Corpus and epochs

The corpus is the active, evidence-backed learning surface—not the complete decisions ledger.

Three layers

Scientific history

Append-only decisions, receipts, audits, sources, trial accounting, and holdout burns remain available for provenance and review.

Active epoch

An epoch identifies the evidence currently eligible to support observations, family principles, and cross-family mechanisms. Every active record is stamped with its epoch.

Derived projections

Graphs, dashboard JSON, queues, reports, and knowledge-store indexes are regenerated from eligible records. They may be archived and rebuilt without erasing scientific history.

Epoch transitions

A safe transition follows:

preflight → activate → replay → rebuild → verify

Preflight pins the engine, catalog, inventory, and capability fingerprints. Activation archives derived stores while preserving protected records. Replay re-adjudicates only claims with sufficient fidelity metadata. Rebuild constructs the new corpus, and verify reconciles every count and invariant.

Legacy Epoch 0

Historical records that predate epoch and search metadata remain visible as Legacy Epoch 0. They may be reviewed but cannot silently support the active corpus. To regain active status, a claim is rerun as a new, preregistered experiment.

Corpus eligibility

Eligibility requires more than a decision row. Evidence must be analysis-backed, non-synthetic where required, faithfully reconstructed, correctly scoped to the active epoch, and supported by honest cohort metadata. Routing states and integrity failures do not become learning evidence.