Rank spirit scores on the average, not the rounded value

[Joe2k]

Kerala's spirit captain reported a three-way tie at 10.4 between Bengal, Kerala and Karnataka that shouldn't have been one:

Karnataka  83/8 = 10.375
Kerala     73/7 = 10.4286

Both displayed as 10.4 — and both were being ranked as 10.4.

Cause

update_tournament_spirit_rankings rounded the average to one decimal before storing it, and rank_spirit_scores built its order from the set of distinct rounded values:

spirit_points = sorted({r["points"] for r in scores}, reverse=True)
score["rank"] = spirit_points.index(score["points"]) + 1

So any teams sharing a displayed figure necessarily shared a rank. Kerala and Karnataka are 0.054 apart — more than half a rounding step.

team    total   n     true avg    before          after
KL         73   7    10.428571    10.4  rank 1    10.43  rank 1
BT       83.2   8    10.400000    10.4  rank 1    10.40  rank 2
KA         83   8    10.375000    10.4  rank 1    10.38  rank 3

The rounding was correct — 10.375 → 10.4 is proper. Ranking on the rounded value was the defect.

Change

• Ranks come from the unrounded average.
• Ties break on final placement, read from current_seeding ({position: team_id}). Ranks are consequently sequential where equal scores used to share one — visible in the updated test_api.py expectation, where six teams on 0.0 move from all-rank-3 to ranks 3–8 in finishing order.
• Scores are kept and shown to two decimals (backend rounds to 2; the table uses toFixed(2)), so these three now read 10.43 / 10.40 / 10.38 instead of three identical 10.4s.
• Averages are compared at six decimals when ranking. Ten matches of 10.4 accumulate to 10.400000000000002, and raw float comparison would rank that above an exact 10.4 on representation alone, silently pre-empting the placement tie-break.

final_standings is optional because migration 0039 still calls this with one argument, on rankings recorded before self_points existed — covered by a test.

Testing

./scripts/lint passes. Five tests in server/tests/test_spirit_ranking.py use the real BT/KL/KA figures. Full suite: 489 tests, 3 errors, all pre-existing and unrelated (TOPSCORE_CLIENT_ID env var, two SeleniumBase UI tests).

Note

Standings won't change on merge. spirit_ranking is stored JSON, recomputed only when a spirit score in that tournament is next saved (api.py:2618, 2652). The affected tournament needs a recompute — happy to add a management command if that's easier than re-saving a score.

🤖 Generated with Claude Code

[Copilot]

🟢 Approval recommended

The change is narrowly scoped, matches the described defect, and is backed by targeted new tests plus updated API expectations.

Pull request overview

This PR fixes spirit ranking ties caused by ranking on a rounded average (1 decimal) instead of the full-precision average, ensuring teams with distinct underlying averages no longer get forced into the same rank when their displayed values match.

Changes:

• Store full-precision per-team spirit averages in update_tournament_spirit_rankings, and defer rounding until after ranking is computed.
• Update rank_spirit_scores to rank by full-precision average (rounded to 6 decimals for float-noise resistance) and break ties using tournament.current_seeding.
• Add focused unit tests for the regression case (BT/KL/KA) and adjust API test expectations for sequential (tie-broken) ranks.

File summaries

File	Description
server/tournament/utils.py	Ranks spirit scores using full-precision averages and uses current_seeding as a deterministic tie-breaker before rounding for display/storage.
server/tests/test_spirit_ranking.py	Adds regression and behavior tests covering full-precision ranking, tie-breaking by placement, float-noise handling, and backward-compat behavior.
server/tests/test_api.py	Updates expected spirit ranking output to reflect sequential ranks after tie-breaking.

Review details

• Files reviewed: 3/3 changed files
• Comments generated: 0
• Review effort level: Lite

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.