Elamur Groupantsworld-quant
The groupant beside world/src/quant.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-quant-behavior-complement | behavior_complement | partial source model | behavior-complement |
world-quant-reverse-reconstruction | reverse_reconstruction | partial source model | reverse-reconstruction |
world-quant-precedence | grammar_relation | partial source model | precedence |
world-quant-origin-prefix | constraint_relation | partial source model | origin-prefix |
world-quant-diagnostic-effects | state_transition | partial source model | diagnostic-effects |
world-quant-transfer-fiber | reverse_reconstruction | partial source model | transfer-fiber |
world-quant-overwrite-loss | information_loss | partial source model | overwrite-loss |
world-quant-sentence-boundary | encoding_relation | partial source model | sentence-boundary |
world-quant-reply-blocks | priority_relation | partial source model | reply-blocks |
world-quant-external-budget | resource_boundary | partial source model | external-budget |
Parts
Model text
model
world-quant
Exact quantity grammar, rational state transitions, sentence-local literal provenance, answer gates and external adapters.
forward
Sentence-prefix validation applies only to numeric literals. References read the current command-order state and carry no sentence provenance. Thus @2 set a.x=19 followed by @1 set b.x=a.x and @1 ask b.x remains green when 19 occurs only in sentence 2. This is a documented current limitation, not complete temporal provenance or a proposed behavior change.
Fold parsed commands over (state, answer, errors). Invalid anchors skip. Extract numbers through largest visited anchor once; gate literals against their current anchor prefix plus IMPLIED, 0, 1 and reciprocal constants. Evaluate exact expressions, skipping on failure. set overwrites; arithmetic requires existing target; move debits and credits only with existing source; ask overwrites answer. check adds parse and gold errors.
invariant
Sentence-prefix validation applies only to numeric literals. References read the current command-order state and carry no sentence provenance. Thus @2 set a.x=19 followed by @1 set b.x=a.x and @1 ask b.x remains green when 19 occurs only in sentence 2. This is a documented current limitation, not complete temporal provenance or a proposed behavior change.
Exact rational state; ordered final keys. Green execution requires an ask and no diagnostics but does not prove semantic grounding. Same-key transfer is identity; distinct-key transfer conserves sum. Negative quantities are allowed.
cost
L script chars, E expression nodes, C commands, N extracted numbers, H state keys, B evolving rational bit length: parsing O(L), literal gate O(E*N) rational comparisons, evaluation sum of rational operation costs at B. Prefix endpoints use O(Smax) integers; final state O(H log H). Recursion and allocations are not bounded.
composition
Parsing feeds ordered execution; literal collection precedes evaluation against state, and check merges parse errors, execution diagnostics and gold inequality. Text extraction is an external exact dependency. Numeric-origin errors do not suppress effects, unlike evaluation and anchor errors; a parent must preserve this distinction. Dataset/teacher adapters do not establish script semantics.
epoch-memory
Sentence-prefix validation applies only to numeric literals. References read the current command-order state and carry no sentence provenance. Thus @2 set a.x=19 followed by @1 set b.x=a.x and @1 ask b.x remains green when 19 occurs only in sentence 2. This is a documented current limitation, not complete temporal provenance or a proposed behavior change.
2026-10-30: regressions first demonstrated backward-anchor number leakage and destination credit from missing transfer sources. Prefix endpoints now restrict each command to its own sentence prefix while preserving incremental extraction; missing-source transfer skips effects. Diagnostic-only literal errors remain unchanged. No corpus or teacher calls.
checks
See tests.md and native quant::wave8_models tests. Models are source-derived partial relations, with finite evidence rather than general proof. Syntax projection excludes macro expansion, name resolution and inferred types. No executable approximant: exact gates and categorical choices have no justified percentage metric here.
behavior-complement
Universe: parsed-command/sentence/result triples with fixed math dependencies and sufficient resources. Inner absent relation excludes green outputs with invalid anchors, undefined references, zero division or foreign literals; missing-source transfers cannot credit destinations. Malformed manually constructed Cmd and resource failure are outside this domain.
reverse-reconstruction
For fixed sentences and observation O, retain every command sequence whose fold gives O. Answer-only fibers erase prior asks and overwritten state: set a; add b; ask yields 4 for all five a,b in 0..4 with a+b=4. Bounded fiber enumeration is complete only on that family.
precedence
Multiply/divide bind tighter than add/subtract, each level left-associative. Unary minus is unsupported despite negative rational results. The 648-expression oracle independently rearranges the numerator.
origin-prefix
Sentence-prefix validation applies only to numeric literals. References read the current command-order state and carry no sentence provenance. Thus @2 set a.x=19 followed by @1 set b.x=a.x and @1 ask b.x remains green when 19 occurs only in sentence 2. This is a documented current limitation, not complete temporal provenance or a proposed behavior change.
