📐world-groupant-wave14

The groupant beside world/tests/groupant_wave14.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

Components

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.

Model text

model

Wave14 native joins

forward

Five supplied-input joins preserve Item/source/class identity, metric boundaries, frame/Hebrew participant rendering and pure knob/candidate identity.

invariant

Aligned dictionary/class coordinates, retained source/predecessor keys, supplied dependency/annotation identity and unchanged ordered finite candidate lists.

cost

Bounded item encoding/tiny fit, row/class metric accounting, supplied rendering and finite knob enumeration; acquisition, adoption and persistence costs excluded.

composition

Each named test is a separate tagged relation. Composition requires identical graph vertices, byte/range coordinates, dictionaries, participant identifiers, schedules and residues on both sides. Distinct tests do not constitute one end-to-end pipeline. Independent finite or literal oracles remain distinct from cross-implementation equality.

epoch-memory

2026-10-30 prospective design retained below; no speed or accuracy gain is claimed.

Wave14 prospective joins

2026-10-30 base245248598c0bbe0522e7c5ac87c0aca10232aaf8. Ten exact original algorithms; four existing Mova source trees; bounded native supplied-input tests only.

Algorithm couplings: cover/matching cardinality in the same unit-capacity network; common preference priority inverted by stable marriage and counted by inversions; primality -> canonical modular root -> modular square; Shift-And/KMP byte matches; minimum cyclic rotation -> BWT row0; Cartesian parents -> LCA -> leftmost range minimum; sorting/reversal inversion identities; normalized subset zeta/Mobius identity; zeta pointwise multiplication -> OR convolution. OR must not become XOR. Shared coordinates and modulo normalization are premises. Every relation has a scoped complement and ambiguous converse; finite agreement is not general proof.

Mova couplings follow supplied Item/source/predecessor/evaluation identities, valid class/metric boundaries, supplied frame rendering and pure domain-knob/tuning helpers. Callback sets used for adoption are tuning validation, not untouched final tests. No live journal, parser, teacher, corpus or GPU execution. Costs compose only reached APIs and bounded oracles; no throughput or accuracy gain predicted. Original algorithms remain exact, no approximation substitution.

checks

5 named tests pass twice in debug and twice in release. The compiled adapter control fails its named assertion and exact source bytes are restored. Partial syntax repeats identically; evidence.md records native and structural receipts. Macro bodies remain opaque; semantic AST and whole-system proof are absent.

catalog

Wave14 supplied-input relations

Source-derived specifications; bounded native evidence and repeated partial syntax are recorded in evidence.md. No general proof or complete inverse engine is claimed.

behavior-complement

Type: behavior_complement. Named check: all five Mova checks.

Universe: supplied Item construction/classification, closed-class metrics, supplied frame/Hebrew rendering and pure tuning observations. Exclude changed source keys, batches or predecessors after JSON transport, wrong constant-field predictions, unequal-length/out-of-range metric success, lost supplied participants, CRLF-only table changes, nonfinite/multicoordinate neighbors within the shipped finite domains::knobs fixture and inadmissible selection. Invalid metric calls may panic as declared. These are supplied-input relations, not unseen prose, journal effects or retained-learning claims.

reverse-reconstruction

Type: reverse_reconstruction. Named check: all five Mova checks.

Retain compatible rows, dictionaries, fitted states and schedules sharing observed Item identities. Constant classes do not recover text or training history. Accuracy/F1 leave many confusion matrices and label sequences. Markdown loses unrendered frame fields and graph details; Hebrew tables cannot recover original line endings. A winner index identifies a candidate only with the same retained ordered candidate list; scalar scores lose configurations and callback history. Passing finite tests do not uniquely determine behavior elsewhere.

item-classification

Type: source_batch_class_identity. Named check: batched_items_keep_source_identity_through_classification.

make_items selects s1,s2,tatoeba:a,tatoeba:b with batches 0,0,1,2 and context only on the first selected tale sentence. JSON roundtrip preserves Item fields; prev_of gives [None,Some(0),None,None], preserving source-qualified predecessor identity. Built slot dictionary and constant-label IgTree fit/predict preserve all six class coordinates and Act scores=(1,1). Exclude changed keys, source leakage and coordinate substitution. Labels admit many texts and fitted histories; this tiny fit tests transport rather than held-out accuracy.

closed-metrics

Type: present_class_metric_boundary. Named check: closed_class_metrics_reject_invalid_observations.

