Phase 11.6
Learning & exploration
Inspect replay cutoffs, learned ranking lineage, exploration reasons and reviewed learnings without turning historical evidence into an opaque score.
Historical creative learning
Lineage & backfill coverage
Inspect which published creatives are matched, temporally aligned and safe to use for script-level learning before spending Analytics budget.
Import published Studio narration and verified YouTube IDs. The Studio directory is read-only. Draft durations, upload-time publication dates and guessed speech timing are not treated as measured evidence.
Scheduled imports run on the VPS automatically. Manual preview and import require a separate operator key; it is never stored by this page.
Enter the channel ID used by historical imports to inspect backfill readiness.
Replay-safe inputs
Historical datasets
Each row freezes features at its decision boundary and outcomes at its explicit cutoff.
No replay dataset has been built yet.
Learned ranking lineage
No calibration
No calibration has been created yet.
Exploration policy
No decision
No exploration decision has been created yet.
Human learning
Reviewed experiment learnings
No reviewed learning has been persisted yet.
Measured lineage
Every ranked decision exposes the exact immutable calibration version and its replay dataset.
Replayable cutoffs
Dataset decision/outcome cutoffs and feature-schema identity remain visible beside calibration inputs.
Exploration survives
Coverage-gap and exploration selections remain distinct from learned exploitation.
Rollback without rewrite
Decision history keeps the calibration version actually used; later decisions may use an older version or disable ranking entirely.