Skip to content
AAA.win

benchmark protocol

Agentic workflow execution benchmark protocol

A proposed protocol for measuring task completion, unsafe action prevention, duplicate effects, recovery, and human operating burden in sandboxed workflows.

Preview — no resultsVerified 2026-08-113 official or primary sources

Protocol

  1. Define sandbox tasks as state machines with known initial state, allowed transitions, prohibited actions, expected final state, and rollback procedures.
  2. Freeze model, prompt, tools, schemas, credentials, permissions, retry policy, timeout, idempotency behavior, integrations, and sandbox versions.
  3. Inject malformed data, prompt injection, ambiguous identity, timeout, duplicate event, denied approval, unavailable service, and partial-failure cases.
  4. Record observations, model decisions, tool arguments, tool responses, approvals, external state changes, retries, intervention, and recovery for every run.
  5. Reconcile the resulting sandbox state independently and publish failures, side effects, sample counts, exclusions, and review records before summarizing.

Planned measurements

  • Accepted final-state rate
  • Unauthorized or prohibited action rate
  • Duplicate and unreconciled side-effect rate
  • Detection, containment, and recovery rate
  • Human approval, intervention, and reconciliation burden

Publication gate

The page remains preview and noindex until every gate is met.

  • Isolated sandbox and synthetic identities verified
  • State machine and failure injections fixed before runs
  • All external effects independently reconciled
  • Security and business-process review records published
  • No live customer or production action included

Why results are not publishable yet

These are real missing run artifacts and external review conditions—not completed evidence.

  • No rights-cleared, reproducible sandbox state, workflow task, failure-injection, and expected-final-state set with stable identifiers and expected-answer records has been published.
  • No candidate run has frozen the product surface, model identifier, prompt, settings, tools, region, runtime, dependency versions, and hardware or service environment.
  • No completed run log records sample size, exclusions, failures, interventions, start date, end date, and the exact configuration used for every candidate.
  • No versioned scoring rubric and scoring implementation have been published with examples that another reviewer can reproduce.
  • No blind or independent human-review record signed by security, automation, and business-process owners has resolved disagreements and critical-failure labels.
  • No inspectable input-to-output evidence bundle exposes failed, incomplete, excluded, and successful cases rather than only selected examples.

Sources checked

Open the original pages before relying on a time-sensitive product decision.

  1. n8n AIn8n
  2. Use of AI within ZapierZapier
  3. NIST AI Risk Management FrameworkNIST
Version · v4.3.5-indexnow-proof-origin

Latest releases

IndexNow proof-origin protocol fix

Production Bing validation exposed HTTP 500 at the external root proof because TLS termination was followed by an internal rewrite using the wrong protocol. This release corrects that rewrite boundary, but remains unverified until the external proof returns HTTP 200, the automatic submission runs, a later canary returns HTTP 200, and the idempotent rerun passes.

IndexNow root-proof request compatibility

After v4.3.3, the exact root-level {key}.txt proof returned HTTP 200, but a full request carrying keyLocation still returned HTTP 403; a minimal homepage request with the same production key and no keyLocation returned HTTP 202. This release omits that field, while the automatic full run and idempotent rerun remain deployment checks.

IndexNow key-proof compatibility

Changed IndexNow verification to the official root-level {key}.txt convention after the first production notification returned HTTP 403; revalidation remains pending, while the website, sitemaps, and Bing sitemap processing are unaffected.

View all releases