> Public sanitized report copy. Local infrastructure paths and opaque contract identifiers are redacted where present, and nonessential internal process/authorization commentary is omitted; financial values and scientific limitations are retained. Historical pending/open wording below records the report-time state, not current program status. The scoped successor program subsequently completed; commissioning-delay and recontracting extensions were deferred, not executed. Historical predecessor holds remain unchanged. This is report-level evidence, not a raw-data reproduction bundle. Referenced checker, CSV, database and provider artifacts are not included. No new experiment or audit was performed for this export.

# Parent acceptance: disclosed settlement-order control

Status: ACCEPTED (bounded control).

Execution: 24 accepted provider decisions, 24 committed native turns, 6 episodes, no stops, 1,780 source pins unchanged, cash conserved at 544.4 per episode. Independent reconstruction (`check_saved.py`, 5,213 assertions, SQLite opened mode=ro&immutable=1, run dir hashed before and after, no side files) passed.

Finding: with the settlement order stated in the brief and nothing else changed, Lab A borrowed before conditional credit was withdrawn in 6/6 episodes (undisclosed comparator 1/6) and went dark in 0/6 (comparator 5/6). Every episode took the same path: request 0,0,5,0; the 5 drawn at 800 bps in 2028-H1. Lab A's contribution less cash interest was -19.17 (fixed rival) and -19.7 (liquidity rival) against -27.5 in the comparator's dark episodes.

Interpretation: the rival-adaptation reversal was largely a disclosure artifact. The agent's poor final-turn outcome followed from not being told the payment order, not from the rival's adaptation. The rival-adaptation acceptance's qualification is upheld and now resolved in this direction. Lab A's plans restate the order in all 24 decisions (none in the comparator); its short rationale still cites withdrawal risk, so which consideration drove the borrowing is not identified.

Combined two-lab totals fell (-39.7, -39.39 vs -37.5, -37.19) because both labs now operate into one unit of final demand and A pays interest; the comparator's higher total came from A going dark and leaving 2.5 unpaid. Not a finance improvement in either direction.

Limits: before/after, not randomized; same model identity (gpt-6-astra-900k -> gpt-6-astra), different run time; identical replicates within cells; no significance, rates or welfare claims; same model family analyst.

Design lesson carried into the information-timing study: state engine settlement order in every brief by default, and treat any undisclosed rule as an explicit, deliberate treatment.
