world-lang-validation

The groupant beside world/src/lang.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	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
`line_admission`	failure_partition	partial source derived model	line_admission
`anchor_domain`	admission_constraint	partial source derived model	anchor_domain
`error_trace`	ordered_failure	partial source derived model	error_trace

Model text

descriptant

world-lang-validation

Forward

parse_line returns None only on trimmed empty input. It scans trailing ^s tokens right-to-left before validating remaining anchors, Args, command, unused arguments and finally required anchor. Cursor sent supplies a default only without explicit anchors. Error stores untrimmed raw and supplied number, success stores trimmed raw. Tail tokens starting ^s parse as u16, so zero and repeated values are admitted, and integer parsing also permits accepted standard u16 spellings. Explicit anchors override cursor sent without equality checking. know's from is a separate stricter positive check. Failure order is token quotes, tail-anchor parse from rightmost, misplaced anchor, no command, Args scanning, command-specific field checks, leftover arguments, missing-anchor. Oracle uses lines with multiple defects to distinguish these phases; no generic lexical-versus-semantic priority is assumed.

Scope: Source bytes [20624,21873): validation; the operation-specific contracts below define observation projections.

Invariant

The parser is pure and never executes a command. IDs are syntax-checked and registry-filtered; their existence is not checked. Cmd projection may lose spelling, while Line retains trimmed raw and supplied no; FormatErr retains original raw. At most one success or error is produced per nonempty input line. Rejected lines do not suppress later parse_all lines.

Composition

The root composes interfaces, tokenization/Args, scalar/collection validators, v1/v2 dispatch, narrative dispatch and line admission. Output of each stage is the exact next-stage input: quotes and consumed-key state cannot be independently reconstructed without matching the same original line. parse_line orders stages as trim, tokens, reverse-tail anchors, misplaced-anchor check, Args, command, unused arguments, fallback anchor. parse_all preserves input positions while partitioning successes/errors and skipping blank lines. Six scoped fragments bind concrete source ranges; the remaining parse_all and test regions are modeled at root. This hierarchy does not include executor or query semantics.

Cost

For total input bytes N and line count L, a conservative whole-file time bound is O(N^2+L) under fixed closed vocabularies: duplicate-key/list scans and vector removals can be quadratic. Tokenization itself is O(N); accumulated output and peak retained parse storage are O(N+L) excluding caller-owned input and allocator overhead. No runtime speedup or finite error budget is claimed.

Limits

Partial syn-2 projection binds the whole file; macros remain opaque and no name resolution, type inference, semantic AST or formal proof is supplied. types::* supplies Id parsing, Closed catalogs, field domains, value normalization and text limits; prag::schema::Act supplies speech acts. No state existence, reference resolution, world mutation, temporal consistency, truth of claims or query behavior is certified. No live commands, annotator, environment mutations, filesystem or external services are used in native tests. No useful approximate semantics or faster Rust surrogate is declared for these categorical gates.

Epoch memory

2026-10-30 prospective model: relation endpoints must be characters, snapshot-looking C1@1 must fail ID parsing, repeated keys fail before command dispatch, quoted text groups whitespace, and explicit placement is exclusive. Complete bounded relation fibers use 784 strings and independent character-only expected antecedents. Current source admits ^s0 and duplicate anchors, allows distinct alias keys at/to to map to repeated Field::At values, and accepts numeric quantities through six digits (999999), despite num's 0-100000 comment. These are recorded behavior/policy gaps, not silently changed. An initial fixture used unsupported act=inform; corrected to the actual Act catalog's assert before final evidence. No production behavior was changed. Native measurements and the compiled endpoint mutant are recorded in evidence.json.

catalog

world-lang-validation

Domain: Finite UTF-8 change-language lines with fixed types::* closed sets and prag::schema::Act; finite line numbers. Observation is Option<Line> or FormatErr. Inverse syntax fibers project to Cmd plus anchors and intentionally omit raw/no; full Line observations retain trimmed spelling and supplied line number. No world state or I/O is an input.

Scope: Source bytes [20624,21873): validation; the operation-specific contracts below define observation projections.

Let U be admitted inputs paired with typed outputs for the named operation, R its fixed-source relation. Behavior complements below are candidate inner bounds of U minus R; source inspection and bounded witnesses do not prove global completeness. Reverse models retain all compatible antecedents in their declared domain, not one preferred spelling. No inverse recovers discarded whitespace unless raw text is observed.

nonbehavior

Nonempty successfully tokenized input with only anchors cannot yield a command. A valid non-cursor command without anchors is rejected only after argument/command checks. Any remaining token starting '^' is a misplaced-anchor error.