Fix various UI performance or UI minor correctness issues

[chondl]

This PR fixes six distinct performance or minor correctness UI issues. Verified with a real browser against a local rig (CockroachDB + fake-gcs + full 2026 season, 3724 teams / 215 events / 18372 matches).

1. Match page: infinite media-request loop

Symptom. On every /match/{key} page the browser fired an endless stream of requests to TBA's /team/frc{N}/media/{year} endpoint — measured ~11,747 requests in 5 seconds on a single open page. Every production visitor has been silently doing this to TBA.

Root cause. frontend/src/pagesContent/match/[match_id]/imageRow.tsx:13-14 recomputed the teams array with .concat() on every render, producing a new array reference each time. The useEffect(..., [teams, year]) at line 19 then called getMediaUrls and setMedia, whose state update re-rendered the component, produced a fresh teams reference, and re-triggered the effect — an unbounded loop.

Fix. Wrap teams in useMemo(..., [data]) so its reference is stable across renders and the effect runs only when the underlying match data changes.

2. Match page: Previous/Next buttons don't update the page

Symptom. Clicking the prev/next arrows changed the URL and document title but left the displayed match unchanged.

Root cause. frontend/src/pages/match/[match_id].tsx:26 guarded the fetch with if (!match_id || data) return;. Because Next.js reuses the same component instance across /match/* route changes, data was already truthy from the previous match, so the guard short-circuited and never fetched the new match.

Fix. Compare the loaded key against the route — if (!match_id || data?.match?.key == match_id) return; — matching the existing pattern in pages/event/[event_id].tsx:25. The effect already depends on [match_id, data], so it now refetches when the route changes and skips redundant fetches once the correct match is loaded.

3. Strength of Schedule renders blank (or NaN) when pre-event EPAs are identical

Symptom. On the Strength of Schedule tab the RP/Rank/EPA/Composite score columns fail to populate. Two manifestations: score columns entirely blank, or NaN in EPA Score and Composite Score.

Root cause. strengthOfSchedule() (in frontend/src/pagesContent/event/[event_id]/worker.ts) runs a "Before Event" pass that computes epaSd, the standard deviation of every team's pre-event start EPA, and builds a Gaussian to score EPA-based schedule strength:

const distrib = Gaussian(0, (epaSd * epaSd * 5) / N);

Early in a season — before ratings diverge — every team at an event shares the same cold-start EPA, so epaSd is 0, the variance is 0, and gaussian throws Error('Variance must be > 0'). Because the worker's message handler calls strengthOfSchedule(...) without await/catch, the throw becomes a silent unhandled promise rejection: no message is posted and the table stays blank. A floating-point variant produces the NaN symptom — with all EPAs equal, Math.sqrt(sum(x^2)/n - avg^2) occasionally takes the root of a tiny negative rounding residue, so epaSd is NaN, gaussian does not throw (NaN <= 0 is false), and every EPA percentile is NaN. Both are the same degeneracy: EPA carries no schedule signal when all ratings are identical.

Fix. Floor the variance with || 1e-9, which treats both 0 and NaN as falsy and substitutes a negligible positive variance; with deltaEPA == 0 the CDF at the mean is 0.5, so every team gets a neutral 0.5 EPA percentile — the correct answer when EPA is non-informative. Any real positive variance passes through untouched. Matches the || 0 fallback idiom already used throughout this worker.

4. Simulation tab shows EPA 0 for every team

Symptom. On the Simulation tab the EPA column reads 0 for all teams (the predicted ranks, RP means, and percentiles are correct).

Root cause. frontend/src/pagesContent/event/[event_id]/simulation.tsx reads the team EPA as teamEvent.epa.total_points.mean. That matches the APITeamEvent TypeScript type, but not the runtime shape: the backend serves epa.total_points as a plain number, so .mean is undefined and ?? 0 renders 0.

Fix. Read epa.breakdown.total_points — the same field the SOS tab and the simulation worker already use — which is present at runtime.

5. Noteworthy matches rank a null/placeholder match incorrectly

Symptom. /v3/site/noteworthy_matches/2026 returned a 0/null-score, fully-DQed placeholder match (2026txmca_sf6m1) at the top of "Highest Clean Scores" (and the other lists), ahead of real high scores.

Root cause. backend/src/db/functions/noteworthy_matches.py — the lists order_by(desc(...)) without specifying null placement. CockroachDB orders NULLs first under DESC. For 2016+ the sort uses the no_foul columns; a match with no clean result on either alliance yields greatest(...) = NULL (and sum = NULL for combined), so it sorts to the top. (The value only surfaces on a NULLS-FIRST-defaulting DB, which is why it appears on staging.)

Fix. Add .nullslast() to every noteworthy order_by. Null-result matches fall past the top-30 cutoff; real high-scoring matches rank first; legitimate single-alliance-DQ matches still rank by their scoring alliance.

6. Every blob/API resource fetched twice on team pages

Symptom. On /team/*, each event/team blob was downloaded exactly twice per page load (observed on the production build, so not a StrictMode artifact).

Root cause. frontend/src/api/storage.tsx query() had no in-flight dedup. Several components request the same resource concurrently; each awaits getWithExpiry(), all miss IndexedDB (nothing is written until a fetch resolves), and all issue their own fetch.

Fix. Add a module-level in-flight promise map keyed by storageKey. Concurrent callers for the same key share one fetch; the entry clears once it settles. The fetch body is extracted into fetchAndStore unchanged.

Verification

• Match page: after the fix the steady-state media-request count is 0 (was ~11.7k/5s), and prev/next navigation updates the displayed match (Qual 43 ↔ Qual 44) with the arrows re-pointing correctly.
• Strength of Schedule: epaSd == 0 event — before: Gaussian throws, SOS blank; after: all columns populate (EPA Score 0.5). epaSd == NaN (float rounding) event — before: EPA/Composite NaN; after: populate (EPA Score 0.5). Event with a genuine non-zero EPA spread: EPA percentile 0.387 before and after — unchanged (the fix is inert on the non-degenerate path).
• Simulation: EPA column 0 before, real per-team EPA after; predicted-rank distributions unchanged and still vary run-to-run.
• Noteworthy: simulating staging's NULLS-FIRST ordering, 2026txmca_sf6m1/_sf9m1 ranked first or second above the real 964-point match; with the patched query, 2026dal_f1m1 (964) is first across all six lists and the placeholder is gone.
• Duplicate fetches: each event/team blob downloaded once per page load after the dedup (was twice).

[vercel]

@chondl is attempting to deploy a commit to the avgupta456's projects Team on Vercel.

A member of the Team first needs to authorize it.