# Production Jev relation qualification

On 2026-09-30, a one-off server-side client used the deployed `jev-1.13.0` request contract at `physics.er33t.com` for a fixed 16-case cohort. The public app routes were deliberately left unchanged. The client sent only code-derived state and a two-choice question (`same` or `opposite`); it did not send the deterministic expected relation.

The first eight cases paired Rule 110 at local index `i` with Rule 137 at complementary index `7 - i`. The expected relation was `opposite`. The second eight paired Rule 42 with the equivalent Boolean expression `R AND NOT(L AND C)` under `index = 4L + 2C + R`. The expected relation was `same`. Candidate order was alternated so each relation appeared first eight times.

All 16 provider responses passed the one-off client's response-shape checks, matching the deployed client conventions for model id, choice, probability keys and normalization, confidence range, and integer input-token count. Jev selected the deterministic relation in all 16 cases. Provider-fetch latency ranged from 107.75 to 301.67 ms; mean confidence was 0.9931. The complete sequential run took 2.640 seconds and used 8,648 validated input tokens.

The result qualifies typed selection over this resolved finite cohort. It does not replace the deterministic truth-table proof or the global induction, and it makes no claim about universality, biological mechanism, or general model accuracy. One earlier unresolved one-cent neural-plan reservation was observed in the durable ledger and was not retried; the qualification added 16 one-cent reservations, all completed.

Operational receipts are retained privately. The public per-case hashes and validated results are in the companion [jev-rule-check.json](jev-rule-check.json); raw provider payloads, credentials, and private operational details are omitted.