# The continuity gate

> In short: Every generated shot is measured against the locked references it committed to, and the measurement is stored with the shot, so "this shot held" is evidence rather than an opinion. Today the gate measures and reports; it does not yet reject a drifting shot.

Applies to: v0.10 · Last reviewed: 2026-10-25 · Canonical: https://docs.zymcos.com/product/continuity-gate/

Generating fifteen versions of a scene and picking the one where the character still looks right
is how most AI production works today. It is cherry-picking by eye, and it does not scale to a
twenty-minute film.

The continuity gate is the alternative: every generated shot is **measured** against the
references it was supposed to be made from, and the measurement is stored with the shot. "This
shot held" becomes evidence rather than an opinion.

> **Read the honest limits before you rely on this.** Today the gate measures and reports; it
> does not yet reject a shot for drifting. [What the gate does not do
> yet](#what-the-gate-does-not-do-yet) explains why, and what is enforced in the meantime.

## Shots, and what they commit to

Your project has a **board**: shots in order, optionally attached to a script. Before a shot can
be generated, it **commits** to the entities it contains — and it commits to a specific locked
version of each, not to the living bible entry.

That distinction is what makes retakes trustworthy. A shot redone six weeks later is
regenerated from exactly the references it used the first time, even if the character's entry
has since gained new photographs. And when a shot does drift, you can point at the specific
locked reference it drifted from.

Two commitments are refused outright:

- **A version with no rights basis, or one whose rights have been withdrawn.** This refusal has
  no threshold and no override. It happens before anything is reserved or generated, so an
  undocumented likeness never conditions work you pay for. See
  [The production bible](https://docs.zymcos.com/product/production-bible/).
- **Two versions of the same entity on one shot.** A shot holding both revision 1 and revision 2
  of the same character would be measured against two contradictory sets of references, with no
  way to say which was meant. This holds even when two people commit at the same moment: one
  commitment lands, the other is refused, and the refusal shows what the shot now commits to.

## Moving a take, and deleting a shot

A take's scores belong to the shot it was measured in. When you move a take to another shot, its
scores stay with the shot they were measured in, and the take starts fresh in the new one. The
gate and the shot's best-take ranking look only at the takes in a shot now.

A shot that holds takes can be deleted, and nothing is lost: its takes move to Unassigned, every
score stays with the take and the shot it was measured in (they are the record a continuity question
is settled from), and **Undo** brings the shot back, takes included, for 30 minutes. The one thing
that stops a delete is a shot with a current final; replace the final first.

## What the gate measures

For every entity a shot committed to, frames from the generated result are compared against that
entity's locked references, on separate axes:

| Axis | What it compares |
| --- | --- |
| Identity | A character's face against the face built from their reference pack. |
| Element | A figure or prop against its references. |
| Set | An environment against its references. |

There is deliberately **no single blended score**. One number tells you a shot is bad without
telling you what is wrong, and a retake needs to know whether to fix the character, the prop or
the set. (A palette axis is named in the scorecard but has nothing to measure against yet, so it
is always reported as unmeasured rather than guessed at.)

Every score is recorded together with the version of the scoring rules that produced it, so
changing those rules later cannot silently reinterpret numbers taken under the old ones.

## Two stages, and why the still comes first

A still image costs roughly a tenth of what a video clip costs. So the gate runs twice:

1. **The keyframe stage** scores a still for the shot. It runs *before* any video is paid for.
2. **The video stage** samples several frames evenly across the finished clip and scores those
   the same way, taking the middle value rather than the worst — a single motion-blurred frame
   is noise, while real drift is sustained.

The point of the order is money. A keyframe that has lost the character should never become a
video bill. The video stage runs after the spend, so it informs whether you accept the take and
what you retake, rather than authorising anything.

## What the gate enforces today

**A clip for a shot with a cast waits until the shot's newest picture has been checked.** That is a
real rule, enforced before anything is reserved or sent to a model, and it is what makes the
two-stage design something you cannot walk around. The newest picture speaks for the shot, so
answering a bad take with a fresh one works the way you would expect.

**You do not run the check.** It runs by itself, within a couple of minutes of a picture landing.
If you ask for a clip before it has, the card says so before you click ("This would be refused right
now…"), in plain words, with what to do first, and the note disappears on its own when the check
finishes. If a click is refused anyway, the answer says why and that nothing was charged.

What the card can say, and what it means:

- **The shot has no picture yet.** Make a picture for the shot first; there is nothing to check.
- **The picture has not been checked yet.** Give it a minute and ask again.
- **A character on the shot is still being prepared.** The Bible is working out the character's look
  from their reference pictures; ask again once it is ready.
- **The picture did not pass.** Make a new one. This is the rule for when thresholds exist; it cannot
  happen today (see below).

