# Machine Learning catalog data (/dev only)

`ml_entries.json` drives https://datasystemsglobal.com/dev/ml.html, so ML models, libraries, research notes and downloaded repos never get missed again.

Shape (a bare array of entries is also accepted):

```json
{"as_of": "YYYY-MM-DD",
 "status_note": "Search in progress: more entries coming from Video_Scrap",
 "entries": [ {...}, {...} ]}
```

- `status_note`: shown as a banner at the top of the page. Set it to `""` to hide the banner.
- Cards render in file order.

To add an entry, append one object to `entries` (comma between objects, keep it valid JSON):

```json
{"id": "short-slug", "name": "...", "type": "model|library|repo|research note",
 "source": "URL / bot / date found (bare domains like github.com/x/y become https links)",
 "box_path": "/workspace/...", "what_it_does": "...",
 "nq_useful": "yes|maybe|no|untested", "verdict_note": "...", "found_date": "YYYY-MM-DD",
 "link": "/dev/some-page.html#section (optional: shows a 'See on ML Models page' link)"}
```

Each card is anchored by `id` (`/dev/ml.html#<id>`). The type and NQ-useful filter buttons and their counts are built from the data automatically. Don't invent data: only add entries that have actually been found or checked. No write API and nothing to rebuild; nginx serves /dev JSON with `Cache-Control: no-cache`.