Audit contextual Workshop semantics and strengthen executable conformance

[Teakowa]

Goal

Audit contextual Workshop semantics that are not fully represented by nominal parameter types or one-per-feature witnesses, and graduate every material finding into the smallest durable owner-side contract: executable tests, implementation fixes, or concise language documentation where needed.

Context

#87 established exhaustive per-feature Workshop conformance from the externally audited native surface. That suite proves that supported leaves have attributable valid cases and checks signatures, domains, rejection behavior, emission, round-trip semantics, and locale behavior where applicable.

A remaining gap is semantic distinctions within a feature or parameter context. Workshop permits context-dependent behavior such as numeric/boolean interchangeability in some positions, while similar-looking contexts can behave differently. For example, Wait(True, Ignore Condition) is accepted even though the duration is numerically interpreted, and OverPy explicitly models 0 -> False / 1 -> True replacements for selected argument positions. These facts must not be generalized into a global Boolean == Number rule.

Trusted sources for this audit include upstream OverPy documentation/source/tests, OSTW/DeltinScript behavior and documentation, and established Workshop documentation/Wiki material. Live-client validation is not required when those sources already establish the semantic contract; it remains useful only for conflicting, seasonal, undocumented, or otherwise unresolved behavior.

Intermediate extraction, comparison, and generated audit data should remain local/ephemeral. The repository should not gain another long-lived evidence database, compatibility harness, or generated semantic inventory solely to support this work.

Scope

• Audit trusted external sources for contextual Workshop semantics that can affect parsing, validation, canonical WIR, semantic comparison, emission, static analysis, or safe source transformation.
• Include at least:
  • numeric/boolean contextual acceptance and substitution behavior such as 0/False and 1/True;
  • null and other context-specific replacement/equivalence behavior where upstream sources establish it;
  • operator/comparison truthiness and coercion distinctions;
  • action/value parameter positions whose accepted behavior is wider or narrower than a nominal type label suggests;
  • context-specific exceptions such as Wait Until behavior where ordinary truthiness assumptions are not sufficient.
• Compare audited findings against the current catalog/signatures, WIR representation, validation, semantic equivalence, and existing conformance/regression tests.
• Use real, structurally valid Workshop programs as witnesses. An Action witness must execute in a valid rule action context; a Value/operator witness must appear in an appropriate semantic position.
• Cover materially distinct accepted/rejected semantic branches rather than selecting one representative value for a union-like or context-sensitive parameter and treating the whole contract as covered.
• Reuse shared structural invariants where the same rule applies across many features; add feature/context-specific cases where behavior materially differs.
• For every material uncovered distinction, add the smallest durable executable test and fix the canonical workshop-rs behavior when the implementation disagrees with the established contract.
• Keep audit extraction/generation tooling local unless a concrete independent long-term use justifies committing it.

Non-goals

• No new checked-in evidence database, semantic-facts inventory, compatibility subsystem, or generic conformance framework.
• No requirement to preserve temporary agent audit output or generated comparison artifacts in Git.
• No live Overwatch client requirement for behavior already established by trusted upstream/documented sources.
• No game-runtime simulator.
• No global Boolean/Number, Null/Number, or other coercion hierarchy unless the audited Workshop behavior actually establishes such a global rule.
• No duplication of OPY or DEL/OSTW source-language semantics in workshop-rs.
• No requirement for byte/text identity where observable semantics are preserved.
• No unrelated catalog, parser, WIR, or test cleanup.

Acceptance criteria

• The selected trusted sources are audited for contextual coercion/equivalence behavior relevant to the declared Workshop surface.
• Every material semantic distinction discovered by the audit is either already protected by an appropriate executable test or results in a focused new test before this issue closes.
• Tests use valid Workshop structural contexts and assert the applicable semantic contract rather than only checking that a spelling/catalog entry exists.
• Materially distinct accepted domains or contextual alternatives receive distinct coverage; one sampled argument does not stand in for all known branches.
• Shared category/structural invariants are not mechanically duplicated per feature unless there is an independent failure mode.
• Known context-specific exceptions remain distinguishable and are not erased by an over-broad normalization/coercion rule.
• Any implementation changes remain in the canonical Workshop owner boundary and are justified by the audited semantic contract.
• No new long-lived evidence/compatibility data subsystem is introduced; temporary audit data remains ephemeral.
• Existing Add exhaustive category and per-feature Workshop semantic conformance tests #87 per-feature conformance and corpus/round-trip coverage remain intact and are strengthened rather than replaced.

Dependencies / relationships

• Parent conformance/evidence track: Establish a tests-first Workshop corpus and seasonal feature census #10.
• Extends the completed per-feature conformance work in Add exhaustive category and per-feature Workshop semantic conformance tests #87; do not reopen or redefine Add exhaustive category and per-feature Workshop semantic conformance tests #87.
• Canonical Workshop semantics remain owned by workshop-rs.
• Downstream opy-rs and deltin-rs may consume the resulting canonical behavior for semantic comparison, but source-language conformance remains in those repositories.