Elamur GroupantsโŒ•
โ˜ฐ

world-store-coupling

The groupant beside world/src/store.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

componentlabelwhere
approximantpartialForward
astpartial../../syntax.json
costsymbolic unmeasuredCost
descriptantpresentmodel.md
invariantpartialInvariant
testsbounded evidence../../model.md#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.

idtypelabelwhere
nonbehaviorbehavior_complementpartial source derived modelnonbehavior
antecedentsreverse_reconstructionpartial source derived modelantecedents
positional-couplingdependency_closurepartial source derived modelpositional-coupling
rank-boundarycounterfactualpartial source derived modelrank-boundary

Model text

descriptant

coupling

Forward

Sampling advances a seed-derived xorshift stream once per ordered candidate, combines the state with ID salt, sorts key/ID pairs, takes k and sorts retained IDs. Filtering first changes stream positions and competitors.

Invariant

Sample output is sorted, unique and a subset of duplicate-free input candidates, of length min(k,n). Changed eligibility can change unchanged output membership. Reverse fibers retain candidate ordering and losing candidates.

Cost

O(n) pair allocation and O(n log n) sorting. This is the same sampling work modeled by the parent and must not be added again.

Epoch memory

2026-10-29: The removed-ID witness disproved that exact eligibility deltas suffice for sampled carry. Fixed-seed determinism does not imply candidate independence.

catalog

world-store-coupling

Source SHA-256 6801677c4f17273e615090a790f2e4018387ca5d006aab899bafda09f01b24d9. Conditional models over successful finite operations; I/O failure, malformed offsets and changing dependency state are excluded unless stated. Algebra is exact only for its declared abstract relation. Source correspondence remains bounded.

nonbehavior

Type: behavior_complement.

Universe: ordered duplicate-free candidate vectors, valid k and seed, and output vectors. Candidate inner complement includes sampled outputs longer than min(k,n), unsorted outputs, duplicate outputs, and IDs absent from candidates. This is a conditional source model of the sample stage, not the complement of the full query.

antecedents

Type: reverse_reconstruction.

Fix k and seed. A sampled observation Y has every ordered candidate vector C whose first min(k,|C|) pairs in sorted(key(C position,seed) xor id salt,id) have numerically sorted selected IDs equal to Y as compatible antecedents. Different candidate orders and losing candidates remain possible. No unique inverse; k=0 maps every candidate vector to empty.

positional-coupling

Type: dependency_closure.

The xorshift stream advances per candidate position. Removing one eligible ID changes ranks of later unchanged IDs as well as competition for k slots. Changed-ID dictionaries therefore do not bound output membership changes. Historical witness eligible [0,1,2] to [0,2], C={1}, pred>=1 sample=1/2 gives old [1], fresh [2], legacy merge []; repaired load must return [2].

rank-boundary

Type: counterfactual.

Hold query order and seed fixed, replace legacy delta-merge by complete target selection for stale sampled queries. Output equals F even if O\C differs from F\C. This intervention fixes candidate coupling but does not validate feature extraction, mutable same-hash indexes or forged stable stamps. Oracle compares repaired witness and rejects historical empty result.

This page in Markdown โ†’ https://md.groupants.elamur.ai/free/en/mova/world/src/store.rs.groupant/selection.groupant/coupling.groupant/