The bench

Travis-ML dose ladder (Qwen2.5-1.5B)

ConcededMatched referencecohort

Matched reference — read against a recipe-matched benign population from the same corpus — the strong regime.

Qwen2.5-1.5B-Instruct LoRA r16 dose ladder (competitor 2605.30189 corpus).

The reading

VerdictHONEST NULL — no dose-response

Spearman rho=0.038 (perm_p 0.84, CI [-0.32, 0.41]); dose-contrast AUC 0.524. The clean-vs-poison AUC 1.0 is a SEED artifact (n_clean=1, seed-42 confound) — reported for completeness, NOT claimed. Published as a null, not forced to a number.

The aggregate

The rank-separability score and the shipped operating point, side by side. The verdicts are a projection at a false-positive budget, not a raw score — coarse reads only, no detector numbers.

AUCwithheld
n (poison / clean)31 / 1

No AUC is reported for this cohort — see the caveat above. We don’t manufacture a rank-separability number where the design doesn’t support one.

Specimens

No materialized specimens. No specimens were materialized for this cohort — see the reading above for what the read returned and why.