See parent task t-3594 for full context.

What#

1. Expand benchmark-types.ts: Add new categories (vision, OCR, audio, function-calling, multilingual, web-search). Add BenchmarkResultDetail type for per-call deep data (TTFT, tokens/sec, raw request/response, error classification). Add BenchmarkRankingOverride type for manual overrides.

2. Extend benchmark-storage.ts: Add methods for ranking overrides (save/load), per-call detail storage, suite import/export. Refactor to support the new types.

3. Refactor API routes: Move from flat /api/benchmarks/ to structured:

   • /api/benchmarks/suites — CRUD for suites
   • /api/benchmarks/suites/[id]/run — run a suite
   • /api/benchmarks/executions — list with filters (category, model, date range, suite)
   • /api/benchmarks/executions/[id] — detail with per-call data
   • /api/benchmarks/rankings — get/set manual overrides

Deliverables#

• Updated lib/benchmark-types.ts with all new types
• Updated lib/benchmark-storage.ts with new methods
• New API route files following the structure above
• Existing tests still pass + new tests for new functionality

Tags#

[benchmarks, api, data-model, refactor]