---
title: "prag-rules-atoms — Mova code model"
description: "Mova code model of prag/src/rules.rs: components with their labels, obernants and the model text."
url: "https://groupants.elamur.ai/free/en/mova/prag/src/rules.rs.groupant/atoms.groupant/"
---
Elamur: [Mova](https://md.mova.elamur.ai/free/en/) · [Dictionary](https://md.dictionary.elamur.ai/free/en/) · [Research](https://md.research.elamur.ai/free/en/) · **Groupants**

This page as HTML: <https://groupants.elamur.ai/free/en/mova/prag/src/rules.rs.groupant/atoms.groupant/>

# prag-rules-atoms

The groupant beside prag/src/rules.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

- **kind:** groupant_fragment
- **status:** partial source bound model
- **source:** [`prag/src/rules.rs`](https://github.com/ElamurAI/mova-free/blob/main/prag/src/rules.rs)
- **source SHA-256:** `ba82334c7c91aaaf…`
- **covers:** frame_atoms projects kind, role presence, flags, links, previous kind, provenance and optional tree/lexicon fields. Pronoun atoms depend on proposed Resolve candidate. PrevAgent searches earlier frames backward; LineAgent uses frame0 for fi>0; Speaker uses nearest Say agent then before_speaker; Before uses number-compatible salience; Unresolve normalizes pron ID. (bytes 28565–37813)
- **part of:** [prag-rules](https://md.groupants.elamur.ai/free/en/mova/prag/src/rules.rs.groupant/)
- **repository folder:** [`prag/src/rules.rs.groupant/atoms.groupant`](https://github.com/ElamurAI/mova-free/tree/main/prag/src/rules.rs.groupant/atoms.groupant)
- **repository commit:** `855da7bd4a7b`

> **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](#forward) |
| **ast** | partial | `../syntax.json` |
| **composition** | partial source model | [composition](#composition) |
| **cost** | partial source model | [cost](#cost) |
| **descriptant** | partial source model | [epoch-memory](#epoch-memory) |
| **invariant** | partial source model | [invariant](#invariant) |
| **tests** | partial source model | [checks](#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 |
|---|---|---|---|
| `prag-rules-atoms-behavior-complement` | behavior_complement | partial source model | [behavior-complement](#behavior-complement) |
| `prag-rules-atoms-reverse-reconstruction` | reverse_reconstruction | partial source model | [reverse-reconstruction](#reverse-reconstruction) |
| `prag-rules-atoms-specialization` | candidate_identity | partial source model | [specialization](#specialization) |

## Model text


### forward
frame_atoms projects kind, role presence, flags, links, previous kind, provenance and optional tree/lexicon fields. Pronoun atoms depend on proposed Resolve candidate. PrevAgent searches earlier frames backward; LineAgent uses frame0 for fi>0; Speaker uses nearest Say agent then before_speaker; Before uses number-compatible salience; Unresolve normalizes pron ID.

### invariant
Features can have multiple values (marks/adverbs); optional absence omits atoms. Candidate action execution sees evolving state although hits use pre-rule atoms. Lexical features may acquire global caches and are outside executed fixtures.

### cost
O(F log F) frame map per atoms call plus O(F+Q+T) scans and external lookup cost. Candidate previous search O(F), Before O(before length); clones cost entity bytes.

### behavior-complement
For empty-verb/entity Event fixtures, candidate atom sets lacking kind:event/has-agent:no or containing has-agent:yes are excluded. With earlier alpha,beta and She at frame2, PrevAgent result alpha is excluded in favor of beta.

### reverse-reconstruction
Kind/presence projections lose attrs, source and exact role identities. Events differing only in attrs share tested atoms. Same candidate ID can originate from different frames; output ID does not recover provenance.

### specialization
Resolve preserves original pron surface. Same candidate ID returns no change even when other metadata differs; this existing ID-based equality is narrower than full entity equality.

### composition
Codec feeds selection without reordering. Sites and action-dependent atoms share the pre-rule Read; ordered action fold sees evolving state, then deletion remaps coordinates. Selection nested in codec retains status distinctions. Exact byte scopes constrain composition; no child alone proves the parent.

### checks
Ten rules::wave15_models tests, tiny supplied fixtures only; two debug and two release passes with identical named outcomes. File-root receipts.md retains hashes. Finite catalogs, nine transfer states and sixteen deletion subsets; compiled production vacancy-guard mutant fails finite_transfer_and_drop_relations.

### epoch-memory
2026-10-30: first rules tree. Review predicted unchanged valid-site transitions and O(1) validation before O(Read bytes) cloning. Old invalid_direct_sites_are_no_change failed because absent Drop returned Some; seven other initial tests passed. Repair rejects absent coordinates/unaligned reads. select_rules exposes the existing production filter; no selection policy change or speedup claimed.

### limitations
Partial syntax, no expanded macros or semantic AST proof. No live parser, teacher, corpus, environment mutation, override or cache acquisition in tests. Empty verbs/entities avoid lexical matching acquisition; populated entity tests use direct actions. Token creation, full lexical/ancestry/action branches, IO/cache lifetime and all multi-site interactions remain unexecuted. Public atoms/drop_frames still require valid coordinates/alignment.