Sentari
Changelog

What we ship, when we ship it.

Public release notes for every detection model update, dataset drop, and product change. We list both wins and fixes — same as our methodology, transparency is the product.

Feature Detection Fix Trust
  1. May 24, 2026v2.7.0 Feature

    Recruiter Verification public beta

    Independent verification of recruiter accounts across LinkedIn, Wellfound, and email outreach — with weighted account-age, employment, and hiring-history signals.

    • ›Four-signal model: account age (30%), employment (30%), hiring activity (20%), outreach (20%)
    • ›Public profile pages with band classification (verified / elevated / fraud)
    • ›Dispute flow for recruiters to submit additional evidence
  2. May 11, 2026v2.6.3 Detection

    Salary-bait detector

    New classifier targets postings that advertise a salary band intentionally inflated relative to the role's market p50.

    • ›Trained on 12.4K labeled examples across 38 job categories
    • ›Precision 94.1% / recall 88.7% on held-out 2026 Q1 set
    • ›Flags now show the inferred market band alongside the posted range
  3. Apr 28, 2026v2.6.0 Trust

    Public Scam Library

    Released the public taxonomy of eight scam patterns with prevalence data, severity ratings, and real-world wording examples.

    • ›Indexed by Google within 48 hours of launch
    • ›Embeddable widget for partner consumer-protection sites
    • ›Quarterly refresh tied to transparency report cadence
  4. Apr 15, 2026v2.5.4 Feature

    Browser extension v1.0

    Chrome and Firefox extensions hit general availability with inline trust scores on 30+ job boards.

    • ›Zero-config activation on LinkedIn, Indeed, Glassdoor, WTTJ, Wellfound, Otta, and 25 more
    • ›Local-first: posting URLs are hashed before leaving the browser
    • ›Average added latency on job pages: 78 ms
  5. Apr 02, 2026v2.5.0 Detection

    Off-platform pressure scoring upgrade

    Detection model now scores conversation-level pressure to move applicants to WhatsApp, Telegram, or Signal.

    • ›Replaces prior keyword heuristics with sequence-level transformer scoring
    • ›Cuts false positives on legitimate technical-recruiter screening flows by 41%
    • ›Inline explanation lists the specific phrases that triggered the flag
  6. Mar 18, 2026v2.4.2 Trust

    Account exports + GDPR controls

    Self-serve data export and one-click account deletion from the account page. No support ticket required.

    • ›JSON + CSV exports of every scan, report, and recruiter check
    • ›Hard-delete propagates to backups within 14 days
    • ›Audit log entries retained for 90 days under cryptographic hash only
  7. Mar 04, 2026v2.4.0 Fix

    Reanalyze button on scan results

    Scan detail pages now expose a manual reanalyze action that re-runs the current model on the original payload.

    • ›Useful after a methodology release to see updated scoring on historical scans
    • ›Old and new scores are shown side-by-side with the model version
  8. Feb 22, 2026v2.3.0 Trust

    Ghost-job index Q1 2026

    Quarterly research drop covering 1.4M postings across 12 markets. Ghost-listing share crossed 22% for the first time.

    • ›Open dataset under CC-BY-4.0 for journalists and researchers
    • ›Methodology, sample frame, and confidence intervals published in full
    • ›Picked up by 3 national newspapers within a week

Want releases in your inbox?

Subscribe via the extension page or follow our quarterly transparency drops.

Get the extension Transparency