Improve L2 market parsing and data handling

docs/goals/store-icon-progress/GOAL.md

@@ -0,0 +1,21 @@
+# Goal: Store Icon Progress Overlay
+
+Use Krypton Execution to execute `docs/goals/store-icon-progress/PLAN.md` after the required Buy-store fixture and catalog icon URL evidence are available.
+
+Core rules:
+- Treat PLAN.md as the source plan.
+- Create `data/fixtures.json` from the reviewed 17-file corpus before changing recognition behavior.
+- Keep fixture expected data human-owned; tests may report actual values but must never bless or rewrite expectations automatically.
+- Keep fixture calibration, templates, catalog JSON/icon bytes and sequence pending/found states human-owned and versioned.
+- Run fixtures through the single `src/frame-analyzer.js#analyzeFrame` path used by live capture; remove the inline implementation from `main.js` and do not create a test-only parser.
+- Use real production OpenCV/Tesseract/parser in the Playwright lane; substitute only catalog transport with committed fixture assets.
+- Preserve the active result as slot-state truth owner.
+- Keep old six-field calibration readable and operational.
+- Do not infer the parsed slot from tooltip position or hover order.
+- Mark a slot green only for a unique best visual match above both the absolute threshold and second-best margin; ties remain yellow.
+- Do not send runtime images or slot state to the backend.
+- Make the active source preview readable at the real source-card scale before accepting overlay evidence.
+- Verify optional grid calibration survives the hosted PUT-to-GET round trip.
+- Capture yellow-to-green browser evidence and record it in EVIDENCE.md.
+- Finish with one `npm run check` gate covering unit tests, all fixture screenshots, the production build and `git diff --check`.
+- Say `implemented but unproven` if Buy/Sell or catalog-icon matching cannot be demonstrated.