Skip to content
C.W.K.
Stream
Lesson 07 of 07 · published

Nothing Prints Before Its Cue

~14 min · timing, motion, visual-design, evolution

Level 0Cold Workshop
0 XP0/43 lessons0/12 achievements
0/100 XP to next level100 XP to go0% complete

The Other Half of the Redundancy Rule

The first lesson of this track said never render what the narration already said. A few episodes later the workshop found that rule's mirror image, and it is the more dangerous of the two. A screen that repeats the voice wastes the second channel. A screen that runs ahead of the voice does something worse: it prints the conclusion while the voice is still building the reason, and the viewer who reads ahead has nothing left to discover.

It surfaced first in the signature beat. An early cut entered every correction scene with the node already struck and the replacement already installed, and one replacement sat on screen eight cues before the narration reached it. The screen was reporting the correction instead of performing it. A reviewer reading the script before any synthesis caught it, and the fix became format law: a correction is staged across three scenes — the premise, the strike, the replacement — each appearing on the cue that speaks it.

Then it surfaced everywhere else. One later episode did it three ways at once: a graph node that arrived thirty-three cues before its name was spoken, a card that printed its closing line for seven cues before the voice got there, and a chart heading that gave away a reveal. The first sweep looking for this compared each card only against its own scene's cues, and found none of them. The check that stuck compares every string on every screen against the cues of every later scene.

The Page Stays Still; Its Elements Move

Staging across scenes solved corrections, but the picture was still coarse: a graph appeared complete, a card printed all its lines at once, a data plot was born finished. Between a camera move over a whole frame and a still board, there was nothing. So on 2026-09-05 the workshop added the missing piece, consolidated from a blind run in which five different brains attacked the same brief: element-level motion on the voice's clock.

A scene may now name which of its elements land on which of its own cues — a card's lines, a graph's nodes and a node's struck state, a plot's points. Because a struck state can land on its own cue, one graph scene can now perform a whole correction — premise, strike, and replacement, each on the cue that speaks it, as the sample below does — which is exactly what the three-scene staging was protecting. The cue is the clock, never seconds, so a re-recorded take that shifts the timing shifts the landings with it. An element that is not named is on from the first frame. And the design protects everything already built on the still picture: the last settled frame is exactly the plate, so the thumbnail source, the read-back, and the lookahead check all see what they always saw. A hidden element keeps its slot, so nothing moves when it lands.

Reading Rules, and the Two Edges

The blind run and its panel of judges also produced rules for how motion behaves around words, and they are worth carrying anywhere text shares a screen with time. Every landing leaves a hold. Text never moves while it is being read — a background may travel, a sentence may not. A camera moves only toward the next thing to be named, and it settles before that cue is spoken. And a card whose lines all land on consecutive cues is not motion design at all; it is a subtitle, which the first lesson of this track already refused.

Two failures mark the edges. One episode passed every check with twenty-two scenes whose first cue played over a bare background, because every element had been keyed to land after it — and the contact sheets of last frames could not show the problem, since every last frame was complete. Now a scene must land something on its first cue. And the reference episode, which built forty of its fifty-six scenes on cue between fifteen whole generated boards, still needed seven direction repairs once real speech timing exposed long holds on completed diagrams. A plan keyed to cues is not the same thing as a film watched at normal speed, which is why direction is judged on the real cut.

Early is a form of redundancy. A screen that shows an answer before the voice reaches it has spent the channel twice — once now, and again when the narration arrives to say what the viewer already read. Key every arrival to the moment its meaning is spoken, and check each screen against what comes later, not only against what has already been said.

Code

A graph scene whose correction lands on the voice·json
{
  "key": "gB3",
  "kind": "graph",
  "nodes": [
    { "key": "b1", "label": "local inference feels slow" },
    { "key": "b2", "label": "because the model is too big", "state": "struck" },
    { "key": "b3", "label": "memory bandwidth sets the pace" }
  ],
  "motion": {
    "at": { "b2": "k112", "b2:struck": "k117", "b3": "k119" }
  }
}

// b1 is on from the first frame: the voice has already reached it.
// b2 lands plain on k112, and is struck on k117, the correcting cue.
// b3, the replacement, appears only on k119, when it is named.
// the settled last frame is the plate: three nodes, b2 struck.
The lookahead check the first sweep got wrong·python
def printed_too_early(scenes: list[Scene]) -> list[str]:
    """No screen may print a sentence that a LATER scene will speak.

    The first version compared each card only with its own scene's
    cues and found nothing. The defect lives across scenes.
    """
    problems = []
    for i, scene in enumerate(scenes):
        for text in scene.screen_strings():      # nodes, lines, labels
            for later in scenes[i + 1:]:
                for cue in later.cues:
                    if same_claim(text, cue.text):
                        problems.append(
                            f"{scene.key}: '{text}' is on screen before "
                            f"{later.key}/{cue.key} says it")
    return problems

External links

Exercise

Take a slide deck, a demo, or a tutorial video you have made and list every moment where something appears on screen. For each, write down the moment its meaning is spoken or explained. Mark every appearance that arrives first by more than a beat. Then decide, for each one, whether the early arrival gives the viewer something to hold while listening, or gives away the thing the explanation was building toward.
Hint
Headings and conclusion slides are the usual offenders: a title that states the finding, shown while the presenter is still walking through the evidence. If the heading can be read as the answer, the evidence that follows is playing to an audience that has already stopped listening.

Progress

Progress is local-only — sign in to sync across devices.
Spotted a bug or have feedback on this page?Report an Issue

Comments 0

🔔 Reply notifications (sign in)
Sign inPlease sign in to comment.

No comments yet — be the first.