Positive-path engineering gate
2026-09-11: 75 tests passed, zero failures, across simulation runner and unchanged live-state tool tests. The new cases are in tools/simulation-runner/tests/paired.test.mjs.
The fixed adapter retrieves a real source ID, returns ADAPTED with eight findings, repairs an invented-ID attempt, survives interrupted simulation delivery, commits, verifies checksums and resumes without another model call. Separate cases remove each causal-fit field and require deterministic failure. A report containing a private disposition label is blocked before mailbox writes. A bounded negative case proposes unsupported settlement collapse and is rejected by the independent scripted review, without commit or delivery. Paired forks preserve exact shared files, omit player and analyst calls at the branch point, and remove selection material from the control wire request. Aggregate accounting deduplicates copied history and stops at the common limit.
These are deterministic engineering fixtures. The independent fixture review has a fixed verdict for the injected unsupported outcome; it is not a general semantic range prover. In live model runs, causal-range support and paraphrased disclosure remain subjects of independent model review. The fixture's positive choice supplies no evidence of voluntary model use or comparative dramatic value.
Supplemental recorded replay
After the real decisive turn completed, replay-positive.mjs preserved that exact accepted analysis, ADAPTED GM output, reviews and private drafts as fixed responses in RECORDED_POSITIVE_RESPONSES.json. broken-measure-positive-fixture-01 successfully committed one deterministic Turn 3 and verified four simulation deliveries. A second invocation made no additional calls. RECORDED_POSITIVE_VERIFICATION.json records the result. This replay uses the Broken Measure's actual causal facts and locked submissions; it is explicitly post-run engineering confirmation, not a replacement for the pre-run tests or an independent observation of voluntary GM choice.