Reseed knockout slots from closed groups and forfeit unscored picks

[borland667]

Summary

Two coordinated changes that take effect once a group's official group.result is published:

• Predict wizard bracket display: knockout slots that reference a closed group (1A, 2B, 3A/B/C/D, ...) now resolve to the real finishers instead of the user's predicted placement. The user's stored group prediction is preserved untouched — this is purely a display change so the user picks a winner among the teams that actually advanced.
• Scoring carve-out for invalidated picks: a knockout match is treated as unscoreable when the user's stored predictedWinner is a team that did not actually play it (e.g. they predicted 1A = MEX, picked MEX to win 1A vs 3B, but the real 1A was Korea so the match was played as KOR vs HAI). The match contributes 0 to earned knockoutScore AND 0 to per-user maximum-possible. The forfeited pointsPerCorrect is summed into a new Score.forfeitedKnockoutPoints column so UIs can compute an effective max as tournament.rules.totalMaximumPoints - forfeitedKnockoutPoints. The user is neither rewarded nor penalised for a pick they could not have made correctly.

Scope is intentionally minimal — the leaderboard UI is not yet showing the per-user effective max, that can be a follow-up.

Files

• src/pages/Predict.jsx: merge group.result over groupPredictions for slot resolution in RoundStep
• api/scoring.cjs: scoreKnockoutPrediction returns { score, scoreable, maxPoints }; calculateTotalScore aggregates forfeitedKnockoutPoints
• api/app.cjs: ship selectedHomeTeamId / selectedAwayTeamId into the scoring context; persist + serialize forfeitedKnockoutPoints on leaderboard responses
• prisma/schema.prisma + prisma/migrations/20260615000000_score_forfeited_knockout_points/: add Score.forfeitedKnockoutPoints Int @default(0)
• tests/scoring.test.mjs: cover the new return shape, the skip rule, and the forfeit aggregation
• docs/ARCHITECTURE.md §10.2 + docs/IMPLEMENTATION_STATUS.md: document the new scoring and bracket-seeding behaviour

Test plan

• npm run verify — 49/49 tests pass, build is clean
• In the Predict wizard, seed a closed group whose real 1A differs from the user's pick. Open a Round-of-16 step that references 1A vs 3B/C/D. Confirm the bracket card shows the real 1A team and the dropdown lists the real participants
• Confirm the user's stored group placement is unchanged after viewing the wizard (no auto-save)
• Run a results import for a knockout match whose participants don't include the user's stored pick. Confirm Score.knockoutScore is unchanged and Score.forfeitedKnockoutPoints increased by pointsPerCorrect for that round
• Confirm correctly-predicted users score normally — the skip rule only fires when the picked team did not appear in the match

Made with Cursor