# Jina — why it has zero stability comparison data (not just "thin")

**Run 1** (live tree, `jina/input-*/outputs/raw/search/`): 16/52 succeeded (`http_status: 200`,
Q01–Q10 + Q42–Q47), 36/52 failed (`http_status: 402`, Q11–Q41 + Q48–Q52). Confirmed two independent
ways: (a) reading every raw JSON file's `http_status` field directly, (b) cross-checking against
`_analysis/all-tools-scored.csv` filtered to `tool==jina`, which shows exactly 16×200 / 36×402 landing
on the same query IDs.

**Run 2** (`_archive_run_2026-08-17-stability/jina/input-*/outputs/raw/search/`, run_id `2026-08-17-r3`):
**0/52 succeeded.** All 52 raw files — including Q01–Q10 and Q42–Q47, which had succeeded in Run 1 —
now return `http_status: 402`. Confirmed both in the raw JSON and in
`_archive_run_2026-08-17-stability/_analysis/all-tools-scored.csv` (`tool==jina` → 52×402, zero 200s).

**Conclusion:** Jina's free-tier quota had fully closed by the time Run 2 executed, wiping out even
the 16 queries that worked in Run 1. The intersection of "completed in both runs" is 16 ∩ 0 = **0
queries** — there is no possible stability comparison to make, not merely an incomplete one.

**Disposition:** "No stability comparison for Jina" is correct and now evidence-backed, not an
unsupported claim. This is a permanent gap for this round (resolved only by re-running Jina on a paid
tier), and should be stated plainly rather than left as a blank cell: *"Jina completed 16/52 queries in
Run 1 and 0/52 in Run 2 (quota fully exhausted by Run 2); the two runs share zero completed queries, so
no stability comparison exists for Jina in this dataset."*