Raven Public Research Report
Decision Timing (Technical)
Does an earlier decision always mean a better decision? This technical version exposes the evidence mechanics, replay boundary and test lineage.
What this report shows
MTC and MTZ
The report separates measured time context from the decision point at which a weak signal can be treated as actionable evidence.
Safe uncertainty
Earlier is not automatically better. Raven must preserve UNKNOWN or HUMAN_REVIEW_REQUIRED when evidence does not support a confident recommendation.
Replay boundary
Replay is useful only when the visible inputs are frozen before the hidden outcome is compared.
Technical boundary
The technical version preserves source identifiers, test families and limitations without exposing private implementation paths.
Engineers should treat the numbers as internally reproducible counts from local artifacts, not as external validation.
Required customer validation
A customer pilot must provide incident history, topology context, ticket or dispatch evidence, decision timestamps and a mutually accepted ground-truth method.
Raven should prefer UNKNOWN or HUMAN_REVIEW_REQUIRED when those inputs are missing or contradictory.