Publication summary: GLM-5.3-Flash ormandj v0.4.2¶
Canonical facts¶
- Decision:
retain-baseline/no-promotion; campaign concluded without production promotion. - Matched cells: candidate improved median TTFT/E2E/decode by -5.19%/-4.91%/+3.93% at 4K n12 and -1.31%/-1.45%/+17.21% at 120K n3.
- Caveat: strict turnover was 58/60, then 57/60 on one unchanged repeat; candidate routed validation was intentionally not attempted.
- Evidence: finding ยท manifest.
Claim ledger¶
| Public claim | Conditions | Evidence |
|---|---|---|
| Candidate is comparatively faster in bounded cells | direct C1, strict eligible 4K/120K populations | comparison |
| Candidate was not promoted; baseline retained | strict turnover misses; user declined comparative exception | human decision |
| Baseline was restored | direct/routed 2/2 baseline checks | restoration |
User-reported other reports were not independently verified and do not affect this decision.