[dayfine]

The runner now also accepts the lean data-authored frlg-parcel-00.05.plan.json via authored-binding.json, with a dedicated JSON Schema and parser. The same file is used by the production story table in dayfine/Arduino-Source#260. Native CI checks byte equality and runs all seven Python scenarios; native engine integration tests pass on Linux and Windows. Captured exports remain separate migration fixtures, and authored reports do not claim a native source revision or observed state.

Load captured checkpoint events through a reusable offline runner with registered handlers. FRLG spatial behavior lives in the game adapter; scene receipt requirements are binding data. The real Parcel capture is the first example, and a synthetic checkpoint with different IDs/order/repeated steps/receipt verifies reuse.

Selected cases set input.check_actions: true and list the entire ordered expectation.actions. These are semantic intents: straight walks for X tiles, facing changes, hops, map transitions and scene requests. They do not count button presses or encode timings. Missing/extra/reordered actions and incorrect directions/distances fail even if the final status says complete. An interrupted walk retains its requested distance; its uncertain displacement is asserted separately.

Start with data/story/parcel-pilot/CASES.md, generated from authored fixtures and checked for drift. The success case parcel-all-events lists all 23 semantic actions. Four cases assert exact actions, including an empty list on failed entry and no scene after an unconfirmed Mart transition. Then review checkpoint_replay.py, games/frlg/story_replay.py, the handler/binding modules and adjacent tests. Schemas enforce the opt-in assertion contract; CI uploads the actual diagnostic trace even when other checks fail.

This remains an offline diagnostic: observations are synthetic, scene dialogue/animation is not simulated, save/retry policy is explicitly stop/no-save, and native execution parity remains open. No native authoring table, approved map bytes or consumer pins change.

Validation: local full suite 520 tests passed (39 optional skips); schema/semantic checks, compact formatting and generated case docs pass. Negative controls alter successful action logs to verify omissions/extras/reordering/direction/distance failures; tests also distinguish requested walk distance from uncertain progress, split hops/warps from walking, and reject unexecuted planner endpoints. CI reruns on this revision.

Next: native trace consumer and actual runtime/save/retry policy export, reconcile initializer/STEP/scene/end-save semantics, then bind recovery and BDSP capture. Native exact-leg comparison requires matching planner policies; these expectations currently describe the pinned Python profile.