fix(parser): UNSUPPORTED seam: Play historic from graveyard once/turn + granted leave-battlefield re

[ntindle]

Summary

Fixes a parser misparse affecting 1 card(s) in the Doctor Who Commander precons.

Root cause: UNSUPPORTED seam: Play historic from graveyard once/turn + granted leave-battlefield replacement rider on the played object

Cards corrected

• The Eighth Doctor

Fix

Implemented the approved plan for Cluster-113 (The Eighth Doctor): a play/cast-from-graveyard permission that GRANTS a per-object leave-the-battlefield → exile replacement to the permanent played this way. Built the general carrier seam, not a Doctor-special-case. No new enum variant — two struct-field additions plus a shared install helper and a shared parser def-builder.

WHAT CHANGED (by subsystem):
Types: added granted_replacement: Option<Box<ReplacementDefinition>> as a fourth leaf rider on StaticMode::GraveyardCastPermission (serde-only, Display/FromStr default None; Hash/Display already destructure with .. so no change there — ReplacementDefinition lacks Hash, confirmed). Added transient GameState.pending_etb_replacements: Vec<(ObjectId, Box<ReplacementDefinition>)> mirroring pending_etb_counters at all four lifecycle sites (field/Default/destructure/PartialEq).
Parser: extended parse_leave_battlefield_rider_ref with a ~ self-ref alt arm; extracted shared parse_leave_battlefield_exile_replacement_def (byte-identical effect-path output preserved via a thin wrapper); replaced the honest-gap return None bail in try_parse_disjunctive_graveyard_cast_permission with a nom split_once_on(". if you do, it gains ") + delimited(char('"'), take_until, char('"')) quote-strip that feeds the shared builder (rider present-but-unparsable still aborts — honest gap preserved).
Runtime: new single-authority GameObject::install_replacement(def, to_base) (add_target_replacement.rs refactored to call it); readers permission_granted_replacement / selected_static_permission_granted_replacement (spell, CR 607.1 self-grant fallback) / graveyard_permission_granted_replacement (land) in casting.rs; spell seam in finalize_cast (casting_costs.rs) and land seam in handle_play_land (engine.rs) both push onto pending_etb_replacements; the drain+base-install lives at the top of apply_zone_delivery_tail gated on to == Battlefield — the single convergence point of the immediate deliver fall-through AND the counter-pause resume, so it fires exactly once; turn-cleanup backstop clear in turns.rs.

VERIFICATION (worktree off-Tilt, cargo run directly): cargo fmt clean; precise parser diff-gate on my added lines = CLEAN (no string dispatch); cargo clippy -p engine -p phase-ai -p engine-wasm -- -D warnings (no --all-targets per disk ceiling) = 0 warnings (covers phase-ai/engine-wasm lib exhaustiveness); cargo test -p engine = 16517 lib + 2958 integration, 0 failed; cargo test -p phase-ai = 0 failed. oracle-gen for The Eighth Doctor now emits the GraveyardCastPermission static with granted_replacement (Moved/SelfRef → Exile), the mill-3 trigger, NO Unimplemented, and parse_warnings: [].

RUNTIME PROOF (new card-test, GameScenario+GameRunner, 3 tests all green): (1) cast a historic permanent from graveyard via the permission → enters battlefield with the redirect on base_replacement_definitions, then Destroy → lands in EXILE (proves SelfRef Moved install fires and binds to the entrant, not the Doctor); (2) CR 611.2c persistence — remove the Doctor, then the grantee's exit still redirects to exile; (3) net-new land path — play a historic land from graveyard via PlayLand → rider installed on the land's base, exit → exile. Building-block parser tests: honest-gap test flipped to assert Some(rider) with correct shape; def-builder tested on both ~ and pronoun forms.

NEIGHBOR REGRESSIONS re-run green: l02_bb3_cast_permission_finality (all), finality_counter_death_to_exile (all 4 death paths), ark_of_hunger land-play, mill_rest_in_peace_redirect, living_death_replacement, statics serde/Display round-trip, casting rules integration.

JUDGEMENT CALLS: (a) followed orchestrator disk-ceiling override (NO --all-targets) rather than the plan's literal --all-targets; lib clippy still catches the non-exhaustive-match class in phase-ai/engine-wasm, which is the whole exhaustiveness risk. (b) install_replacement doc cites CR 613.1 + CR 614.1a (accurate to base-restore + replacement semantics) rather than the plan's tentative CR 611.2b, since 611.2b is specifically 'for as long as' durations; the preserved 611.2b in the add_target_replacement invariant comment remains correct for the ControllerControlsSource ('for as long as you control') case.

STOP-AND-RETURN: none — no CR ambiguity, no multi-PR seam. The adjacent reanimation-rider class (Gruesome Encore etc.) uses a different carrier (effect rider) and is explicitly out of cluster-113 scope; the extracted def-builder + install helper are the seams that unlock it later.

