Overall gate Pass

Depends on both the required-quality gate and the performance gate.

Required suites 12/12

Required suites currently passing.

English 1K p95 47.9 ms

Current gating latency check for English 1K text.

Frozen cases 534

Total cases across the current frozen suites.

02

Performance snapshot


The public page highlights the current 1K-character gating snapshot rather than publishing the full internal performance report.

03

Detector precision / recall snapshot


Per-entity-type precision/recall is published against an adversarial-FP and Saudi-name-recall corpus. Dated in-repo snapshots on curated corpora — not an external audit and not a claim of production-wide coverage.

Precision / recall gate Pass

Dated detector precision/recall snapshot.

Total cases 350

Total measured cases in the public JSON artifact.

False positives 2

Total false positives across the published slices.

False negatives 0

Total false negatives across the published slices.

04

Current benchmark suites


These are the suites currently included in the public benchmark gate.

Published suites

    05

    Research corpora coverage


    These suites were folded in from the 2026-04-26 detector research package. They expand evaluation coverage across public-domain Arabic literature, Saudi code-switched business text, and adversarial PII attacks. This section is shown separately from the frozen public gate until the new corpora complete trend-history stabilization.

    Dated in-repo benchmark over curated public-domain and synthetic research corpora; not an external audit or production-wide coverage guarantee.

    Research corpora gate Pass

    Separate from the frozen public gate.

    Total research records 1283

    Across all research suites.

    False positives 4

    Across research corpora.

    False negatives 5

    Across research corpora.

    Per-suite breakdown

      Adversarial attack-class metrics

        06

        Artifact and method


        This page reads the published public JSON summary. It is intentionally narrower than the internal benchmark report.

        Last public update: 2026-04-29T14:40:52Z
        07

        Claim boundary


        This page is a benchmark snapshot, not a blanket claim that every customer payload or every future detector change is perfect. Public benchmark language should stay tied to a fresh artifact.

        The public benchmark page shows only the suites currently included in buyer-facing claims.


        Evaluate the product with the evidence in hand.

        Evaluate →