Longitudinal Skill State & Mastery Model

The real problem

A latest score, a completion flag, or one strong project is not the same thing as mastery. This lesson builds Longitudinal Skill State Ledger & Mastery Transition Model v1 so every state change can be traced to time-stamped evidence, context, independence, quality, contradiction, staleness, and provenance.

The end product must support bounded, reproducible promotion, hold, downgrade, or re-open decisions without inventing certainty.

Lesson boundary

This lesson does not own the competency taxonomy and does not design the diagnostic/adaptive algorithm. MKN-003 owns competency/prerequisite structure; MKN-005 owns adaptive assessment decisions. MKN-004 owns the longitudinal history and transition logic for one competency.

Keep three layers separate: event ≠ sufficient evidence, evidence ≠ mastery, mastery state ≠ credential.

Mental model: ledger plus transition

Each ledger row is a traceable observation, not a final verdict. A transition engine reads prior state, evidence type, context, independence, quality, contradiction, and staleness before proposing a next state under a named rule.

When evidence is weak or incomplete, typed states such as UNKNOWN, DEVELOPING, PROVISIONAL, HOLD, CONTRADICTED, and REVIEW_REQUIRED remain visible.

The 16-field contract

The artifact binds learner, competency, evidence, event time, evidence type, context, independence, quality, state before, transition rule, state after, confidence, contradiction, staleness, review due, and provenance.

event_time orders observations; review_due schedules re-evaluation. evidence_type separates activity-only observations from competence-bearing evidence. provenance lets a reviewer reconstruct the synthetic source.

Why the newest event cannot overwrite history

Recency is only one evidence dimension. A strong artifact from a new context can be valuable while still missing prerequisite, transfer, or independence dimensions. Older evidence can remain informative while also becoming stale and needing refresh.

Contradiction must remain explicit. If a learner was previously MASTERED but a new transfer failure challenges that claim, the model must re-open review rather than hide the conflict.

Caliper can standardize learning-activity events such as assessment, media, reading, session, and tool-use events. Mapping those events to mastery transitions is a separate local design rule.

Worked baseline

A learner has a strong diagnostic from six months ago, a failed transfer task last month, several lesson-completion events this week, and a strong independent project artifact today. The completions are activity signals. The project is competence-bearing, but the transfer failure creates an unresolved contradiction. A bounded result is review/re-open, not automatic mastery.

The working table includes baseline, invalid shortcut, and changed-context transfer cases.

Common failure modes

1. Newest-event wins.
2. Completion/access is inflated into competence.
3. New contradictory evidence is erased by a strong artifact.
4. Stale evidence is treated as permanently current.
5. Context changes are ignored.
6. Provenance is hidden so the transition cannot be reproduced.

Choosing a bounded state

Use MASTERY_EVIDENCE_REVIEW_READY only when evidence is multi-dimensional, sufficiently independent, current, and contradiction-free. Use MASTERY_DEVELOPING when evidence is growing but incomplete, MASTERY_REVIEW_STALE when core support has aged out, and MASTERY_REOPEN_CONTRADICTION when new evidence conflicts with the prior state.

These are evidence states, not credentials and not authorization to mutate a live learner record.

Production implementation bridge

A production ledger should preserve history. Transition rules should be versioned so the system can later explain why a state changed under a specific rule. A derived current-state view may be materialized for performance, but historical evidence must not disappear when state changes.

This lesson does not prescribe a statistical estimator or Bayesian model; such choices need separate policy and validation.

Guided and independent practice

Run the baseline fixture and populate all 16 required fields. Change the newest event to activity-only: promotion must stop. Add a contradictory transfer failure to a prior MASTERED state: the model must re-open review.

For independent practice, make all supporting evidence stale except one weak recent signal. For transfer, use a different competency with one strong artifact but sparse evidence dimensions and recompute from scratch.

Verification evidence

Ask four questions: Are all decisive fields valid? Are activity-only and competence-bearing evidence separated? Are contradiction and staleness visible? Can a reviewer reconstruct state-before → rule → state-after from provenance?

The offline lab runs these gates against baseline and mutated fixtures without network access or real learner data.

Assessment and transfer

Assessment covers the owned boundary, baseline ledger construction, diagnosis of the newest-activity shortcut, and transfer to a different competency with sparse evidence plus one strong artifact.

The baseline label is never an oracle for transfer. Context, dimension coverage, independence, quality, contradiction, and staleness must be re-evaluated.

References and freshness

FACT: 1EdTech Caliper Analytics 1.2 provides a framework and shared vocabulary for consistently capturing and presenting learning activity, with multiple metric profiles. INFERENCE: Keeping activity events separate from mastery state improves auditability. DESIGN: MAKNOW owns the state vocabulary, transition rules, contradiction/staleness policy, and review cadence used here.

Source: 1EdTech Caliper Analytics — https://www.1edtech.org/standards/caliper (freshness checked 2026-09-14; review due 2027-01-15 or earlier on material change).