VERIFICATION NOT RUN (disk preservation, ~6Gi free): cargo semantic-audit, cargo coverage, and frontend pnpm type-check/lint. The targeted oracle-gen already proved this specific card parses clean (no Unimplemented, no swallow warnings); no frontend code changed and the new field is an optional serde field (backward-compatible, tsify auto-regenerated). Also note: the full ./scripts/check-parser-combinators.sh reports FAIL, but it diffs against an ancient fixed base commit (ff799f8, PR #904) and every flagged line is pre-existing code in oracle.rs/other files — the precise per-diff gate on my added lines is clean.

Files changed

• crates/engine/src/types/statics.rs
• crates/engine/src/types/game_state.rs
• crates/engine/src/parser/oracle_effect/mod.rs
• crates/engine/src/parser/oracle_static/restriction.rs
• crates/engine/src/parser/oracle_static/tests.rs
• crates/engine/src/parser/swallow_check.rs
• crates/engine/src/game/game_object.rs
• crates/engine/src/game/effects/add_target_replacement.rs
• crates/engine/src/game/casting.rs
• crates/engine/src/game/casting_costs.rs
• crates/engine/src/game/engine.rs
• crates/engine/src/game/zone_pipeline.rs
• crates/engine/src/game/turns.rs
• crates/engine/src/game/engine_tests.rs
• crates/engine/src/game/effects/create_emblem.rs
• crates/engine/tests/integration/rules/casting.rs
• crates/engine/tests/integration/main.rs
• crates/engine/tests/integration/eighth_doctor_granted_leave_battlefield_exile.rs

CR references

• CR 305.1
• CR 601.2a
• CR 603.6c
• CR 607.1
• CR 608.2c
• CR 611.2c
• CR 613.1
• CR 614.1a
• CR 614.6
• CR 614.12
• CR 700.6

Verification

• cargo fmt --all — clean (exit 0, no changes; nothing to commit)
• check-parser-combinators.sh (scoped to upstream/main merge-base 7c63d29fe) — clean (Gate A PASS, exit 0)
• cargo clippy -p engine -p phase-ai -p engine-wasm -- -D warnings — clean (exit 0)
• cargo test -p engine -p phase-ai — clean (exit 0, 22 suites ok, 0 failed)
• oracle-gen data --filter "the eighth doctor" — clean (exit 0, AST matches oracle text, no Unimplemented/Unknown)
  Cards confirmed re-parsed correctly: The Eighth Doctor

🤖 Generated with Claude Code

[gemini-code-assist]

Warning

You have reached your daily quota limit. Please wait up to 24 hours and I will start processing your requests again!

[github-actions]

Parse changes introduced by this PR · 1 card(s), 2 signature(s) (baseline: main 6d7821dced96)

1 card(s) · static/GraveyardCastPermission(Play,once_per_turn) · added: GraveyardCastPermission(Play,once_per_turn) (affects=historic land or historic permanent)

Examples: The Eighth Doctor

1 card(s) · ability/static_structure · removed: static_structure

Examples: The Eighth Doctor

[matthewevans]

Changes requested — the new leave-battlefield rider is recovered from its source rather than from the permission that actually authorized this play/cast.

🔴 Blocker

• crates/engine/src/game/casting.rs:2255-2273 re-scans every graveyard permission on the source and takes the first one with a rider. It does not retain the selected permission definition (or rider) whose affected filter admitted this particular spell. A source with two overlapping GraveyardCastPermission definitions can therefore attach the wrong rider. Carry the selected permission provenance/rider through GraveyardPermissionSource and CastingVariant::GraveyardPermission, rather than looking it up after the cast. Add a collision regression proving that a spell admitted by a no-rider permission does not receive a different permission's rider.
• crates/engine/src/game/casting.rs:2282-2287 repeats the same source-wide lookup for lands; cover that path in the collision regression too.
• CI currently fails with E0063 at crates/mtgish-import/src/convert/player_effect.rs:243: the new StaticMode::GraveyardCastPermission field is not initialized. Thread granted_replacement: None (or the appropriate mapping) through every external constructor.

🟡 Non-blocking

• crates/engine/src/game/casting.rs:3173-3178 also changes emblem/static-definition behavior as an unrelated side effect. Please split it out or add focused evidence that it is required for this permission path.

✅ Clean

• The parser uses a typed static mode and the runtime tests cover the core permanent, persistence-after-source-leaves, and land-play scenarios. The verified Oracle text for The Eighth Doctor does require the granted leave-battlefield replacement.

Recommendation: retain the selected authorizing permission at choice time, use it for both spell and land finalization, add the conflicting-permission regressions, and repair the initializer before requesting re-review.