Search

Search benchmark pages, protocols, published runs and the docs

Rout
Method

Method changelog

Dated notes on every change to the method. When a change makes earlier figures not comparable, the note says so and the affected runs link here.
  1. Protocols made partner-neutral

    The head-to-head guard protocol is now A2-guard-vs-incumbent, with arms rout_guard and incumbent_guard; Block-0's control arm is partner_path. Wording only: no parameter, metric or rule changed.

  2. Bundle index and site

    rbench.index writes the index the site reads. No method change; no figures exist yet.

  3. E0 primary comparison is nonce vs blockhash

    A sender's docs say durable nonce can hurt landing on some paths, so nonce vs blockhash became the primary comparison. The reference-RPC comparisons are secondary.

  4. A1 compares two guard implementations

    external_flow judges net flow at one slot; a typical launch guard sums gross buys. A1 now pre-registers a slot ladder and a provider-aggregated gross window, and counts the ≥1-slot structural delay against Rout.

  5. Verifiable assignment, timing and anchors

    Arms are recomputed from sha256(run_id:seq); probes exist only at hash-scheduled slots; anchors are re-fetched from chain; cluster-bootstrap intervals beside Newcombe; finalized commitment by default.

  6. STEP0 reports the Block-0 cohort both ways

    Create-slot buyers on a bundled launch are usually the creator's own bundle, so STEP0 publishes the result with that cohort counted as external and as insider.