Elamur Groupantsrule-derivation-packing
The groupant beside world/src/derive.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 | Forward |
| ast | partial | ../syntax.json |
| composition | partial | Composition |
| cost | symbolic unmeasured | Cost |
| descriptant | present | model.md |
| invariant | partial | Invariant |
| tests | bounded 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.
| id | type | label | where |
|---|---|---|---|
nonbehavior | behavior_complement | partial source derived model | nonbehavior |
antecedents | reverse_reconstruction | partial source derived model | antecedents |
lexical-projection | information_loss | partial source derived model | lexical-projection |
finite-projection | state_projection | partial source derived model | finite-projection |
widths | admission_constraint | partial source derived model | widths |
Model text
descriptant
rule-derivation-packing
Partial source-derived groupant, 2026-10-29. Full source inspected; this node specializes its explicit syntax envelope. Macro expansion and semantic AST are absent.
Forward
scan_store captures latest version, selects ordered IDs, reads raw records and interns classes in first-encounter order. It copies UPOS/relation/head bytes and computes finite flags from tags. unpack creates synthetic S with empty forms, class-lemma projection, widened heads and decoded enums.
Invariant
Equal Packed vector lengths, valid enum/class indexes and fewer than 65536 classes are assumptions. Downstream heads must be in range. Missing lemma IDs become empty strings, whereas malformed enum/class indexes can panic. Class IDs are dictionary-relative, not stable cross-run identities.
Cost
W selected tokens and C classes: O(W log C + class-string bytes), plus selection/raw lookup. Unpack O(W+lexical bytes). Numeric vectors use 6 bytes/token plus one bool byte, vector allocation overhead and class strings. Seven bytes/token describes logical element sizes, not total retained or allocated memory.
Composition
Closed-class lemma changes affect class and unpacked lemma; tag membership in VBD/VBZ/VBP/MD affects finite and potentially VERB root choice. Store identity is external. scan_store fixes hash then calls lemma_table(dir), which separately reopens latest; concurrent ref changes require compatible lexical tables. No transaction/snapshot guarantee follows.
Epoch memory
Initial epoch: 90 assertions over twelve extracted pure source items with minimal enum dependencies. Closed-class erasure mutant rejected. No full learner, parser, GPU, corpus, filesystem-training or external state execution. Source-derived models and proposed oracles are distinct from executed bounded probes. See ../tests.md.
catalog
rule-derivation-packing typed models
Subject: derive.rs SHA-256 2726211da9538f3f5c225292219dfda4d4fd6f15d6b68deaaaf53c834ba60842. Domain: finite admitted inputs with fixed dependency/schedule state. First type is a candidate inner behavior-complement bound in explicit input/output universe, second is an ambiguous compatible-preimage relation. Other models specialize observations or costs. See descriptant for assumptions and ../tests.md for evidence; structural acceptance is not semantic proof.
nonbehavior
Type: behavior_complement.
U consists of valid record/table projections and successful results. Inner absent behaviors include unpack recovering original surface forms, obtaining a nonempty lemma from a colon-free class, and finite=true for a decoded store tag outside VBD/VBZ/VBP/MD. Form/lemma cases executed; real decode_tag inspected only.
antecedents
Type: reverse_reconstruction.
Given unpacked S with fixed dictionary/enums, Packed must preserve observed enum/finite/u16-head values; each class slot may be any slot whose split(':').nth(1).unwrap_or(empty) equals the lemma. Nonempty forms or heads exceeding u16 have no preimage. Multiple slots and suffixes share observations; original forms are unrecoverable. Probe witnesses ADP:x:y versus ADP:x:z and open-class collisions.
lexical-projection
Type: information_loss.
Closed ADP/AUX/PART/SCONJ/CCONJ/DET/PRON map to POS:lowercase(lemma), others to POS. unpack retains only colon component one, so ADP:x:y yields x: class/unpack is not an identity for colon-containing lemmas. Missing lexical IDs yield empty suffix. Probe covers lowercase/open-class/colon collapse and to/of propagation; no full Unicode normalization claim.
finite-projection
Type: state_projection.
scan_store finite is decoded-tag membership in VBD/VBZ/VBP/MD; parsed scan instead checks feature text for VerbForm=Fin. These are distinct dependency predicates. Finite=true admits four decoded tags, false all others; root preference further requires VERB POS. Proposed exhaustive real Tag/decode oracle not executed by stub probe; store sibling fixtures are separate evidence.
widths
Type: admission_constraint.
New class gets classes_m.len() as u16. After 65536 classes, the next index wraps to zero and dictionary inversion overwrites a slot: injectivity needs the capacity assumption. Raw head little-endian u16 widens without validation; probe retains 65535. Book scan narrows parser heads to u16. Width recovery needs range constraints. Capacity-boundary allocation oracle remains prospective.
This page in Markdown → https://md.groupants.elamur.ai/free/en/mova/world/src/derive.rs.groupant/packing.groupant/