**A shot with no cast is not held back at all.** A scene with no character has nothing to measure its
picture against, so its clips go ahead. And if a shot has a cast but **nothing about the character could
be measured** (no usable reference picture, or no face showing), the check still counts as having run:
the clip goes ahead, and the answer is a plain note that neither the character nor the clip could be
checked, never a block.

The rights refusal described above is also fully enforced. It is a legal control rather than a
quality one, and it fails closed.

## What the gate does not do yet

**The gate does not currently refuse a shot on continuity grounds.** It measures every axis it
can, records the scores, and leaves the verdict blank.

The reason is calibration. Turning a similarity number into "passed" or "failed" requires a
threshold, and a threshold invented rather than measured is worse than none at all: set it too
high and it rejects perfectly good work until people learn to ignore it; set it too low and it
waves drift through while looking like a check. Thresholds here come from a curated set of
shots — pairs that genuinely held and pairs deliberately drifted — that the gate has to
separate reliably before any number is trusted. Until that measurement exists, no axis has a
threshold and no score carries a ruling.

So, plainly: **do not rely on the gate to catch drift today.** Review your shots. What you get
in the meantime is the measurement history, the two-stage structure enforced, and scores that
become verdicts with no change to how you work once the thresholds land.

## Reading a take's marks

A take that is in a shot with a cast shows its **marks**: under the picture in the Viewport, and under
the take on the stage in the History tab. For each entry the shot is cast with, one line gives the
check where this take did worst, with a **Show the other checks** under it.

- **A mark is a number and a place.** "Looks like 0.81 · Strongest of 3 takes" says how similar the
  take is to the locked pictures, and that it ranks first of the three takes measured for this shot.
  The checks are **Looks like** (the character's face), **Matches** (a prop), **Same place** and **Same
  colours**.
- **The place is an ordering, not a verdict.** *Strongest*, *Middle* and *Weakest* split the shot's
  measured takes into thirds; *Only take measured* and *Tied* are said as they are. It tells you which
  take held an entry best. It does not tell you that any take is good.
- **Nothing says pass or fail** while no check has a measured threshold, which is today. A line under
  the marks says so. The identity check carries a starting point borrowed from face verification; it
  is shown as a starting point, never as a ruling. The day a row carries a verdict it appears in that
  cell, against its own threshold, and the line says that instead.
- **A cell with no mark says why**, in words: nothing to compare it against (a place has no colours
  to match), its identity checks are not ready (still preparing, could not be measured, or could not
  be run just now), it was not in the shot's cast when the take was made, or the check has not run on
  the take yet. It is never blank, and never a low mark.
- **A shot with no cast** says "Cast this shot to see how its takes match."
- **Which version** is stated for each entry ("As locked in version 2"). If you cast a newer version
  after a take was made, the take keeps the version it was measured against.

Under a take's marks, one line says what they are (*Similarity to the locked pictures, not a pass
or a fail*); **Why?** opens the full explanation.

**The scorecard** puts every take of a shot side by side: takes down the side, cast entries across,
each cell a mark. Open it from a take's marks (**Open the scorecard**) or from **Takes and marks** in
the shot editor. A take measured under an earlier version of the scoring rubric is listed in its own
section and never ranked with the rest, so changing the rubric never rewrites an old judgement.

You can ask CODA about the take on the stage ("how does this one hold up?"). It reads the marks and
quotes the numbers; it will not say a take passes or fails continuity, because nothing does yet.

## When a score is missing

A missing score is never a pass. Absence of a measurement is not evidence of continuity, and the
gate keeps two different absences apart because they call for different things:

- **Not yet calibrated.** The axis was measured; there is no threshold to judge it by. This is
  our gap, and it is the one above.
- **Nothing to measure.** The shot committed to a character and no frame showed a face, or the
  character's reference pack never produced one. This is a fact about your material, and it is
  the one worth acting on. A character deliberately framed from behind is fine and will read
  this way. A character who should have been clearly visible is telling you something — either
  about the shot or about the references.

Because the second case is legitimate often enough, it is reported rather than refused. A
quality check whose first act is to reject correct work teaches people to route around it.

## Where this lives today

Zymcos is in active development. You cast a shot and read its marks and scorecard in the app; the
checks run by themselves and Zynth operates them for your project. A mark is evidence to look at, not
a gate: nothing in this page as enforced depends on the screens that show it.