At anchor j, accepted literal set is union of extracted numbers in sentences 1..j plus fixed allowances. A previous command at later anchor must not widen membership. Regression visits anchors 2,1,2 and detects one illegal 19 at anchor 1.
diagnostic-effects
Foreign literal adds an error but can still modify state/answer. Invalid anchor or evaluation failure skips effects. check rejects any diagnostics, so an errorful run.state is not accepted knowledge.
transfer-fiber
For distinct existing keys, fixed amount d maps (x,y) to (x-d,y+d), with inverse (x'+d,y'-d). Missing destination uses zero but its prior absence is lost; alias keys preserve x for every d. Sum alone leaves infinitely many rational antecedents.
overwrite-loss
set erases prior target value/existence; ask erases prior answer. Converse retains all erased values compatible with earlier commands. Final state cannot identify overwritten command count or ordering.
sentence-boundary
Split after .?! only before whitespace or end; $2.50 stays intact; trim each segment and keep unfinished tail. Whitespace information is lost. Four-segment fixture checks decimal and punctuation boundaries.
reply-blocks
Trimmed === starts a block; preamble and code-fence lines are dropped; last duplicate ID wins. HashMap iteration supplies no ordering contract. Reverse reply fiber retains overwritten blocks and ignored text.
external-budget
vmm_run checks cap before calls but retries and concurrent in-flight calls can exceed it. batch=0 panics, par=0 processes nothing. These are source observations only; no strict spending guarantee or live-call evidence.
tests
Bounded evidence
Initial measurement epoch
2026-10-30. Five native tests in quant::wave8_models; both files total ten. Four focused runs (twice debug, twice release) pass all ten; sorted test execution uses --test-threads=1. Normalized named-result lines match byte-for-byte, SHA-256 dba383da848a3393a0766618919041d307b687a2a8b88d65b833bd9f77aa633d. Raw durations are excluded from identity comparison.
Run with MOVA_NOMEANS=1 MOVA_NOKINDS=1 MOVA_NOCLASSES=1 and cargo test -p world --lib wave8_models -- --test-threads=1; add --release for release. Synthetic mentions and prepared learner state avoid annotation, corpus, teacher and live state. The two existing quant::tests also pass.
Coverage: 648 rational expression evaluations against a rearranged independent formula; six malformed expressions and zero division; 162 transfer/alias cases over initial and amount 0..8; backward anchors 2,1,2; missing-source destination preservation; complete 25-pair answer fiber with five antecedents; decimal sentence boundaries and duplicate reply IDs.
Pre-fix native regressions failed: backward-anchor diagnostic count 0 instead of 1 and destination 9 instead of 7. Compiled production controls restored global seen membership and removed the missing-source continue; the respective named tests each failed, then source was restored.
Limits: finite CPU evidence does not prove all grammar/feature behavior, learner overflow safety, lexical correctness, persistent storage, teacher transport, budget adherence or full task accuracy. No speedup or numerical surrogate claim. Partial syntax projections repeat identically; structural checks validate bindings and references, not semantic correctness.
Artifact identities
Source SHA-256: 3c192932d6786041d0d57962ec6736240e026985e99814429f99cabeffe16793.
Compiled prefix control raw-log SHA-256: ddba0f2bf24a1aebd5d15431fc265808b67e51abf31c5d6915727dfc8a142ad0.
Compiled transfer control raw-log SHA-256: 8af2d3983e0effd0afacb23dccd40c3c8bedca1938c3a59c43d8da11489e0536.
Structural validation: an isolated Git-visible copy of this file and its complete sidecar tree passed mova-modelcheck groupants twice with identical reports and no diagnostics. Each root declares ten typed views and three fragments with four views each (22 views per file). Whole-repository registry and ancestor integration are delegated to the parent task; isolated success does not validate those inter-file bindings. Restored production sources passed all ten native tests after compiled controls.
Reference provenance scope correction
2026-10-30 follow-up: backward_anchor_references_use_command_order_state documents that references retain command-order state across decreasing anchors. Only literal membership is sentence-local. No production behavior changed in this follow-up. Six quant::wave8_models tests pass twice debug and twice release; normalized names/results match. Current source SHA-256: dd5cf73c5e8092c17566b87e0f6f465c3cdec5f1d87ac5a02ddc04c092987dbd. Earlier source hash and ten-test measurements above belong to the initial epoch.
Normalized six-test result SHA-256: 8daa317331c966e1080dd615414044901c9ac32240c8b5c093fcce936c43ddf8.
Follow-up source projection repeated byte-for-byte. Refreshed root and all three fragment hashes; existing fragment ranges remain valid because the new test follows their scopes. Isolated groupants reports repeat valid with no diagnostics; changed reports only world/src/quant.rs and no missing registrations. Full ancestor integration remains the parent task.
This page in Markdown → https://md.groupants.elamur.ai/free/en/mova/world/src/quant.rs.groupant/