership's frame, then zoom into the team.** The leader's frame first, the phrases held without quote marks, then a dive into the team's dot. [2]
- **Turn the campaign image into a fast building animation.** The graphic was taken apart and built on the beat. It enters through the headline's full stop and leaves as the ring's dot. [3.8]
- **Some pictures are low resolution, the zooms are not smooth, and a word on screen differs from the script.** These were all notes on a run that had no eyes. The photo rule, the layout audit and the motion audit now catch them before the client does. [3.7, 6.4, 8.3]
- **Less slide template, more of the bright colors.** Each moment became a full field of its template color, opening from where its number sits, and the template's corner furniture went. [3.9]
- **Weave the opening back in, and bring back the leads.** The score grew by whole bars, every line was re-placed by rule, and the leads got a tour of two beats each. [3.7, 5.2, 8.5]
- **Weave in the team's slogan.** Every spoken take breathed out its last word, so the slogan plays in type, on the music. [3.1]

## From the reel

- **A different male voice would be better.** Three voices were cast on the turn line by measurement, auditioned in context, and the pick voiced the script. The next note: none of them is dramatic. An alto the client had approved on an earlier film read the whole script in one pass, and the note after that kept the first narrator. Measurement rejects bad takes; the client's ear chooses. Check the speech provider's credit balance before a full re-record: the last recast got two of its four takes. [3.4]

## From the lesson

The first cut passed every technical gate: every word recognized, the layout lint clean, loudness on target. It was rejected on first viewing, and the notes were about the story, not the craft.

- **The script doesn't say what this is for.** It showed edits and evidence but never the problem the product solves or its core thesis. Now the brief starts with the purpose, in plain words, taken from the repo or site, and the creative gate checks for it. [2, 4]
- **You should have reviewed it with me first.** An hour of generation and rendering went into a script nobody had approved. Now a storyboard and script go to a person within minutes of the brief, and nothing is generated until they approve it. [4.1]
- **It isn't accessible to non-technical viewers, and it isn't obvious what we are talking about.** The first storyboard used insider words (evaluations, rubric, scenarios) and an outage report as its example. The second used one everyday object anyone can judge in a second: a yard-sale sign seen from a passing car. [2]
- **Define the ideas and teach people to use them; the product comes second.** The film became a lesson: two signs that fail in different ways, seen before either failure is named; each term defined on the sign that lacks it; a map and a question with a real silent pause; the two fixes; a three-second check; the same questions on an email, a slide, a room and an AI answer; three habits; and the product last. [2]
- **The animation carried more than the meaning.** Motion now has to explain: one idea per beat, edits drive the moves, no kinetic type for its own sake. The gates measure craft, not meaning. [3.9, 8.4]
- **The sun is in the wrong place.** On the storyboard, a child's sun doodle sat on the sign's date line. It moved to the top corner and stayed there in every frame. Screenshot the sketches and keep a recurring motif in one place. [4.1]

## From the agency teaser

A 60 s teaser on the author's own research into human agency. It passed the creative gate, the cold read found both ahas at their declared times, and every claim traced to the papers.

- **Just jumping into this video, I'd be confused.** Four simulated first-time viewers (a student, a second-language reader with the sound off, a busy nurse, a retired teacher who does not use feed apps) rated their understanding 2, 2, 3 and 2 out of 5. Most were lost at the same places: a research reversal said in abstract words, a term ("the strongest lever") never shown, the jump from a 1967 study to a phone feed never explained, and key lines that existed only in the voice. Every earlier reviewer had read the brief and judged understanding in passing. Now a first-time panel is a gate at the storyboard and again on the render, every idea the story needs is shown as well as said, and real people outside the field say the storyboard back before generation. [3.2, 4, 4.1, 8.2]
- **Super robotic, not emotional, and a little depressing. Not the high-energy sizzle I expected.** The panel measured understanding only, and each round of clarity fixes slowed the film into a lecture. The director had also chosen a quiet tone ("a film about noticing") without naming it to the user, over a subject (shocks, helplessness, a dark lamp) that pulls down. Now a sizzle defaults to rousing, the brief names the feeling the viewer leaves with, the storyboard states the energy plainly, the panel reports how the film made each viewer feel and whether they would watch it again, and clarity fixes must not cost energy. [defaults, 2, 4, 4.1]
- **Here are the two major ahas from this research.** After four storyboards built on the director's reading of the papers, the author named the two realizations they wanted, and two essays of their own that carry them. The director had never asked. Now, when the person wrote the source, the first question is what two things they most want a viewer to realize. [2]
- **Calibrating the panel.** Over four rounds of rework the ratings rose to 3, 3, 4 and 3.5, and every viewer retold the proposition, explained the motif and read the hero line correctly. Yet all four still listed the motif's opening as "lost" every round, because it is shown before it is explained. Asked to list where they got lost, simulated viewers always find something. A moment now counts as lost only when the confusion lasts into the retell and the answers. [4]

## What the audit found after the eyes had looked

On the opener's v5, after five rounds of looking, the layout audit still found a person icon crossing the title, two phrases overlapping at a handover, and a four-word hint card on screen for 0.9 s. On v6, teaching the audit about masks, clips and covers took it from 49 false issues to none, so the real ones stood out. [6.4]