Gold [Request,Assert] and prediction [Request,Request] yield accuracy 1/2 and macro F1 1/3 over present classes. Unequal lengths and class IDs at the class count panic. Empty vectors with zero classes yield (0,0). Exclude zip truncation and silently ignored invalid classes. Equal metrics retain row permutation and confusion-matrix ambiguity. Cost is row/class bookkeeping; no semantic label accuracy follows.

frame-rendering

Type: supplied_participant_projection. Named check: supplied_frame_identity_survives_markdown_projection.

The supplied Ada/carries/box tree yields one frame with agent id ada and patient id box. Markdown contains carry/Ada/box in one event row; roles preserves (1,2,agent); empty input states no event. Exclude exchanged participants and duplicated rows. Selected lexical/role observations lose other graph/state details and retain many antecedents. Only supplied Word trees are used; no parser acquisition.

hebrew-rendering

Type: macula_identity_projection. Named check: hebrew_annotation_rows_keep_macula_participant_identity.

For two supplied annotation rows, hwords retains participant head 1 and MId identity referenced by Frame=A0:2. Hebrew output includes v/did work, x/worker and finite/qal. LF-to-CRLF conversion leaves the table equal. Exclude missing referenced participants and newline-dependent labels. The table loses line endings, unused columns and gloss punctuation. This fixture does not establish external corpus coverage.

tuning-identity

Type: neighbor_and_selector_coordinate. Named check: domain_knobs_and_tuning_selection_preserve_candidate_identity.

From domains::knobs and its finite base map, each neighboring_configs candidate retains map keys, changes exactly one value and has only finite values. select_candidate on the supplied observations returns index 0, referring to that same candidate vector; separate inadmissible observations yield None. Exclude index substitution, multiknob mutation and inadmissible selection. An index without the list cannot recover configuration, scores or callback history. Adoption callbacks are tuning validation, not untouched final assessment; no adoption executes.

evidence

Wave14 bounded evidence, 2026-10-30

5 new native joins. Source SHA-256 48dcfb18de805d9999e51e9a4b753d6f5d8b03dd427c31aebbe4bd128ea66d68; twice-identical partial syntax SHA-256 8dc2a2707528f5028cd3b01d054f5ccbba3ddeb012b401c0501da9bf3035e5d5.

464 algorithm tests (416 unit, 47 external, one doc) and 101 selected Mova checks pass twice in debug and twice in release with identical normalized named outcomes. Mova checks comprise the previous 75 and 26 reached checks: eval 4, data 6, frames_md 7, tune 4 and five new joins. These selected Mova checks do not represent the complete Mova suite.

Ten new algorithm production mutations and eight false-oracle mutations fail named assertions in owner receipts. Four Mova production mutations fail named assertions. Both root adapters compile and fail their intended named assertions: zeta_product_inverts_to_or_convolution (wrong inverse stage) and batched_items_keep_source_identity_through_classification (wrong source window). Exact source bytes are restored before positive runs. Compilation failures do not count as semantic rejection.

Strict all-target algorithms Clippy and algorithms formatting pass. Historical world/prag warnings remain. No benchmark, throughput, GPU, teacher/parser acquisition, full-school run, held-out accuracy, persistent adoption or general proof is claimed. Tiny supplied-input fitting exercises actual classifier calls. Tuning validation participates in candidate choice and is not an untouched final test.

Raw receipts: $MOVA_DATA/runs/octopus-wave14-integration-2026-10-30. native-evidence-hashes.json SHA-256 d95a91417adeb7f8e755d26f900b6508ea17759291407f33153917a42396280f; algorithm-adapter-mutant.log d2279a28cc474377b1d9321e15d4c779b0af8507f81ac39e83338bc9084a02a2; mova-adapter-mutant.log 60567fa4d984ac58834be268cdb221a37caf30bd5d07e9a3d6a2d89c16fdac87; mova-all-normalized.txt fb492ea8097fd74fe50cdceb5d61cb49efc74baa6cb152f7699e546dcbb3ac20; structural-controls/summary.json a50d4387189f34044579c0d068475201acf1f6e7ee8e3297c0db6909f3180975. The structural controls reject stale parent-source bindings and wrong child identities, restoring exact bytes. Source syntax remains partial, with opaque macros and no semantic AST.

This page in Markdown → https://md.groupants.elamur.ai/free/en/mova/world/tests/groupant_wave14.rs.groupant/