feat(scaling): unified run index

[0x00101010]

Summary

Why

Base-bench runs and load tests live on separate /benchmark pages with different columns, and the load-test list shows only dates, so comparing nights or setting a load test beside a bench run means opening each one. /scaling/runs puts both in one table with the same columns, and says why no run counts as sustained-capacity evidence yet.

What

• Move the report-api data layer (dataService, types, metricDefinitions, formatters, formatTransactions) from app/benchmark to app/scaling/runs/lib unchanged; app/benchmark imports it from there.
• toUnifiedRun maps a bench run or a load test to one row: env, machine, duration, profile (sustained at ≥ 1 h, else burst), workload, gas/s, latency, status, areas, and qualifies: { ok: false, reason }. State is unknown until producers record it; bench duration is null (shown as —) because metadata has no block count.
• /scaling/runs: newest first, grouped by UTC week, with when / kind / env / profile / workload / Mgas/s / latency / status / duration / machine, and filter chips for kind, env and qualifying-only. Rows link to /scaling/runs/bench/<id> and /scaling/runs/load/<network>/<timestamp>, which do not exist yet and 404 for now.
• A "Runs →" link in the first-screen header; /scaling/runs added to the CI check that internal-only routes 404 on the public build.

Notes

• Stack: feat(scaling): snapshot schema v2 with ladder, ceilings and guardrails #194 → feat(scaling): add ceilings and guardrails to the snapshot #195 → feat(scaling): first screen with ladder, ceilings, guardrails and progress #196 → feat(scaling): ceiling cards and guardrails #197 → feat(scaling): progress from Linear and collapsed detail sections #198 → feat(scaling): detail pages and unified run index #199 → feat(scaling): per-area detail pages #200 → this PR.
• Only the newest 30 load-test results are fetched (one request each); the header says "the latest 30 of N". Only sepolia is listed because report-api has no endpoint listing networks.
• Bench latencies are labelled get-payload / new-payload, not p99: metadata does not say which percentile they are. Load-test latencies are block and Flashblocks p99.

Screenshots

Internal build against a local server serving the committed synthetic fixture (invented numbers, not a real network).

/scaling/runs at 1280 px:

/scaling header with the Runs link:

Test plan

• npm test, npm run typecheck, npm run lint, npm run build, npm run build:internal, npm run llms -- --check, npm run agents -- --check; the public build serves 404 for /scaling/runs.
• unified.test.ts: bench and load mapping from the synthetic fixture, including an incomplete bench run, a load test without config, a fatal load test, a missing result, and week boundaries.
• runs/page.test.tsx: both kinds ordered newest first under week headings with detail links; the data-source error is shown.
• Rendered /scaling/runs in npm run dev:internal against the live report-api and against the fixture; kind, env and qualifying-only chips filter as expected.

[vercel]

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

Project	Deployment	Actions	Updated
omni-ui	Ready	Preview	Oct 16, 2026 1:00pm UTC

[0x00101010]

GitHub marked this merged because its commits were fast-forwarded into the branch of #199 (consolidating 8 PRs into 3). Nothing reached main; review continues on #199.