SURVIVES 2026-07-10 ยท all 16 regression tests re-run on current code + fresh data ยท 0 failures in ~44 min ยท clean 1/3
"All 16 closed gaps hold SIMULTANEOUSLY on the current, further-hardened code (post iters 15/16/17: driving-cell attribution, determinism pin, slate dedup) and current data." Non-vacuous twice over: iter 14's sweep found 2 temporal re-openings and the knife-edge finding; the probe files have changed three times since.
| Test | Status | Time |
|---|---|---|
| A5 substrate ยท A7 nulls ยท A1 metamorphic ยท A6 thin-cell ยท A2 blind-spot ยท A9 epochs ยท A10 h_norm ยท A12 kernel-path ยท A13 diagnosability ยท A15 resolution ยท A17 sibling | PASS ร11 | 3โ169s each |
| A3 block-length | PASS | 169s |
| A4 stitching + MC gating (previously re-opened by drift; sound oracle) | PASS | 379s |
| A8 pooled order-invariance | PASS | 867s |
| A11 label-shuffle null (previously re-opened by drift; sound oracle) | PASS | 545s |
| A16 determinism | PASS | 271s |
16/16 โ including the two previously drift-fragile closures under the iter-14 certification-stability oracles, on data hours fresher than their last proof: the temporal re-opening class is controlled.
| File | What |
|---|---|
| run_suite_v2.sh | The 16-test compound runner โ now the standing compound regression (supersedes iter 14's runner). |
| suite_v2_results.json / suite_v2_run.log | Per-test status + timing; full per-attack logs. |
The strongest compound evidence yet: every defect class found across 17 iterations โ estimator, substrate, aggregation, transfer axes, library rot, production path, auditability, input hygiene โ remains closed at once on live data. Clean rounds: 1 of 3. (4 batched firings collapsed into this single iteration per the one-iteration-per-firing contract.)
iter 18 ยท expanded-suite sweep ยท verdict SURVIVES ยท one atomic commit:
iter 18: expanded-suite sweep โ SURVIVES ยท append-only.