Neutral-AI main study: collection ended; nine outcomes remain unknown

All 960 scheduled assignments have been processed across the two authorized sessions: 951 completed (99.1%), nine service-failed outcomes remain unknown, and none are untouched. The continuation completed 701 and skipped four, preserving the original 250 completed and five failed records. Collection ended on 4 November 2026 at 14:22 UTC, with exit code 0 and release of the owned sleep guard. Processing every assignment does not meet the all-outcomes-complete criterion. No failure was retried, replaced or imputed.

The independent combined audit accepted the records: 11,171 unique attempted calls, 11,166 saved responses, and 11,162 validated receipts/applied actions. All 49,406 release bindings and 108,995 actual archive hashes matched. Original and continuation processes are absent. Five generation failures have no completed response; four billing-lookup failures have saved but unapplied responses. Their charges remain unresolved, with reserves retained; neither late receipts nor a remote failure cause is assumed.

The planned primary buyer-minus-seller gain point estimate and ordinary confidence interval remain suppressed because eight of the 720 feasible negotiations are missing. The frozen full-sample legal identification bounds are -2.2431 to +1.0000 game points. The approximate population bound interval is -3.6942 to +2.4214, using family-level Welch endpoint limits and a Bonferroni adjustment. These are missing-outcome bounds, not an ordinary confidence interval around a point estimate. They allow both buyer and seller advantage, and do not establish equivalence. The four-point planning target was not an equivalence margin, and complete-data planning power is not a claim about this incomplete sample. Seventy-one of 80 families have all 12 outcomes.

Observed outcomes only: tradeoff negotiations reached agreement in 351 of 357 completed cases (three unknown out of 360 scheduled); narrow-zone negotiations reached agreement in 257 of 355 completed cases (five unknown out of 360); all 239 completed no-zone negotiations ended in walkaway (one unknown out of 240). No harmful agreement was observed among the 951 completed negotiations. These descriptive counts exclude failed outcomes and do not establish universal safety or full-study agreement rates. Pilot outcomes are not pooled. Scope is one model in this artificial negotiation game; generalization to other models or real negotiations is untested.

Conservatively rounded settled charges are $53.382054300. The nine unresolved calls retain $0.254865 in reserves, for a total recorded commitment of $53.636919300 within the $330 whole-main ceiling. The reserve is not confirmed spend. Prior pilot spending of $7.891284039 remains separate.

Disclosure: before main collection, the public category prior was amended to 3:3:2, adding six ASCII bytes and corresponding request allowance, and the outage guard was fixed at three consecutive or ten lifetime service-failed episodes. After the first session stopped and its partial economic outcomes had been observed and reported, the owner requested continuation following restored Wi-Fi. The separately reviewed amendment allowed only the 705 untouched assignments in original order, reset the new-session consecutive-failure count to zero, retained all five prior failures toward the lifetime limit, and added an owned idle-sleep guard with cooperative stop handling. The reported Wi-Fi outage is not an independently established cause of the service failures. Frozen scientific inputs, scoring, estimand and missingness rules remained unchanged; original artifacts and the earlier partial report are preserved.

The descriptive appendix SECONDARY-RESULTS-v1.json reports every category/start/orientation cell regardless of sign, full and partial family means separately, gain and slot distributions, and applied-action/query/invalid-action diagnostics by role. Unknowns are retained, and no secondary significance tests are added. No-zone results represent 20 independent families, not 240 independent scenario samples. The endpoint interval relies on independent-family and normal/CLT approximations; the appendix preserves family distributions for heavy-tail inspection.

Reporting review accepted: REPORTING-REVIEW-v1.json. Sources: PRIMARY-v1.json, RESULTS-v1.json, SECONDARY-RESULTS-v1.json and ../COMBINED-MAIN-REVIEW-v1.json. No additional collection, retry or publication is authorized by this report.