Elamur Groupantsschool-course
The groupant beside world/src/knowledge.rs in the Mova repository: what the code computes, what it never does, how to go back from its outputs, its invariants, cost and checks.
Facts
Implementation material. A Mova code model describes this code at one source revision: partial models, finite checks and their stated scope. It is not a reviewed general exposition, and its labels are shown as written, never raised.
Components
| component | label | where |
|---|---|---|
| approximant | partial source model | forward |
| ast | partial | ../syntax.json |
| cost | partial source model | cost |
| descriptant | present | epoch-memory |
| invariant | partial source model | invariant |
| tests | existing scoped checks | checks |
Obernants
The typed relations around the forward model: the behaviour complement (what the code never does), the reverse reconstruction (every input compatible with an output) and specialised views.
| id | type | label | where |
|---|---|---|---|
school-course-behavior-complement | behavior_complement | partial source model | behavior-complement |
school-course-reverse-reconstruction | reverse_reconstruction | partial source model | reverse-reconstruction |
school-course-control-partition | experimental_relation | partial source model | control-partition |
school-course-output-failure-prefix | failure_projection | partial source model | output-failure-prefix |
Model text
Source-bound partial models, epoch 2026-10-29. Subject: world/src/knowledge.rs at the manifest SHA-256. Observation: Diagnostic course orchestration from preprocessing through trace/results persistence. Every relation below concerns this frozen source/configuration. Structural validation does not establish semantic correctness.
forward
Load/preparse lessons; read kinds; optionally fork two controls; for each lesson optionally forget, read it, test-before, learn twice, test-after, the two question controls (question_controls: the next test's question of the same lesson; a question of other_lesson, half the course away) and exam; join controls and write optional results/trace.
invariant
Controls use separate students and shared cloned parse data, not the primary mutable weights. Reporting a better after-score is not a retained-learning or causal improvement proof.
cost
Preparse + main learner + concurrent controls; latency is not the sum of thread CPU costs. Skip-before removes an assessment pass; trace memory scales with emitted rows.
checks
knowledge::control_tests and trace_tests validate local reporting; full course execution intentionally not run. See ../tests.md for executed scope. A proposed oracle below is not a completed check.
epoch-memory
2026-10-29: inspected the existing implementation before modeling. Preserved current behavior, including lossy decoding, index-sensitive ties and conditional updates. Existing witness tests distinguish active-scope learning from a wrong global witness. Counts represent distinct observations, not independent proofs. No source changes, performance gain, full-course result or autonomous combiner is claimed. AST remains explicitly absent; ranges bind source bytes only. Parser resources, panic/allocation behavior and complete inference semantics remain gaps.
2026-11-04: the course loop prints gap_diag lines to stderr for each test not understood when MOVA_GAPDIAG=1; the trace TSV and the report are unchanged (byte-identical traces, ten courses, two runs).
2026-11-04 (student-control): the fourth control «questions from another lesson» (question_controls, other_lesson). Prediction: C_cross ≤ C_same, both red; a topic-only switch raises C_same, not C_cross. Measured on the ten courses: trace rows byte-identical to the base, C_cross 2 178 against C_same 2 354 of 7 283 (student 3 204), red in all ten; lower than C_same in seven. student-qents-mods then: student +30, C_same +20, C_cross +2 (kept). Test the_cross_lesson_control_tells_topic_from_question: a topic reader is red only against the cross-lesson control, an options-only reader against neither; the mutant other_lesson = li fails it.
behavior-complement
Type: behavior_complement. Domain and observation: Diagnostic course orchestration from preprocessing through trace/results persistence, as narrowed below. Operator/target: U is terminating configured course traces, R is run_course_lessons. Inner-bound U\R excludes before-test calls when before_tests=false, control forks when controls=false, and school results writes when save_school=false. Optional explicit trace writes remain independent of save_school.
Approximation: source-derived scoped relation, conditional on valid internal state and fixed flags/resources; unmodeled behavior remains unknown. Evidence/oracle: Source inspection; no filesystem side-effect oracle executed.
reverse-reconstruction
Type: reverse_reconstruction. Domain and observation: Diagnostic course orchestration from preprocessing through trace/results persistence, as narrowed below. Operator/target: For aggregate (right,total,understood,exam), compatible histories include all per-row outcomes satisfying sums and per-lesson order. Totals cannot recover labels, exact weights, lesson histories or parse cache contents. Trace rows refine but do not collapse that preimage to one state.
Approximation: source-derived scoped relation, conditional on valid internal state and fixed flags/resources; unmodeled behavior remains unknown. Evidence/oracle: Existing trace test validates selected row projections; aggregate inverse enumeration is proposed.
control-partition
Type: experimental_relation. Domain and observation: Diagnostic course orchestration from preprocessing through trace/results persistence, as narrowed below. Operator/target: Control students share initial skeleton/kinds and cloned parse trees; no-lesson and key-removed tracks alter lesson reading. Shuffled-question control keeps options but rotates questions only when there are >1 tests. The cross-lesson control keeps options and takes the question at the same place (cyclically) of lesson (l + n/2) mod n, never l, when n > 1 and that lesson has tests. Therefore both denominators can differ from the main total and control_red compares ratios. The same-lesson shuffle removes question-specific evidence only; the cross-lesson one removes it and the lesson's topic, leaving an option prior.
Approximation: source-derived scoped relation, conditional on valid internal state and fixed flags/resources; unmodeled behavior remains unknown. Evidence/oracle: Existing control-share unequal-denominator negative case; no empirical causal claim.
output-failure-prefix
Type: failure_projection. Domain and observation: Diagnostic course orchestration from preprocessing through trace/results persistence, as narrowed below. Operator/target: After inference and joining controls, optional school results write precedes optional trace directory/write. A late trace error can coexist with already written results; operations are not transactional. Conversely preparse failure prevents lesson loop and those writes. Reconstruct failures as compatible prefixes, not rolled-back state.
Approximation: source-derived scoped relation, conditional on valid internal state and fixed flags/resources; unmodeled behavior remains unknown. Evidence/oracle: Proposed injected filesystem failure oracle; no mutations to runtime source.
This page in Markdown → https://md.groupants.elamur.ai/free/en/mova/world/src/knowledge.rs.groupant/course.groupant/