Elamur Groupantsworld-induce
The groupant beside world/src/induce.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 | approximant |
| ast | partial | syntax.json |
| composition | partial | composition |
| cost | partial | cost |
| descriptant | present | descriptant |
| invariant | partial | 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 |
|---|---|---|---|
behavior_complement | behavior_complement | partial source derived model | behavior_complement |
reverse_reconstruction | reverse_reconstruction | partial source derived model | reverse_reconstruction |
world_induce_invariants | world_induce_invariants | partial source derived model | world_induce_invariants |
world_induce_resource_cost | world_induce_resource_cost | partial source derived model | world_induce_resource_cost |
world_induce_representation_join | world_induce_representation_join | partial source derived model | world_induce_representation_join |
world_induce_domain_partition | world_induce_domain_partition | partial source derived model | world_induce_domain_partition |
Parts
Model text
descriptant
world-induce
acode(f,v)=((f<<16)|v)+1; pack1..3 sorted codes at21bit offsets. Fix=number of head-correct/label-wrong words carrying subset and current/from,target/gold; Break=head-and-label-correct words carrying(from,subset). Sort (-Fix+Break,atom_count,Rule). apply changes labels using per-rule feature snapshot; run needs precision>=.85 and Fix>=min+Break. explore_all gd>0,gh>=0 reports p without significance filter; selfplay adoption needs p<.05.
Modeling epoch 2026-10-31
Prospective source/branch/cost hypotheses were saved externally before source edits. This partial source-derived model retains observed boundaries rather than claiming complete verification. Pure codes/counting/parsing/changes/threshold tests; feature extraction, persistent effects and live learning remain partial, no surrogate.
approximant
Forward relation
acode(f,v)=((f<<16)|v)+1; pack1..3 sorted codes at21bit offsets. Fix=number of head-correct/label-wrong words carrying subset and current/from,target/gold; Break=head-and-label-correct words carrying(from,subset). Sort (-Fix+Break,atom_count,Rule). apply changes labels using per-rule feature snapshot; run needs precision>=.85 and Fix>=min+Break. explore_all gd>0,gh>=0 reports p without significance filter; selfplay adoption needs p<.05.
Domain: Synthetic aligned Sent/gold/Feats passed directly to real scoring; no annotator, corpus, journal, shipped-cache, training/adopt or GPU paths.
No executable percent surrogate: these categorical/effect observations lack a useful approximation metric. Future supported caps1/2/5/10% need a defined domain/metric and real reference evidence; no error or speed bound is claimed.
invariant
Explicit total order removes FxMap iteration dependence. Empty LAS0, bootstrap1;1000 fixed-seed resamples, all positive deltas p0. parse_rules preserves valid atom order/duplicates. Add exact-dedups while holds_in checks broader coverage, not identical operation effect.
Cost component: cost.md.
cost
Subset work W*(A+C(A,2)+C(A,3)), candidate sort O(G log G), local per-worker counts/maps. Feature extraction scans sentence dependents roughlyO(L^2) plus delegated graph costs. u32 counts/indices and i64 sums require representability. No timing claim.
composition
family transfers Change/scope then adds admission/guard rules. big matrix pseudo-labels exclude matrix feature predicates for new rules. GPU adapter only flattens head-and-label-correct words; equivalence requires valid widths/count bounds.
models
behavior_complement
Wrong-head words excluded from both counts. apply cannot alter heads/forms. Invalid atom fragments silently dropped; empty conjunction accepted. explore_changes can return p>=.05. Atom/parser feature domain not enforced at public parse_rules; malformed values can panic show/encoding.
reverse_reconstruction
Valid packed subset uniquely decodes its1..3 atoms, but leaves omitted atoms unknown. Count vectors admit word/sentence permutations and arbitrary wrong-head additions. Changed label to admits prior from matches and prior to; inverse needs ordered rules/feature context. Change removal preimage preserves arbitrary removed multiplicity.
world_induce_invariants
Explicit total order removes FxMap iteration dependence. Empty LAS0, bootstrap1;1000 fixed-seed resamples, all positive deltas p0. parse_rules preserves valid atom order/duplicates. Add exact-dedups while holds_in checks broader coverage, not identical operation effect.
world_induce_resource_cost
Subset work W*(A+C(A,2)+C(A,3)), candidate sort O(G log G), local per-worker counts/maps. Feature extraction scans sentence dependents roughlyO(L^2) plus delegated graph costs. u32 counts/indices and i64 sums require representability. No timing claim.
world_induce_representation_join
family transfers Change/scope then adds admission/guard rules. big matrix pseudo-labels exclude matrix feature predicates for new rules. GPU adapter only flattens head-and-label-correct words; equivalence requires valid widths/count bounds.
world_induce_domain_partition
Synthetic aligned Sent/gold/Feats passed directly to real scoring; no annotator, corpus, journal, shipped-cache, training/adopt or GPU paths.
tests
Finite actual-code evidence — 2026-10-31
Names: ["induce::wave29_worldlearn::counting_requires_correct_heads_and_orders_candidates", "induce::wave29_worldlearn::packed_subsets_and_empty_bootstrap", "induce::wave29_worldlearn::permissive_parse_and_set_valued_changes"]
Selected CPU command: CARGO_BUILD_JOBS=2 CARGO_TARGET_DIR=ISOLATED_TARGET cargo test --offline --locked -p world --lib --example retention_probe [--release] wave29_worldlearn -- --test-threads=1
15 selected cases passed twice debug and twice release with identical sorted names/outcomes, exits0. control-induce: compiled source mutation failed with exit101; source restored byte-for-byte and15positive tests passed. No live teacher, corpus, state activation, GPU, or full-school run. Temporary synthetic files use unique process-id paths and are removed after passing tests.
Pure codes/counting/parsing/changes/threshold tests; feature extraction, persistent effects and live learning remain partial, no surrogate.
Source SHA256: 0297cbe15b922cb5aeebd8632fbb9bcb3d0d8da3e6a6dd287b0978638c1fe2ee. Syntax generated independently twice and compared; lexical only, no semantic proof or CUDA compilation. Strict whole-crate Clippy was not repeated merely to reproduce existing dependency/source lints; Cargo warnings retained in receipts.
This page in Markdown → https://md.groupants.elamur.ai/free/en/mova/world/src/induce.rs.groupant/