0 passed0 failed

Scenario History And Proposal Normalization

Scenario history and proposal normalization keep protected checks reusable across tuning loops.

Map keys: promise.evaluation, promise.improvement, rule.cost-and-proof-freshness, rule.agent-human-resumability. Evidence path: deterministic plus held-out eval. Evidence status: open gap. Next action: connect scenario proposal sources, scenario history cadence, normalizer coverage, and held-out selection to eval and improve packets. Terms covered here: scenario proposal, scenario history, protected check, held-out, iterate cadence, train cadence, normalizer, proposal packet, context recovery, skill failure episode.

Maintainer Promise

Cautilus preserves useful scenarios as durable candidates that can be reused for eval, held-out validation, and improvement loops, and normalizers produce inspectable proposal packets rather than hidden one-off shapers.

Subclaims

  • Scenario proposals are stored as durable candidates that another agent or operator can reopen.
  • Held-out selection and protected-check rules stay explicit so iterate-vs-train cadence does not silently drift.
  • Normalizers produce inspectable proposal packets rather than hidden one-off shape transformations.
  • Scenario history cadence is observable so reuse across tuning loops is auditable.

Evidence

Evidence Gaps

  • Held-out eval result packet attached to a scenario-history cycle so the reuse contract has a reopenable end-to-end proof. Owner: maintainer. Next action: capture a self-dogfood scenario-history cycle with held-out validation under artifacts/self-dogfood/ and link the summary; no checked-in artifact today.