Lossless Changelog
product

Corpus refresh — 16 new repos join the context-v rollup

The tree grew for two and a half months while the corpus stood still. Tonight the discovery walk ran again: 16 new context-v sources opted in, 3 relocated repos reconciled, 1,001 files re-collated, the Chroma collections re-embedded, and the splash rebuilt so every new repo shows on the corpus page.

Michael StatonMusic Editoraugmented with Claude Code on Claude Fable 5

Why Care?

The whole premise of this kit is that the corpus tracks the tree. Since the last collation on 2026-05-08, the tree added augment-it, corpora-builder, id-didi-sh, the entire dididecks-ai client-sites family, four new studies, and more — none of it visible to the corpus page, the Chroma collections, or any agent querying "did we already think about this." A stale corpus is worse than no corpus: it answers confidently from a world that no longer exists. This refresh closes the 2.5-month gap and reconciles three repos that physically moved in the meantime.

What's New?

  • Discovery re-run — assemble-context-v-sources.py walked the tree and surfaced 20 new context-v/ directories. sources.md now holds 53 curated entries.

  • 16 real sources opted in: augment-it, corpora-builder, dididecks-ai + its 5 client-sites (calmstorm, chroma, humain-vc, lossless, reach-edu-hub), id-didi-sh, 4 studies (agent-harnesses, conversational-ui-and-native-shells, memory-layers-for-agents, vector-databases), astro-knots/sites/arthouse-site, self-host-stack, and site.

  • 4 false positives excluded with notes: three Astro page routes (ai-labs/splash, augment-it/splash, id-didi-sh/splash — they render context-v, they aren't sources) and ai-labs/splash/src/rollup/context-v (a build-time copy that would double-ingest everything).

  • 3 stale entries removed, all explained by relocations: investment-memo-orchestrator → memopop-ai; calmstorm-decks and reach-edu-hub moved from astro-knots/sites/ into dididecks-ai/client-sites/.

  • Full --clean re-collation — corpus/ rebuilt from scratch: 1,001 files across 40 repo slugs, stale slugs gone.

  • Manifest rebuilt — corpus-manifest.md buckets: 107 worked-on, 540 idea-started, 354 stub.

  • Chroma re-ingested — context-vigilance-corpus upserted 8,649 chunks from the 41 included sources; lossless-changelog picked up 153 new/updated entries across 35 changelog dirs.

Verification

$ .venv/bin/python scripts/collate.py --clean
collated 1001 file(s) into corpus  (41 of 53 sources included)
$ .venv/bin/python scripts/build-corpus-manifest.py
total files: 1001  (worked-on 107 / idea-started 540 / stub 354)
$ cd splash && pnpm build
904 page(s) built in 35.82s — Pagefind indexed 904 pages

The built corpus index page carries entries from every newly opted-in repo (id-didi-sh, corpora-builder, agent-harnesses, chroma-decks, self-host-stack, …) and the homepage stats read the fresh 1,001-file manifest. A smoke query against Chroma for "didi.sh identity service" returns the new Id-Didi-Sh-Identity-Service.md spec as the top hit.

What's next

The corpus lags the tree whenever new repos land — the discovery walk (assemble-context-v-sources.py → curate sources.md → collate.py --clean → build-corpus-manifest.py → ingest-all.sh) is cheap enough to run on a cadence rather than on memory. The ~800 pre-refresh chunks still in the Chroma collection are harmless but could be purged with a one-time ingest-all.sh --reset if stale paths ever surface in query results.

  • [[2026-05-08_01]] — the original collation that this refresh supersedes.

  • context-v/explorations/ChromaDB-as-Context-Improvement-Across-Everything-Everyone.md — the exploration that produced the ingest pipeline.