# What a format conform does not fix (2026-09-11)

Evidence for `182.C.22.05`: reconcile provider clips with CFR/VFR, time base,
colour, resolution, grain, motion cadence, and audio ambience differences.

Those seven do not behave the same way, and the split is the finding.

## Method

Three 16-frame sequences built from the committed frame: a clean pan, the same
pan with Gaussian grain, and the same pan in a 3:2 (24-in-30) cadence. Each
measured for flat-region noise and for the identical-frame-pair pattern. The
first two were then conformed through `h264 / yuv420p / 30 fps` — what
`182.C.16`'s `planConform` does — and measured again.

## Grain survives the conform

| Clip      | flat-region noise | after conform |
| --------- | ----------------- | ------------- |
| clean     | 1.021             | 0.854         |
| grainy    | 3.896             | 3.368         |
| **ratio** | **3.82×**         | **3.94×**     |

The conform normalised the container and the codec and removed **none** of the
difference.

### And the obvious way to measure grain does not work

| Clip   | whole-frame residual | flat-region noise |
| ------ | -------------------- | ----------------- |
| clean  | 13.078               | 1.021             |
| grainy | 13.597               | 3.896             |
| ratio  | **1.04×**            | **3.82×**         |

A whole-frame high-frequency residual is dominated by image detail — hair,
fabric — so it reads about 13 whether or not grain was added. Measured in the
flattest 10% of 16×16 patches, where there is no detail to dominate, the same
pair separates cleanly.

## Motion cadence survives it too

| Clip              | identical pairs | longest run |
| ----------------- | --------------- | ----------- |
| clean (native 30) | 0               | 0           |
| 3:2 (24-in-30)    | **9**           | **2**       |

That is the judder signature `182.C.19.06` measured. The duplicated frames _are_
the clip, so conforming to 30 fps re-encodes them rather than removing them.

## The conclusion

**Format properties conform away and perceptual properties do not.** Frame rate,
time base, pixel format, colour tagging and resolution are `182.C.16` conform
targets; grain, motion cadence and audio ambience are measured, compared between
clips, and reported as a reconciliation gap a conform will not close.

`assertPerceptuallyReconciled` is not a prohibition on cutting between providers
— a caller can accept the gap or apply a matching pass. It is a refusal to let a
_conform_ stand in for that decision, which is what "reconcile" quietly becomes
when the format gaps are the only ones anybody looked at.

An unmeasured perceptual property is reported separately from a matching one,
and refused first: a property nobody measured is not one that matched.

## Per-item status this evidence supports

| Item          | Was           | Is                                                                                 |
| ------------- | ------------- | ---------------------------------------------------------------------------------- |
| `182.C.22.05` | `not-started` | done; format gaps named conformable, perceptual gaps measured and named unclosable |
