Decide behavioral-anomaly ranking and impact-validation methodology

[justsml]

Question

How should the system capture and rank surprising behavior as Research Observations, form citation-backed Investigation Assertions and Research Priorities, and validate impact across surfaces, identities, and accounts before promotion to a Finding? Cover malformed-type coercion, timeouts instead of validation errors, unexpected state resets, alternate credentials or JWTs, inconsistent endpoints, evidence sufficiency, protected read/write checks, cross-account effects, privilege change, secret exposure, integrity loss, deletion, rejection and coverage records, confidence, reachability, deduplication, and safe stop conditions.

[justsml]

Implemented and pushed a reversible impact-validation boundary on dan/wayfinder-research-flywheel: e18a65d. It keeps anomalies as Research Observations until cited, authorized, controlled, independently reset evidence demonstrates a reproducible protected security effect; rejected leads and coverage remain separate outcomes. The methodology decision ticket remains open for remaining ranking and workflow choices.