# A composed site, small enough to diff by hand

This is the output of `compose-site.ts` over the repo's smallest real corpus
fixture, captured at the tip of **one-core phase 2 slice S1**
(`one-core/phase-2-s1`, branched from `c7f7b90`). It exists so the parallel
slices — S2a, S2b, S2c, S3 and S0-pause — can prove "the contract did not move"
against a committed artifact instead of each rebuilding a baseline and hoping
the two recipes matched.

`plans/FACTS.md` § "How to compose a fixture site offline" is the recipe; this
directory is one run of it plus the two things the recipe leaves to the reader:
the hand-written `site.json`, and a diff that knows what is allowed to change.

## What is here

| Path | What |
|---|---|
| `site.json` | The hand-written `sites/testsite/site.json`. The recipe's only authored input. |
| `public/` | 16 files — what a deploy serves. |
| `index/` | 7 files — the intermediate index the compose step copies from. |
| `compose-diff.sh` | `./compose-diff.sh <expected> <actual>` — a normalising diff. |

## How it was produced

The corpus is `editor/e2e/fixtures/test-transcripts/one-youtube-channel-with-data`
(one channel, one video, one VTT). Nothing here reads or writes the real
`transcripts/`.

```sh
F=$(mktemp -d)                      # the corpus root
O=$(mktemp -d); mkdir -p $O/public $O/index

cp -r editor/e2e/fixtures/test-transcripts/one-youtube-channel-with-data/channels $F/
mkdir -p $F/sites/testsite
cp plans/tools/compose-fixture-one-youtube-channel/site.json $F/sites/testsite/site.json
echo '{}' > $F/settings.json

cd common
env TRANSCRIPTS_DIR=$F SITES_DIR=$F/sites \
    EXPORT_PUBLIC_DIR=$O/public EXPORT_INDEX_DIR=$O/index \
    SETTINGS_FILE=$F/settings.json SITE_ID=testsite \
  pnpm exec tsx bin/build-index.ts
env TRANSCRIPTS_DIR=$F SITES_DIR=$F/sites \
    EXPORT_PUBLIC_DIR=$O/public EXPORT_INDEX_DIR=$O/index \
    SETTINGS_FILE=$F/settings.json SITE_ID=testsite \
  pnpm exec tsx bin/compose-site.ts
cd ..

plans/tools/compose-fixture-one-youtube-channel/compose-diff.sh \
  plans/tools/compose-fixture-one-youtube-channel/public $O/public
plans/tools/compose-fixture-one-youtube-channel/compose-diff.sh \
  plans/tools/compose-fixture-one-youtube-channel/index  $O/index
```

Both should print `IDENTICAL modulo the build clock`. Anything else is a real
change to the wire, and belongs in a commit that says so.

**`build-index.ts` writes `$F/index.mdb` and caches a schema version in it.**
Compose the fixture into a fresh `$F` each time, or the second run reports
`0 built, 1 up to date` and copies nothing.

## What the diff normalises, and why only this

- **`"generatedAt": "<iso>"`**, in both the compact and the pretty-printed
  spellings. It appears in `corpus.json`, `site.json`, the five manifests, and
  the archive zip's `channel.json`.
- **The archive zip's own bytes.** A zip's entry mtimes are the build clock, so
  two runs of the same input differ by a byte or two in the compressed stream —
  and `archives/manifest.json` records that size. Zips are therefore compared by
  their unpacked, `generatedAt`-normalised **entries**, which do diff clean, and
  the `bytes` field is normalised with them.

Nothing else is excused. In particular the URL shapes in `corpus.json` — the
absolute-with-`siteUrl` form this fixture exercises — are the thing phase 2's
`lib/archive/contract.ts` must reproduce, so a diff there is the failure this
directory exists to catch.

## The site.json is deliberately minimal but NOT trivial

It sets `siteUrl`, which is what makes every manifest pointer in `corpus.json`
absolute. A fixture without one would pass a reader that had quietly dropped the
origin.