refactor: polish pass

[adibarra]

Note

Medium Risk
Compare page static-param removal and redirect logic affect SEO/caching; trend interpolation and JSON-LD escaping change user-visible chart gaps and structured-data safety; feedback failure events include truncated error text sent to analytics.

Overview
This polish pass adds server-side observability via a lightweight trackServer helper (PostHgo /capture/ or console fallback), wired into feedback API failures (E_CRYPTO / E_INSERT) and per-dollar PNG render failures (502 + short cache instead of opaque 500s).

Compare routes stop pre-enumerating slugs in generateStaticParams on the HTML pages (still fully dynamic SSR) and only 308-redirect when the slug differs from canonical after case-insensitive comparison, avoiding redundant CDN cache entries for mixed-case URLs that already match content.

Chart/trend correctness: interpolated trends return null (gaps) when a metric is missing on any frontier point instead of drawing a misleading zero line; lightweight trend points gain optional measured power fields and drop unsafe casts. Dynamic x-axis reads use typed AggDataEntry accessors in chart and trend hooks.

JSON-LD now escapes >, &, and < in inline script payloads (with tests). Feature gate re-lock is centralized in relockFeatureGate() with unit tests. Latest-image age/outdated helpers move to latest-image-utils with Vitest coverage. FeedbackForm loads via next/dynamic in the nudge registry. Root lint/fmt scripts pass --no-error-on-unmatched-pattern consistently.

^{Reviewed by Cursor Bugbot for commit 7c171aa. Bugbot is set up for automated code reviews on this repo. Configure here.}

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
inferencemax-app	Ready	Preview, Comment	May 27, 2026 6:22pm

[adibarra]

@claude review

[claude]

Claude finished @adibarra's task in 3m 12s —— View job

---

Review

🟢 LGTM — no blocking issues found.

Read the diff across all 21 files. The polish pass is well-scoped, with new tests covering the extracted code (29 tests for latest-image-utils + use-feature-gate, additional tests for JsonLd escaping and interpolateMetricAtInteractivity missing-metric path).

Verified

• analytics-server.ts — fire-and-forget fetch with keepalive, falls back to console when key missing, distinct_id is intentionally synthetic. Both call sites (feedback/route.ts, PNG route) truncate error_message to 500 chars before sending, so internal stack details don't bloat PostHog payloads.
• JSON-LD escaping (json-ld.tsx:12-15) — order of < / > / & replacements is safe because each replacement produces \u00xx sequences containing none of the three trigger chars, so no double-escape risk. Tests cover the </script> breakout vector explicitly.
• Trend null-propagation (useInterpolatedTrendData.ts:127-133) — bails out entirely if any frontier point lacks the metric, including the single-point branch (now routes through extractMetric). Prevents the flat-zero regression that motivated the change.
• Compare lowercase redirect (compare/[slug]/page.tsx:85, compare-per-dollar/[slug]/page.tsx:80) — canonical is always lowercase, so comparing against slug.toLowerCase() correctly avoids redundant 308s for mixed-case URLs that already match canonical content. SEO consolidation still works via alternates.canonical.
• PNG route try/catch — scope is limited to ImageResponse; parsed/sequence/precision are declared outside the try and remain in scope in the catch. 502 + short s-maxage=60 cache is the right call vs. pinning a permanent failure.
• useChartData.ts:313-314 / useTrendData.ts:118-123 — typed Partial<AggDataEntry> accessor with typeof === 'number' narrowing replaces the as any. Future field rename would now fall back to d.x cleanly instead of silently masking a regression.
• relockFeatureGate() — centralizes the localStorage key + event dispatch; three callers (FeedbackViewer, GpuPowerDisplay, SubmissionsDisplay) all migrated. Tests cover idempotency and event firing.
• Lazy FeedbackForm (registry.tsx:18-21) — next/dynamic with ssr: false is correct since the modal only renders after the eligibility event fires client-side.

Non-blocking observations

• registry.tsx:18-22 declares const FeedbackForm = dynamic(...) between import statements (GitHubIcon import follows it). ES module imports hoist, so this is functionally fine, but it's an unusual order — moving the const below all imports would read more conventionally. Not worth blocking on.
• The PNG route's outer DB/data calls (getCachedBenchmarks, computeCompareTableData) still throw uncaught — failures there will surface as opaque 500s without PostHog tagging. Likely intentional (different failure mode from the Satori-render path the new tracking targets), but worth confirming if you want full observability coverage on that endpoint.

---

• Read PR diff and analyze changes
• Review feedback API server-side analytics
• Review compare-per-dollar PNG route changes
• Review interpolated trend data null handling
• Review JSON-LD escaping
• Review latest-image-utils extraction + tests
• Review use-feature-gate refactor + tests
• Check for missing tests on new code
• Post final review comments
  — branch refactor/polish-pass