📐world-nerves-scope

The groupant beside world/src/nerves.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

descriptant

Concept normalization and manual sentence extraction

Define A(s)=concatenate lowercase expansions of alphanumeric chars of s. If A is empty, in STOP/NUMBERS/ASKING, or all ASCII digits, N(s)=None. Otherwise, if byte_len(A)>4 and A ends ies, replace ies by y; else if byte_len(A)>3 and A ends s but not ss/us/is, drop final s; else retain A. content_words(t)=set(filter_map(N, split t at non-alphanumeric chars)). sentences_of trims lines, toggles fenced-code state on leading backticks, resets open at fences/blank/headings, skips code, and appends a line to the prior sentence iff open; open becomes false for final .!?»")]: punctuation.

2026-10-31 review epoch: replaced generic fragment wrappers with scoped transitions, concrete exclusions and compatible antecedents. The exact byte range/hash and shared parent AST are retained. This fragment supplies norm/content_words to the parent graph seed/build/ranking code and norm to ideas::words. It also supplies manual sentence units to graph construction. The parent three-step diffusion recurrence is outside this child, so its activation arithmetic is not copied here.

approximant

Concept normalization and manual sentence extraction

Define A(s)=concatenate lowercase expansions of alphanumeric chars of s. If A is empty, in STOP/NUMBERS/ASKING, or all ASCII digits, N(s)=None. Otherwise, if byte_len(A)>4 and A ends ies, replace ies by y; else if byte_len(A)>3 and A ends s but not ss/us/is, drop final s; else retain A. content_words(t)=set(filter_map(N, split t at non-alphanumeric chars)). sentences_of trims lines, toggles fenced-code state on leading backticks, resets open at fences/blank/headings, skips code, and appends a line to the prior sentence iff open; open becomes false for final .!?»")]: punctuation.

Categorical source relation; no executable percent surrogate or approximation error bound is claimed.

invariant

Stop filtering precedes suffix reduction. norm removes punctuation within its argument, while content_words splits punctuation before normalization: norm(cat-dog)=catdog but content_words(cat-dog) contains cat,dog. Sentence extraction discards headings and fenced code; it joins wrapped prose with one space and does not tokenize multiple sentences on a single line.

Cost: norm is O(input chars plus lowercase output bytes) with fixed cue-table scans; content_words adds ordered-set insertion O(W log U). sentences_of is linear in line/input/output bytes under amortized String growth. No input cap or fallible-allocation interface exists.

composition

This fragment supplies norm/content_words to the parent graph seed/build/ranking code and norm to ideas::words. It also supplies manual sentence units to graph construction. The parent three-step diffusion recurrence is outside this child, so its activation arithmetic is not copied here.

models

behavior_complement

Exclude Some for empty, 123, where or three. N(BERRIES!) is berry and N(glass) is glass. N(ones)=Some(one) while N(one)=None: the source does not reapply stop filtering after suffix reduction. sentences_of cannot emit the body of a properly toggled code fence, a heading line, or a standalone empty line.

reverse_reconstruction

N^-1(berry) contains berry, berries, BERRIES!, and punctuation-inserted b.e.r.r.y because norm filters before suffix handling. N^-1(None) contains empty strings, pure punctuation, ASCII digit strings and all removed cue words. The sentence output [One wrapped line.] admits one input line or two wrapped lines joined by one space, with arbitrary ignored headings/fenced examples; original wrapping is unrecoverable.

normalization_order

A(s) -> removal predicates -> suffix rule is ordered and lossy. These rules are surface morphology, not linguistic lemmatization; alphanumeric Unicode digits outside ASCII are not removed by the ASCII-number predicate alone.

token_boundary_transport

content_words deduplicates normalized tokens using BTreeSet. Repeating apples or adding case variants does not create multiple seed concepts. norm on an entire punctuation-separated phrase is not interchangeable with content_words.

manual_state_machine

State=(code,open,out). Fence toggles code and clears open; blank/heading clears open; code content is skipped. A prose line either extends last(out) or creates a new element, then trailing punctuation controls open.

terminal_punctuation

A colon or closing bracket ends wrapping, just like sentence punctuation. A line ending a comma stays open. An unmatched opening fence suppresses the remaining file; this parser does not report a fence error.

tests

nerves::wave28_dialogue::normalization_and_manual_boundaries checks BERRIES!, glass, empty/digits/where/three, wrapping, heading removal and code fencing. synthetic_spread_units_and_dedup consumes content_words through spread, but diffusion itself is outside this child scope. No claim that finite examples prove all Unicode/suffix behavior.

Commands and repeated outcomes: parent evidence, source evidence. No new execution was required for this model-only refinement.

This page in Markdown → https://md.groupants.elamur.ai/free/en/mova/world/src/nerves.rs.groupant/scope.groupant/