learner-model

learner-model stores learning evidence in Loro 1.13. Observations and reviews are immutable records with unique IDs; mastery, evidence counters, FSRS cards, and due dates are deterministic folds, not independently merged mutable fields.

Why evidence is the authority

Two peers may import the same evidence in different orders. Recomputing from a stable (concept, evidence-id) order makes the result commutative, associative, and idempotent. Conservative merge rules keep evidence counters non-decreasing and select the earliest due date when schedules disagree.

Mastery uses the configured PFA-style fold after sufficient evidence. The exact derived value can change when new evidence arrives, but importing an already-seen evidence ID cannot increment counters or move a card again.

Migration

Legacy Loro snapshots are read into the new evidence model and written to a new snapshot. The original snapshot is preserved. Certification checks semantic continuity, evidence uniqueness, deterministic refolding, and conservative due dates before the migrated pointer can become current.

The worker and Sovereign Sync import path both call the same fold after local writes or remote Loro updates.

Canonical source: substrate/learner-model.