worker/data is stale: index missing 116 repos, total_repos disagrees with itself, phantom verisim-data entry

[hyperpolymath]

The registry’s served data has drifted from the seed. No uuid is wrong — I checked, and every uuid in worker/data/repos.json matches its derivation exactly. The problem is coverage, counts, and a phantom entry.

Found while creating metadatastician/cadastra and cross-checking identities against the registry.

1. The index is missing 116 repos (~36% of the estate)

index.json  "generated"   : 2026-06-20T18:09:58.001Z     <- ~4 weeks stale
index.json  total_repos   : 203
index.json  by_name count : 204                          <- disagrees with its own total_repos
repos.json  entries       : 320
seed repos.a2ml entries   : 319

116 repos present in repos.json are absent from index.json’s by_name, e.g. a2ml-ecosystem, a2ml-estate-normalizer, a2ml_ex, a2ml_gleam, a2ml-showcase, …

worker/src/verisim.js:211 fetches repos.json at runtime, so this is what the worker actually serves.

Note worker/test/data.test.js opens with:

// so snapshot drift (e.g. a repo added to repos.json but not re-indexed) fails CI.

which is precisely this condition. Either that test is not running in CI, or it does not assert the count. Worth checking — the guard appears to be there but not catching it.

2. Phantom entry: verisim-data

worker/data/repos.json (and 5 places in index.json) carry:

{
  "name": "verisim-data",
  "uuid": "22772417-fc00-5263-a78a-7567aee04984",
  "github": "hyperpolymath/verisim-data"
}

• hyperpolymath/verisim-data → 404. metadatastician/verisim-data → 404. The repo does not exist.
• It is not in verisim/seed/repos.a2ml.
• The real repo is verisimdb-data, which exists and is in the seed (line 374) and already in the JSON.

This looks like unfinished business from #46 ("dedupe verisimdb-data"): the seed was deduped, but worker/data/ was never regenerated, so the phantom survives in the served data. Its uuid derives from a URL that does not resolve.

3. export-json.sh is not what produced the committed data

This is the reason I did not just regenerate and open a PR — it is not clear which side is canonical:

	committed worker/data/repos.json	sync/export-json.sh output
format	pretty-printed, 3646 lines	compact, 321 lines
unicode	escaped (—)	literal (—)
entries	320 (incl. phantom)	319

awk printf cannot produce — escaping, so the committed file was not generated by export-json.sh. Either the script is stale/superseded, or the data is maintained by another path and the script is vestigial. Running it today would fix the phantom and the 116 missing repos, but reformat ~3,300 lines and assume it is canonical.

That decision is the owner’s, which is why this is an issue and not a PR.

4. Also stale (minor)

verisim/seed/repos.a2ml [meta] total-repos = 200, actual entry count 319.

Suggested order

1. Decide which generator is canonical (or that the data is hand-maintained and export-json.sh should be removed).
2. Fix that generator to emit the canonical format, so regenerating is a no-op except for real changes.
3. Regenerate → drops the phantom, restores the 116 missing repos, fixes total_repos.
4. Make worker/test/data.test.js actually fail on this drift, since it is written as though it should.

Related

• fix(sync): derive uuid from the repo's actual owner, not a hardcoded one #48 — deploy-clade-a2ml.sh / export-json.sh hardcoded the owner segment to hyperpolymath, giving a wrong uuid to any repo hosted elsewhere. Separate bug, same files. metadatastician/gossamer was a live victim (fix(clade): derive gossamer's uuid from its real owner — the file contradicted itself metadatastician/gossamer#113).

🤖 Generated with Claude Code