Oklok
Accuracy Dashboard

The numbers we're measured on, in public.

Rolling 30-day accuracy, dispute outcomes, and latency for every tier. Updated nightly from production. The published numbers below are the SLA — the same ones the Master Service Agreement references.

Standard tier launches May 6, 2026.

0 verifications processed yet. The methodology below is published, contractually binding, and won't change retroactively. Public live counters update nightly once production verifications begin.

We chose to ship this page empty rather than seed it with synthetic numbers. Every figure that appears here from launch onward is real production data.

Last updated Apr 30, 2026, 09:47 AM UTC · regenerated nightly via Supabase scheduled function.

Accuracy Methodology · v1.0 · Effective May 1, 2026

How we measure accuracy

This page defines the SLA measurement Oklook commits to in every Master Service Agreement. The numbers above are computed nightly from the rules below. Customers can verify the calculation themselves via the public counters and dispute resolution audit log.

Showing locally-cached fallback copy. The latest version is always served from the Oklook database; this fallback only appears when the live source is briefly unreachable.

What "99.5% accuracy" means at Oklook

We measure accuracy on a rolling 30-day window, ending at the last full UTC day. Three independent signals feed every accuracy number you see on this page.

1. Shadow audits — 70% of the score

For 8% of production verifications during your first 90 days as a customer (5% steady-state), we silently route the same job to a second qualified reviewer. Neither reviewer knows it's happening. We compare their per-check verdicts.

If the two reviewers agreed on the objective checks, the original verdict counts as accurate. If they disagreed, our adjudication panel reviews the asset and picks the right answer; if the original reviewer was wrong, that counts as an accuracy miss.

2. Customer disputes — 25% of the score

When you dispute a verdict, our adjudication panel reviews within 1 business day. If the panel finds Oklook was wrong, that counts as one accuracy miss. If the panel finds we were right, no impact on the metric. If the dispute is over a subjective check, the dispute is logged but does NOT count for or against the SLA — see the next section.

3. Trap performance — 5% of the score (internal)

We continuously inject known-good verifications into the queue at the same rate as production work. Reviewer accuracy on these traps is the third input. This signal is internal-facing only — it doesn't appear in the per-customer numbers but feeds our composite score.

Subjective checks are excluded

Some checks (like "tone matches the brief") are inherently subjective. We tag those subjective in the rubric, route them only via consensus tiers (Standard or Enterprise), and exclude them from the SLA calculation. The consensus verdict is treated as the right answer.

Adjudication

Through 2026 Q3, our internal team adjudicates disputes within 1 business day. From Q4 we move to a three-person panel, and post-revenue we'll add a rotating third-party panel. Customers can escalate to binding arbitration only if we can't agree.

Our commitment

The numbers on this page are contractually binding. The Master Service Agreement references this methodology directly (§5.1). If we miss your tier's SLA over a calendar month, we automatically apply a 25% credit to your prepaid balance within 5 business days. You don't have to ask.


Last updated: 2026-04-30 (v1.0) Tier targets: Standard 99.5% / Enterprise 99.9% / Fast (best-effort, no penalty)

Monthly Summary

Last six months, in one table.

Calendar-month rollups, the same numbers that drive SLA-credit issuance under MSA §5.1.

No completed months yet.

The first monthly summary will appear after May 31, 2026. Each row reports the rolling-30-day accuracy for both consensus tiers, the dispute outcomes that fed the calculation, and whether an SLA credit was automatically issued under MSA §5.1.