bao/genshin-assistant
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat(scanner): add native artifact pipeline

Browse Source 
Add native IK-style capture processing, Artifact Inventory, explicit promotion and single-result review. Confirm the three live OCR corrections in the eval corpus and preserve extraction/value separation.
This commit is contained in:
AzuTear
2026-07-09 23:30:42 +02:00
parent 28d60eb915
commit 639b0b7f59
92 changed files with 13606 additions and 1703 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs/MERGE_READINESS.md
+9 -10
View File
@@ -1,6 +1,6 @@
# Scanner Merge Evidence
Merged branch: `codex/ik-scanner-progress`
Merged branch: scanner baseline branch
Target branch: `main`
Merge commit: `c025daa`
Merged on: 2026-07-09
@@ -13,17 +13,17 @@ This document records the evidence used to merge the scanner branch into
| Area | Status | Evidence |
| --- | --- | --- |
| TypeScript/lint gate | Passed | `npm run lint` |
| Unit/regression suite | Passed | `npm test` with 207 tests, including equipped footer, lock detection, and PNG lock-crop regression coverage |
| Unit/regression suite | Passed | `npm test` with 209 tests, including equipped footer, lock detection, and PNG lock-crop regression coverage |
| Production build | Passed | `npm run build` |
| Whitespace check | Passed | `git diff --check` |
| OCR eval gate | Passed | `npm run eval` with `23/23` exact-match cases and `100%` critical fields |
| Scan assessment self-test | Passed | `npm run scan:assessment:test`; the fixture intentionally expects `ik-traineddata` to win its synthetic `limit=100` case while real live-winner claims stay tied to archived live assessments |
| Scan assessment self-test | Passed | `npm run scan:assessment:test`; the fixture rejects fast but low-quality synthetic runs |
| Live runtime preflight | Passed | `npm run scan:live:preflight`; signature `2026-07-08-direct-gdi-reviewfix`, elevated yes, Genshin found |
| Safe visible-inventory scan path | Passed | 2026-07-09 live run: `/scanner/start?entry=visible-inventory&limit=20&engine=current` completed `20/20` verified and parsed, `19` stored, `1` duplicate, `0` review, `0` misses, `8047 ms` elapsed |
| Safe visible-inventory scan path | Passed | 2026-07-09 live run: `/scanner/start?entry=visible-inventory&limit=20` completed `20/20` verified and parsed, `19` stored, `1` duplicate, `0` review, `0` misses, `8047 ms` elapsed |
| Equipped character OCR/persist path | Passed for live smoke | 2026-07-09 live captures read equipped footers for `Citlali` and `Linnea`; parser regression covers `Equipped: Linnea l` -> `Linnea` |
| Unlocked lock state | Passed for live smoke | 2026-07-09 Smart Capture reported `locked: false` on an unlocked artifact; after restart, `lockSignal.ratio: 0` with threshold `0.06` |
| Positive locked lock state | Passed | 2026-07-09 Smart Capture on a visibly locked artifact reported `locked: true`, `lockSignal.ratio: 0.14797913950456323`, threshold `0.06` |
| Lock-state persistence | Passed | 2026-07-09 `/scanner/start?entry=visible-inventory&limit=1&engine=current` stored `A Note in Spring's Leich` with `equipped: "Citlali"` and `locked: true` |
| Lock-state persistence | Passed | 2026-07-09 `/scanner/start?entry=visible-inventory&limit=1` stored `A Note in Spring's Leich` with `equipped: "Citlali"` and `locked: true` |
| Lock-state diagnostics | Passed | Capture results include `lockSignal.ratio`, `lockSignal.threshold`, and the lock crop rect. Detection uses decoded PNG crop pixels to avoid native bitmap channel-order ambiguity |
| Unsafe auto-entry default | Mitigated | Normal guided Auto-Scan now blocks when no artifact detail card is visible instead of falling back to `auto-entry` |
| Review-to-eval export | Passed by tests | `npm run eval:review-candidates` script is covered by `src/eval/reviewEvalCandidatesScript.test.ts` and output is Git-ignored |
@@ -37,7 +37,7 @@ npm run build
git diff --check
```
Latest run after the lock-state fix: passed with `207` tests.
Latest run after the cleanup: passed with `209` tests.
`npm run eval` and `npm run scan:assessment:test` also passed after the
lock-state fix.
@@ -49,10 +49,9 @@ lock-state fix.
- The optional `3 artifacts/second` target is not proven. The stable current
path is closer to `2.5` to `2.75 artifacts/second` on clean 20-artifact live
runs.
- Native Tesseract/IK-traineddata is not the default. The latest documented
qualified live 100-artifact comparison in this environment had `current` as
winner; the assessment self-test is a synthetic validator fixture, not a live
winner claim.
- Alternate OCR-engine comparison paths have been retired. The current scanner
path is the maintained baseline; future engine work should be reopened only
with a new explicit product reason.
## Merge Result