Elamur Groupantsworld-principles-teacher-retention
The groupant beside world/src/principles.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 |
| composition | partial source model | composition |
| cost | partial source model | cost |
| descriptant | partial source model | epoch-memory |
| invariant | partial source model | invariant |
| tests | partial source model | 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 |
|---|---|---|---|
world-principles-teacher-retention-behavior-complement | behavior_complement | partial source model | behavior-complement |
world-principles-teacher-retention-reverse-reconstruction | reverse_reconstruction | partial source model | reverse-reconstruction |
world-principles-teacher-retention-coupling | constraint_relation | partial source model | coupling |
Model text
forward
round gates on disabled flag, daily stop and call budget, obtains pending batch, makes one external teacher call, then appends call count, verdict rows and ideas before write_world. Verdicts default U; parsed numbered lines overwrite earlier ones, accepting first T/F/U character; absent/invalid indices are ignored. write_world emits every cached T row with sources/notes then atomically replaces the derived world and attempts a brain update.
invariant
Teacher T is a reported judgment, not independently established truth. Cached records have six TSV columns but parser takes the first verdict character and does not validate the timestamp. File read failure becomes an empty cache. Calls/checked rows/world/brain updates are not one atomic transaction; concurrent callers can exceed budget or duplicate checks. Source frequency/conditions lost by key are not recovered by teacher wording.
cost
O(cache bytes + batch text + response bytes + true-row output), excluding teacher/network/knowledge costs. Budget bounds intended successful sequential calls, not concurrent or failure-recovery resource use.
composition
Source callers: main dispatches main_args; family invokes round once per research round and reports returned counts/errors; round consumes pending and statement then external teacher/cache/world operations. Native public join: caller-supplied know::Fact -> statement. Selection tests call the exact private pure helper used by pending with supplied keys, avoiding filesystem state.
epoch-memory
2026-10-30: initial source-bound review. Prospective cost model predicts no asymptotic change: selection helper extraction preserves sorting and cloning; list repair replaces one scalar division with checked division. Old zero-limit native witness divides by zero. Final implementation returns a nonzero stride and take(0) produces no rows. Mind production behavior is unchanged. No surrogate, speed or full-system correctness claim.
checks
cargo test -p world --lib principles::wave12_models -- --test-threads=1. File evidence.md binds repeated native profiles and compiled production controls. Partial syntax leaves macros opaque; source-bound syntax is not semantic AST verification.
behavior-complement
Domain and observation scope: Returned round/write_world traces with supplied cache, numbered response text and sequential successful I/O premises; no native teacher/filesystem evidence.
Conditional on supplied actual response/cache and successful sequential I/O, exclude verdict lengths other than batch length, non-U defaults for unmentioned indices, assignments outside indices 1..batch length, and an earlier value surviving a later accepted numbered overwrite. write_world's selected rows must be precisely cached T rows in order, preserving statement/source/note; selecting F/U rows is excluded. This partial absent-behavior model excludes neither false T judgments, racing budget overshoot nor partial publication after failures. Evidence is source inspection; no truth/transaction guarantee.
reverse-reconstruction
Final verdict fibers retain all response texts whose ordered in-range assignments end at that vector; U includes omitted lines, invalid initials and explicit U. Published T-row fibers retain all cache snapshots with exactly those T rows plus compatible non-T rows. Overwritten lines, timestamps and failed publication history remain ambiguous. Actual filesystem compatibility is a premise; displayed sources/T are not unique causal or truth witnesses.
coupling
No teacher/cache/world I/O executed. Truth, consistency of cached keys and state snapshots are explicit premises. General filesystem atomicity and fresh-process retained learning are outside this finite suite; no exam or lesson data is loaded.
Typed entries bind the manifest source revision, this section's domain/operator/observation, and the named finite checks. Status is partial source-derived model. No complete reconstruction search, universal proof or factual-truth guarantee is claimed.
This page in Markdown → https://md.groupants.elamur.ai/free/en/mova/world/src/principles.rs.groupant/teacher-retention.groupant/