Decouple MasteryScore from SessionRotationReady (cross-session evidence requirement)

[davidortinau]

Background

Filed as a follow-up to #191 (Wash's vocab-quiz scoring proposal + language-tutor SLA consultation).

The current MasteryScore algorithm conflates two distinct signals:

1. Session rotation readiness — "I've shown enough evidence in this session to stop hammering this word; let spacing handle it."
2. True acquisition / mastery — "This word has been acquired in long-term memory; we can confidently call it 'known'."

Today both are driven by the same MasteryScore field, which can climb 0 → 1.0 in a single session. Per SLA literature (Cepeda et al. 2006 spacing effect; Bahrick & Phelps 1987; SM-2/Leitner/FSRS families), this overclaims true mastery — eight back-to-back correct answers in one session demonstrates working-memory access, not consolidation.

Tutor's recommendation

Cap single-session contribution to MasteryScore at ~0.60–0.70, and require a successful retrieval after a ≥24h gap to climb past that. This is exactly the Pimsleur/Leitner/SM-2 family logic and is well validated (Karpicke & Roediger, 2008 — the testing effect with spacing, not just testing alone).

Proposed split

• SessionRotationReady (UI/session layer) — aggressive thresholds OK; this is the "stop drilling in this session" signal. Latter quiz rounds rapidly empty #191's resolution is appropriate here.
• MasteryScore (long-term acquisition estimate) — require evidence across ≥2 sessions on different days before crossing ~0.80; should decay between sessions; should NOT reach 1.0 from a single session.

Acceptance criteria (sketch)

• Single-session correct streaks cap MasteryScore contribution at ≤0.70 (configurable constant).
• Climbing past 0.70 requires a successful retrieval ≥24h after the previous correct retrieval.
• Existing user data preserved monotonically.
• Schema: verify/add LastCorrectRetrievalAt per VocabularyProgress.
• Repro tests asserting cross-session gating BEFORE implementation.

Cross-references

• Latter quiz rounds rapidly empty #191 — current iteration
• Wash's proposal: .squad/decisions/inbox/wash-vocab-quiz-scoring-proposal-191.md
• Language-tutor SLA review key quote: "The math is fine; the label is wrong."