Corpus State Report
Generated by cogentia.js v2.
Consolidated current-state of corpus caches/serving layers (no vector or body dumps).
Generated: 2026-07-23T06:50:53.116Z
Health signals
| Level | Code | Message |
|---|---|---|
| ok | index_built |
Index built (12971 chunks). |
| ok | embeddings_present |
8782 local embedding rows. |
| info | remote_skipped |
Remote Supabase probe skipped. |
Local SQLite index cache
This is the workstation / node cache (.cogentia/index/corpus.sqlite): documents, FTS chunks, edges. It is not mirrored wholesale to the Views Store.
| Field | Value |
|---|---|
| Built | yes |
| OK | yes |
| Schema | 0.1 |
| Index hash | 0f0cfa1d245d4155… |
| Updated | 2026-07-11T22:21:04.444Z |
| Mode | update |
| Documents (all / public) | 1128 / 545 |
| Chunks (all / searchable public) | 12971 / 12863 |
| Edges | 2407 |
| FTS5 | yes |
| DB file (basename) | corpus.sqlite |
| DB size | 358448 KB |
sqlite-vec acceleration
Status: ok · embedding rows in vec path: 8782
| Table | Dims | vec_rows | Status |
|---|---|---|---|
| vec_embeddings_1536 | 1536 | 8782 | ok |
Chunks by repository (local)
| Repo | Chunks | Searchable public |
|---|---|---|
| barons-Mariani | 4930 | 4930 |
| cogentia | 2737 | 2737 |
| FractaVolta | 1025 | 1025 |
| survey | 1015 | 1015 |
| marenostrum | 847 | 847 |
| operium | 650 | 650 |
| Inox | 326 | 326 |
| inseme | 303 | 303 |
| JeanHuguesRobert | 268 | 268 |
| Kudos | 197 | 197 |
| ubikia | 185 | 185 |
| serra | 162 | 162 |
| registre-mariani | 108 | 0 |
| privai | 71 | 71 |
| gouvernance | 61 | 61 |
| institut-mariani | 35 | 35 |
| marianivillage | 27 | 27 |
| acorsica.org | 12 | 12 |
| .github | 12 | 12 |
Local embeddings cache
Vectors live inside the same SQLite file (and optional sqlite-vec tables). Views Store only carries this summary.
| Field | Value |
|---|---|
| Built | yes |
| Rows | 8782 |
| Model | text-embedding-3-small |
| Dimensions | 1536 |
| Coverage of chunks | 67.7% |
| Created (min) | 2026-07-12 14:44:14 |
| Provider | Model | Dims | Count |
|---|---|---|---|
| openai | text-embedding-3-small | 1536 | 8782 |
Embeddings by repository (local)
| Repo | Embeddings |
|---|---|
| barons-Mariani | 4114 |
| cogentia | 2025 |
| FractaVolta | 868 |
| marenostrum | 781 |
| Inox | 229 |
| Kudos | 185 |
| ubikia | 157 |
| inseme | 150 |
| JeanHuguesRobert | 127 |
| privai | 59 |
| gouvernance | 49 |
| institut-mariani | 23 |
| marianivillage | 15 |
Remote Supabase / Postgres (serving layer)
Public retrieval may use public.retrieval_chunks (pgvector). Do not publish vectors to the Views Store; this section is inventory only.
Remote probe skipped (--skip-remote).
Why not dump the data?
- SQLite cache can be hundreds of MB with FTS + embeddings; regenerable via
index rebuild/embeddings import-cache. - Supabase vectors are a serving layer; the useful public artifact is freshness, coverage, and hash alignment, not the float arrays.
- Future uses (drift alerts, multi-node sync dashboards) still only need these state fields plus optional targeted samples.
Related commands
node scripts/cogentia.js index status --json
node scripts/cogentia.js embeddings status --json
node scripts/cogentia.js corpus-state export
node scripts/cogentia.js publish push corpus-state