[0x00101010]

Stacked on #220 (feat/scaling-limit-categories); will retarget main after #220 merges.

Why

The bench's proof-node step now judges two limits separately and reports each with its own budget: proof history must run at least 2× real time (head lag is recorded only), and witness generation (debug_executePayload p99) has a 3 s budget. The page still applied its own hard-coded budgets (lag ≤ 10 blocks, ≤ 1 s) to one merged "Proof node" row, so passing runs showed as over budget and unranked.

What

• Read both limits and their budgets from the run's proof_node step outputs; no proof-node budget is hard-coded anymore. Runs from before the bench reported them show their lag and p99 with the budget as unknown, and stay unranked.
• Split "Proof node" into two Execution limits in What breaks first, the ladder ticks, the subsystem cards and the guardrails page. Proof history: ceiling = run rate × headroom, shown as ≥ (the bench underestimates throughput). Witness generation: ceiling ≈ run rate × budget ÷ p99, tagged Estimated (assumes linear scaling).
• The guardrails page replaces the head-lag row with proof history headroom against the run's minimum, and judges executePayload against the run's budget.
• Disk rows take storage growth from the swaps run when one has it, falling back to the heaviest gas per transaction. The calibrated mix now has the heaviest transactions, but its storage stand-ins overwrite the same slots and overstate history bytes per gas until they are tuned. Subsystem disk cards still show each workload's own numbers.

Notes

• /scaling/guardrails swaps the head-lag row for proof history headroom; it has no screenshot here (only /scaling was retaken), and firstScreen.test.tsx covers it.

Evidence

Review: ready (one blocker fixed, re-reviewed) | high, 85% | deep (220 hand-written lines): claude-fable-5-1 + gpt-6-astra

View	Before	After
/scaling, full page

/scaling at 1280 px, from the running app. Synthetic data only: a copy of the synthetic fixture (network set to mainnet, the live archive-size fetch blocked) and a synthetic bench host served through Playwright with two 200 Mgas/s runs (mix: headroom 2.6×, p99 1400 ms; swaps: 3.4×, 900 ms).

Not verified beyond CI otherwise.

[vercel]

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

Project	Deployment	Actions	Updated
omni-ui	Ready	Preview	Nov 3, 2026 2:55am UTC

The merge-base changed after approval.

[cb-heimdall]

✅ Heimdall Review Status

Requirement	Status	More Info
Reviews	✅ 1/1	Denominator calculation Show calculation 1 if user is bot 0 1 if user is external 0 2 if repo is sensitive 0 From .codeflow.yml 1 Additional review requirements Show calculation Max 0 0 From CODEOWNERS 0 Global minimum 0 Max 1 1 1 if commit is unverified 0 1 if authored by an app that is not allowlisted and lacks verified human attribution 0 Sum 1