{"id":"research.feedback_predicts_paying.v1","title":"Does feedback volume predict that an agent starts paying","unit":"ROC AUC (0.5 = chance) with a 95% bootstrap interval","chain_set":"eip155:8453 (Base)","window":"feedback before T0; outcome in the 30 days from T0","definition":"Cohort: Base ERC-8004 agents registered before T0 with a wallet binding valid at T0 and no counted payment before T0. Score: the number of reputation feedback entries the agent received before T0, after removing every entry from the ten raters with the most entries before T0. Outcome: at least one counted payment (pay.usd.total.v2 rules) in [T0, T0 + 30 days). AUC of the score against the outcome, ties counted half.","method":"SQL over registrations, bindings, reputation_feedback and money_events; AUC computed twice (grouped exact pair count in SQL, rank sums in JavaScript); 95% percentile bootstrap, 1,000 resamples, seed 20261023, rows in a fixed order. Decision rule fixed in advance: better than chance only if the interval's lower end is above 0.5, worse only if its upper end is below 0.5. Main T0 2026-09-15. Method fixed 2026-10-23T09:13Z before the query ran.","excludes":["Agents already paying before T0.","Agents with no wallet binding valid at T0 (their paying cannot be observed)."],"does_not_assert":["That feedback causes or prevents paying.","That any rater is fake or any agent bad.","That paying is a measure of quality."],"assumptions":"A payment we cannot attribute to the agent's bound wallets counts as not paying.","known_gaps":"Binding records begin 6 Sep 2026, so a T0 before then has no cohort (the pre-registered 15 Aug robustness run could not be computed). 912 of the 24,727 agents in the main cohort have a bound wallet the payment scan had not read by 23 Oct.","intended_use":"Testing whether reputation volume carries information about later economic activity.","not_intended_for":"Ranking, scoring or judging agents or raters.","source":"AFG-Pulse, one-off research query","since":"2026-10-23","status":"answered 23 Oct 2026: AUC 0.4524 (0.4218-0.4866) on the 15 Sep cohort of 24,727 agents, worse than chance; a 22 Sep cohort added afterwards (not pre-registered) gives 0.5344 (0.4985-0.5722). The manifesto expected no better than chance. Results: /research/manifesto-answers-2026-10-23.json","revisions":[],"body_sha256":"b5fa0308084b4867a164eab1e55c55cdc6bb5d6498921d1cd8f1ea68c3d72e99","url":"https://agenticfinancegraph.com/def/research.feedback_predicts_paying.v1","ladder":"https://agenticfinancegraph.com/ai-agent-liveness-methodology-l0-l9-ladder","license":"https://creativecommons.org/licenses/by/4.0/","attribution":"Agentic Finance Graph — https://agenticfinancegraph.com"}