fix(parser): swallow audit — typed, KIND-SPECIFIC duration evidence (CR 611.2a)

[matthewevans]

Duration is a BOUNDED fact — its parse-time carriers are a small closed set —
so it is hand-typed over an enumerated carrier list rather than probed. Three
vacuities died with the old code:

1. static_has_duration(s) { let _ = s; true } — returned true for ANY static
   ability whatsoever (a keyword grant, an anthem, a cost modifier).
   StaticDefinition has no Duration field at all; a static's duration lives on
   the Effect::GenericEffect that wraps it. The stub was not a conservative
   approximation of a duration — it was unrelated to one.

2. any_ability_has_duration (its only caller, and the sole evidence gate on
   detect_duration_until_eot) accepted a duration of ANY KIND, so a Permanent
   grant discharged an "until end of turn" expectation.

3. detect_duration_next_turn's markers were ["YourNextTurn", "NextTurn", "UntilYourNextTurn"]. NextTurn is a MEASURED SUBSTRING COLLISION. Counting
   [A-Za-z]*NextTurn[A-Za-z]* over the full 35,396-face export:

   UntilNextTurnOf 216 | UntilEndOfNextTurnOf 182   <- real durations
   SkipNextTurn     10 | ControlNextTurn        8   <- NOT durations
   

   So an "until your next turn" expectation was dischargeable by
   Effect::SkipNextTurn / Effect::ControlNextTurn — facts about extra/skipped
   turns, not about how long a continuous effect lasts. Same defect as APNAP's
   player_scope, arriving by substring instead of by wrong field.

Evidence is now the kind the expectation asked about: unit_has_duration(parsed, &duration_is_end_of_turn) / &duration_is_next_turn. The walk reaches every
container (Token / GenericEffect statics, GrantTrigger, CreateEmblem, FlipCoin,
ChooseOneOf, CreateDelayedTrigger, Vote, SeparateIntoPiles, RevealFromHand,
AddTargetReplacement), which is what the json_has_any fallback existed for —
the old structured walk could not descend into Effect::Token.

THE CARRIER LIST IS THE CORRECTNESS ARGUMENT, so the population is pinned:

rg ':\s*[^,/]*\bDuration\b' crates/engine/src/types/{ability,statics,triggers,replacements,mana}.rs

10 parse-time carriers: AbilityDefinition.duration; Effect::{BecomeCopy,
GainActivatedAbilitiesOfTarget, GenericEffect, CastFromZone, ForceAttack}.duration;
Effect::PreventDamage.prevention_duration; CastingPermission::{ExileWithAltCost,
PlayFromExile}.duration; ManaSpellGrant::AddKeywordUntilEndOfTurn.duration. Plus
the duration-EQUIVALENT Effect::Mana.expiry (ManaExpiry, CR 106.4), and the
structural CR 614.6 / CR 514.2 lifetime of a one-shot ReplacementEvent::DamageDone
shield (which has no duration field to carry it).

THAT GREP IS THE SECOND ONE, and the first one's failure is the lesson. It read
:\s*(Option<)?(Box<)?Duration\b over ability/statics/triggers/replacements and
MISSED ManaSpellGrant — boxed, fully qualified, and in mana.rs. The full-pool
delta caught it as a false-positive cluster, and deleting any_ability_has_duration
wholesale also dropped its DamageDone leg, caught the same way. A carrier list is
a CLAIM ABOUT A POPULATION; state the population, then measure the claim.

MEASURED, full pool (35,396 faces; pool sha 2d168167…), vs the previous commit:

Duration_UntilEndOfTurn 9 -> 25 (+16, 15 distinct faces)
Duration_NextTurn 1 -> 2 (+1 — cone of cold)
every other detector +0
LEDGER B (losses) 0 — no warning disappeared
export minus parse_warnings 0 faces changed (byte-identical)

Iteration, stated rather than hidden: +23 -> +19 (ManaSpellGrant carrier restored)
-> +16 (DamageDone lifetime restored). Both regressions were MINE, and both were
found by the full-pool delta, not by review.

16,237 engine lib tests pass. Refs: Plan 02 / U2 commit 3 (Duration leg; the
DynamicQty / Condition_* detectors still ride ast_json).

[gemini-code-assist]

Code Review

This pull request refactors the duration-swallowing detectors (detect_duration_until_eot and detect_duration_next_turn) in swallow_check.rs to use precise, typed AST traversal instead of fragile JSON substring matching or overly broad blanket checks. It replaces the old any_ability_has_duration and static_has_duration helpers with structured checks (unit_has_duration) parameterized by specific duration predicates (duration_is_end_of_turn, duration_is_next_turn) and duration-equivalents like mana expiry. No review comments were provided, and the changes are highly idiomatic, aligning perfectly with the repository's architectural rules for strict rule fidelity and robust type-safe validation; therefore, I have no feedback to provide.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

[github-actions]

Parse changes introduced by this PR

✓ No card-parse changes detected.