Elamur Groupantsworld-absurd-knowledge-votes
The groupant beside world/src/absurd.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-absurd-knowledge-votes-behavior-complement | behavior_complement | partial source model | behavior-complement |
world-absurd-knowledge-votes-reverse-reconstruction | reverse_reconstruction | partial source model | reverse-reconstruction |
world-absurd-knowledge-votes-coupling | constraint_relation | partial source model | coupling |
Model text
forward
learn_cells filters lowercase last subject and first object tokens longer than two bytes on Can/Does facts. Negative unconditional facts vote subject 4, negative unconditioned-object facts vote subject 3, negative conditionals abstain. Positive facts vote subject 0 or Rarely 1 and final nonconditional object 0. Highest count wins with lower-score tie break; each score retains its first source.
invariant
Negated facts never generate object votes. learn_args omits a whole cell when either role has a rated counterpart. Fact provenance and lexical heuristics do not establish physical truth. This fragment is source inspected, not natively exercised.
cost
F facts and V vote keys: O(F log V) map operations plus token work; O(V) retained vote storage. Knowledge loading and writing the layer file are external.
composition
Actual callers: domains_with projects Word to Node then consumes judge indices without reordering; rerank::items sums judgment penalties per candidate, and induce calls judge on supplied vectors. Only the domains join executes in these native tests.
epoch-memory
2026-10-30 initial source-bound epoch. Old native witnesses exposed role-index panic and spurious formal activation with empty overrides. Guards/default resolution repair these; explicit finite values and sparse-cell gap semantics remain. No optimization, surrogate or percentage bound is claimed.
checks
cargo test -p world --lib absurd::wave11_models -- --test-threads=1. Root evidence.md records repeated profiles and mutants. Partial syntax leaves macro bodies opaque; no semantic AST, parser, teacher or corpus execution.
behavior-complement
Universe: terminating scoped calls and their outputs. Inner absent-behavior model excludes outputs contradicting: Negated facts never generate object votes. learn_args omits a whole cell when either role has a rated counterpart. Fact provenance and lexical heuristics do not establish physical truth. This fragment is source inspected, not natively exercised. Unknown external behavior remains outside this exclusion.
reverse-reconstruction
A winning cell retains all vote multisets whose maximal count and lower-score tie rule choose that score. Losing counts and alternate sources are lost; winning count is not total evidence count.
coupling
Source-order counterfactual: permuting equal-score votes may retain score/count while changing the example source. Multiple fact sets have identical outputs; source examples are not unique causal explanations.
All typed entries bind this scope and manifest source revision. Domain and observation/operator are stated above. Status: partial source-derived relations with finite evidence named in checks, assuming stable dependencies and bounded successful execution. Reverse relations preserve ambiguity, with no complete-search or causal claim.
This page in Markdown → https://md.groupants.elamur.ai/free/en/mova/world/src/absurd.rs.groupant/knowledge-votes.groupant/