Elamur Groupantsschool-student-vidtvorant-rust
The groupant beside world/src/knowledge.rs.groupant/student-vidtvorant.groupant/rust_approximant.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 | source derived | forward |
| ast | partial | syntax.json |
| cost | measured 2026 11 04 | cost |
| descriptant | present | epoch-memory |
| invariant | source derived | invariant |
| tests | executed bounded checks | tests |
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-student-vidtvorant-rust-behavior-complement | behavior_complement | partial source derived model | behavior-complement |
school-student-vidtvorant-rust-reverse-reconstruction | reverse_reconstruction | partial source derived model | reverse-reconstruction |
Model text
File model of rust_approximant.rs, the executable code of the student's vidtvorant. The vidtvorant itself (subject,
domain, observations, error and performance profile, conformance) is described in ../model.md; this
file models the code.
forward
words = knowledge::words (alphanumeric split, length > 1, stop and negation words out, mentions::stem).
Memory::read_line appends a line (lesson, word set, negation flag, numbers) and indexes its words.
predict(mem, lesson, question, options, params): tail cut; question words; gap flag selects the gate; anchors by
shared-word counts (current lesson first, strongest only, ten); targets (anchor/next/linked); per option own lines;
features [asked, echo, missing, neg-mismatch, route×4, current]; score w·f; structured and ≥ gate to count; best
and best-current per option; lesson-first pool; argmax (score, current) with earlier index on ties; subset/extra
rule outside lesson-first; unique counted number override. replay reads a course lesson by lesson and calls back
per traced question. fit: averaged integer perceptron over the training fold, updating only on supported
disagreements. TimeModel::fit: least squares on ln ns with ridge 1e-6, Gauss–Jordan with partial pivoting.
invariant
Deterministic: ordered maps and sets only, no randomness, no clock in prediction or fit. predict never panics on
the declared inputs (empty question words or no options return the guess). Weights change only by
φ(student choice) − φ(predicted choice). The time model's coefficients are finite when the design matrix has full
rank; a zero pivot leaves the coefficient 0.
cost
read_line: O(w log V). predict: O(P_q + Σ_options (P_o + T·w_o)) for posting-list sizes P, targets T ≤ 30 and
option words w_o. fit: epochs × replay. TimeModel::fit: O(n·7² + 7³). Measured 2026-11-04: 40–48 µs median per
prediction on one P-core thread.
behavior-complement
The code does not parse, tag, look up WordNet, read the kind base or rules, touch the file system, the network, the student's state or any production path; it does not certify a bound; it does not predict exercises or learning.
reverse-reconstruction
A returned prediction is compatible with every question/option/memory triple with the same word bags, gap flag, negation flags, numbers and line-to-lesson map; fitted weights are compatible with a cone of weight vectors.
tests
Six unit tests, run with cargo test --release -p mova-approx --example student_vidtvorant (the runner includes
this file): stemming and stop words; supported option wins and an unread topic guesses first; the support gates
(and the no-gap gate only without a gap); current lesson first; fit moves only on disagreement and only inside the
training fold; time model recovers a planted law to 0.1%.
epoch-memory
2026-11-04: first version; the module is included by path from the runner, so it compiles only with it.
This page in Markdown → https://md.groupants.elamur.ai/free/en/mova/world/src/knowledge.rs.groupant/student-vidtvorant.groupant/rust_approximant.rs.groupant/