feat: model longitudinal CEFR development and drift from immutable result events

[seonghobae]

Goal

Consume immutable CEFR domain-result references over time and estimate language-development, context, membership, and drift effects without re-owning assessment execution or base psychometric scoring.

Boundary

TEPP owns temporal/event, multilevel, cross-classified and multiple-membership analytical artifacts. Psychometrics Commons owns result snapshots; fast-mlsirm owns reusable psychometric measurement/calibration kernels; the LRS owns learning-event evidence; the LMS owns placement/completion decisions.

Required vertical

• ingest a versioned CEFR result-observation event that pins result reference/digest, target language, domain, claim status, scoring/cut-score/instrument versions, observed/recorded/ingested clocks, and authorized context-membership references;
• preserve domain profiles and uncertainty rather than converting A1–C2 labels to equally spaced integers;
• model within-person versus between-person change, occasion effects, delayed availability, item/rater/scoring-version drift, and time-varying multiple membership;
• distinguish measurement-scale drift from actual learner development;
• produce source-text-free longitudinal summaries and review triggers, not rewritten historical results;
• route any LLM interpretation through contextual-orchestrator, while deterministic/Rust results remain authoritative.

Acceptance evidence

• true trajectory/state, occasion, context, membership and drift recovery with bias/RMSE/coverage and failure denominator;
• irregular event-time, out-of-order arrival, correction/supersession, duplicated event, missing domain, scoring-version change and cross-tenant cases;
• one-hot nesting parity, cross-classified and weighted multiple-membership recovery;
• CPU multithread and supported accelerator determinism/parity;
• 100% production statement, branch, edge-case, and public-doc coverage;
• exact consumer contract tests against the released artifact corresponding to learning-interoperability-contracts PR feat(temporal): add typed six-clock values and uncertain intervals #5;
• APA 7th doctoring, ADR, PRD/TRD, UML/ERD, operability, changelog, traceability, and docs/product-technical-gap-baseline.md updates.

Non-claims

A temporal trend is not automatically causal growth, an intervention effect, or evidence of CEFR linking. Historical result snapshots remain immutable.