laya-nli-conflict-v10-s2 — research archive (round 10, v9-config noise baseline), NOT delivered

⚠️ Research archive — NOT a delivered model. This checkpoint failed its round's acceptance gates and was never shipped. The current production head is slow-stack/laya-nli-conflict-v12 (v12, delivered 2026-10-07; it superseded v4). Uploaded 2026-10-02 for provenance/backup while round 11 (multi-run verdict protocol) waits for Kaggle GPU quota.

Round 10 (2026-10-01) of the laya NLI memory-conflict head program by modusensus. S2 = the noise-baseline run of the v9 corpus and configuration (kernel-direct, complete artifact set incl. its own SHA-verified weights). Sibling runs: the original v9 round checkpoint (laya-nli-conflict-v9) and the L2 rerun (laya-nli-conflict-v10-l2).

Headline results

• main val 0.894 (✗), bias diagnosis 13/14 PASS
• Why this run matters: with v9 (0.9050) and L2 (0.8890) it forms the same-config noise band — main val 1.6pp swing, τ(noul) 1.03–1.19, and B2 = 0.889 / 0.3877 (this run) / 0.6951: a 50pp range crossing the 0.5 decision line ⇒ single-run B2 readings have no decision power; round 11 moves to a multi-run median protocol with a 9-case isomorphic family instrument.

Artifacts

file	value
model.safetensors	SHA256 5cbcc083…8724b6 (full hash in archive_sha256_manifest.txt)
rl_agent_config.json	τ(noul) = 1.1939; encoder jhu-clsp/mmBERT-base; bf16
metrics.json	val_accuracy 0.894, val_ece 0.0411, n_val 1000, no_rl true
val_probs.json	frozen-val probability dump (calibration analyses)

Provenance

• Training: Kaggle GPU kernel daphnelaurent/laya-nli-conflict-ce (three-version sequence), dataset daphnelaurent/nli-conflict-pairs v15
• Round record & full gate table: kaggle_eval/HANDOFF_NLI_V10.md