fix: Score only confirmed predictions at market resolution and make claims select the confirmed row

[innocentharuna]

Summary

Resolution now scores only confirmed predictions, late confirmations are scored, and claims pick the confirmed row deterministically. I couldn't run anything here, so none of it has been run, including the unit and integration tests, lint, tsc and the openapi check. The EXPLAIN output and the repair script's dry-run output the issue asks for are not included.

• Resolution: atomicResolve in src/services/marketResolutionService.ts now updates only rows with status = 'confirmed' and result IS NULL. The FOR UPDATE transaction is unchanged. Pending and failed rows keep result = NULL, and replays never flip an existing won/lost.
• Late confirmation: the confirm UPDATE in src/workers/predictionsConfirmer.ts now sets result in the same statement. It takes the value from markets.winning_outcome when the market is already resolved and keeps any existing result.
• Claims: claimWinnings in src/services/claimService.ts selects only status = 'confirmed' rows, ordered by id, and fetches up to two. Two confirmed rows return 409 ambiguous_prediction and log both ids. If only unconfirmed rows exist, it returns 400 prediction_not_winning before any RPC. If the user has no rows at all, it still returns 404.
• OpenAPI: the 409 response is added to the claim operation in src/openapi/registry.ts. I also edited openapi.yaml by hand to match, since I couldn't regenerate it. If npm run openapi:check fails, run npm run openapi:generate (check the script name in package.json) and commit the output.
• Repair script: scripts/repair-unconfirmed-results.ts runs dry by default and writes only with --apply. It resets result on unclaimed, non-confirmed rows in resolved markets. Rows with a claim_tx_hash are listed, not changed. It also lists users with more than one confirmed row in a market.
• Docs: a "When is a prediction claimable" section is added to docs/predictions-api.md.

Changed files

• docs/predictions-api.md
• openapi.yaml
• src/openapi/registry.ts
• src/services/claimService.ts
• src/services/marketResolutionService.ts
• src/workers/predictionsConfirmer.ts
• tests/marketResolver.test.ts
• tests/predictionsConfirmer.test.ts
• scripts/repair-unconfirmed-results.ts
• tests/integration/resolutionScoring.test.ts

Test plan

CI should run npm test -- tests/marketResolver.test.ts tests/predictionsConfirmer.test.ts and npm run test:integration -- resolutionScoring against a migrated DB. Then run npm run lint, npx tsc -p tsconfig.json --noEmit, npm test with coverage, and npm run openapi:check.

• tests/marketResolver.test.ts: covers a market with confirmed, pending and failed rows (only the confirmed one is scored), and a replayed resolution that changes nothing. It uses an in-memory repo, so it doesn't exercise the SQL filter.
• tests/predictionsConfirmer.test.ts: one test checks that the confirm update's generated SQL scores late confirmations from winning_outcome without overwriting.
• tests/integration/resolutionScoring.test.ts: covers plan rows 1–8 against the real DB, including the failed-row claim (400, no RPC), the confirmed row chosen over a failed duplicate, the 409 for two confirmed rows, late confirmation, and the repair script's dry run and --apply. The two regression cases (rows 1 and 2) should fail on main and pass now.

Still to do for the PR: the EXPLAIN output on a 10,000-prediction market, the coverage summary, and the repair script's dry-run output against a seeded DB.

This PR was created as a draft by the issue solver bot. It will remain a
draft until repository CI passes.

Closes #984

[drips-wave]

@innocentharuna Great news! 🎉 Based on an automated assessment of this PR, the linked Wave issue(s) no longer count against your application limits.

You can now already apply to more issues while waiting for a review of this PR. Keep up the great work! 🚀

Learn more about application limits