Downstream review
The primary comparison is Turn 3. Later player plans were generated separately from each branch's own history; all twelve paired player submissions for Turns 4–6 differ in bytes. These later trajectories cannot isolate the effect of Turn 3 exposure from new player choices and model variation. See FINAL_EVIDENCE.json for hashes and exact per-turn outcomes.
| Point | Enabled branch | Control branch |
|---|---|---|
| Turn 4 | Thirty jars formally received at Measure House; new river/road logistics established | Thirty jars received at Measure House and transported downriver into Barleybank's storehouse; disclosure defects repaired before delivery |
| Turn 5 | Barleybank collects thirty restitution jars by padded cart; a secondary consignment of 35 jars and 15 tiles is staged for harvest exchange | Forty additional jars are moved to Barleybank by Reedhaven boats; return-grain schedules are confirmed |
| Turn 5 institutions | Exchange Commons remains a hypothesis; initial restitution obligation marked performed | Exchange Commons remains a hypothesis; initial restitution performed, two bilateral obligations remain active |
| Turn 6 | Harvest exchanges and household reintegration approved and delivered; Exchange Commons activated as ACTIVE_VICTORY | Analysis and a proposed adjudication exist, but review rejected the proposal; corrective request exceeded the pinned input limit; no approved Turn 6 state or reports |
| Last approved leaders | Four active leaders, all SUPPORTIVE influence | Four active leaders, all CONDITIONAL influence at end of Turn 5 |
The control's Turn 4 proposal required three adjudication rounds. Earlier versions overexposed private storehouse and shipment details; the accepted version split disclosures by participation and acquisition path. No rejected version was delivered. This demonstrates the review gate operating, not that ablation caused a privacy defect.
At Turn 6 the control reviewer rejected a binding ceramic-vessel obligation that had only one party's proposal, rule terminology in a proposed public victory fact, and direct-observation claims for remote events. The corrective packet was 250,447 bytes, above the pinned 240,000-byte limit, and no corrective provider call started. The original request was 191,646 bytes. Source evidence and exact findings are in ../ablated/TERMINAL_DIAGNOSIS.json and the original run's call records.
The enabled endpoint is completed and verified. The control endpoint is technically inconclusive after five committed turns. Its unapproved sixth-turn proposal must not be counted as a harvest completion, victory activation or delivery. The full six-turn downstream comparison is therefore inconclusive. No winner, relative game quality, or stronger long-term dramatic effect can be inferred from the asymmetric stopping points.