Readiness Drift Timeline

Did we keep getting safer — across every run, not just A vs B?

The A/B bake-off answers did this one change help?. AgentSmack's drift timeline answers the longitudinal question a hardening team actually has: feed it an ordered series of prompt-check readiness reports and it projects the multi-run trajectory — the overall trend, the peak and trough, the per-step regression history, and the per-dimension journey from first run to current. It reuses the same readiness diff engine the bake-off uses for every adjacent step, so the deltas are byte-identical to a pairwise compare. The verdict is sticky-maximal: a series that ends less ready than it started is net_regressed and can never launder up to improving. Pure, deterministic, migration-free — the caller supplies the series, no DB. Load the sample series below to watch the trajectory render with no live infra.

Want the whole-stack scorecard trend instead of the prompt-check roll-up? Posture Drift Grader → grades a series of unified production scorecards across the 22+ production dimensions.