R01: P3: image reward recognition

Go implementation tracking epic

Selected stack: Go + Wails v2 + React/TypeScript + SQLite. Implementation starts on the next machine. Each child below is a separate PR; do not implement this epic in one PR.

Child tasks

• R01a: Reward recognition from image fixtures
  • Merged prerequisites: C02a, F03a.
  • Acceptance: R5: bounded Tesseract invocation, slot-indexed candidates, confidence/unknowns; held-out 1-4 reward and negative-image corpus with sample counts and recognition metrics.
• R01b: Reward image import and comparison UI
  • Merged prerequisites: R01a, I01c.
  • Acceptance: R5/R8: user-selected image, correct slot order, prices/ducats/ownership hints and uncertain-match correction; cached latency and scale demonstration.

Workflow

See TASKS.md, ROADMAP.md, and CONTRIBUTING.md. Aim for 100-400 handwritten changed lines per PR; split above 600 unless inseparable. Merge prerequisites before dependents. Update task evidence and this checklist after each completed PR. Close this epic only when every child passes; optional work can remain deferred.

All tasks are not started. Live game access and publication/merge actions follow the user's explicit authorization. The planning handoff does not claim runtime validation.

Start by reading TASKS.md, ROADMAP.md, and CONTRIBUTING.md, then separate the R01a recognition work from the R01b import and comparison UI work as described. R01a is done when its prerequisites, bounded Tesseract use, slot-indexed candidates, confidence and unknown handling, and held-out corpus metrics are covered; R01b is done when its listed image-selection, ordering, hints, correction, latency, and scale criteria pass.