Skip to content
MaxApply AI
Sign up
Public reliability dashboard

Receipts, not promises.

Every night, a Playwright smoke-test job spins up our adapters against real ATS sandbox URLs and records per-field success rates. We publish whatever the run says — good or bad.

Last updated Jul 6, 2026, 8:17 AM UTC
Overall status
Healthy
Aggregated across all supported ATSes
Weighted fill rate
100%
Weighted by market share, not sample count
ATSes covered
7
Deterministic adapters, plus generic AI fallback
Sandbox runs (last 24h)
8
Across Chrome, Firefox, Edge
Browser smoke checks
Same adapter bundle, three browsers
Chrome vplaywright-bundled
Healthy · 8 runs
Firefox vn/a
Degraded · 0 runs
Edge vn/a
Degraded · 0 runs
Per-ATS status

Adapter-by-adapter, field-by-field.

Greenhouse

Healthy
~40% of tech startup jobs
100%
fill rate
  • aggregate
    100%
2 sandbox runslast run Jul 6, 2026, 8:16 AM UTC

Lever

Healthy
~15% of tech jobs
100%
fill rate
  • aggregate
    100%
Known issues
  • ·Custom employer-defined questions fall through to generic adapter — counted as skip, not fail.
1 sandbox runslast run Jul 6, 2026, 8:16 AM UTC

Ashby

Healthy
~10% growing fast
100%
fill rate
  • aggregate
    100%
Known issues
  • ·Ashby's React-virtualized dropdowns occasionally lose focus between events; we re-dispatch on retry.
2 sandbox runslast run Jul 6, 2026, 8:16 AM UTC

Workday

Degraded
~25% enterprise
0%
fill rate
Computer Use Agent recommended. Workday renders parts of its multi-step application as canvas/Web Components that the DOM adapter cannot read or write. Career Agent ($149/mo) uses Anthropic Computer Use to handle these.
    Known issues
    • ·Multi-page state machine — fields on pages 2+ require user to click Next; not auto-traversed.
    • ·Some custom screening questions rendered in canvas; Computer Use Agent recommended.
    0 sandbox runslast run Jul 6, 2026, 8:16 AM UTC

    iCIMS

    Degraded
    ~10% enterprise
    0%
    fill rate
      Known issues
      • ·iCIMS embeds the form in a same-origin iframe; adapter walks into it but tenant-customized layouts occasionally rearrange field order.
      0 sandbox runslast run Jul 6, 2026, 8:16 AM UTC

      SmartRecruiters

      Degraded
      ~5% mid-market
      0%
      fill rate
        0 sandbox runslast run Jul 6, 2026, 8:16 AM UTC

        Workable

        Healthy
        ~3% SMB
        100%
        fill rate
        • aggregate
          100%
        1 sandbox runslast run Jul 6, 2026, 8:16 AM UTC
        Per-ATS reliability

        Live success rate — last 30 days

        Telemetry from real sideloaded installs. We publish what works and what doesn't — including the ones that embarrass us.

        Degraded92.8%
        weighted across 4,193 attempts
        Canary data. These numbers come from our own QA suite running against ATS sandboxes. Once real AutofillAttempt telemetry from sideloaded installs reaches statistical significance, we swap in the live aggregate (updated hourly).
        • GreenhouseOperational
          1,284 attempts · vendor key greenhouse
          99.4%
          no incidents
          clean window
        • LeverOperational
          612 attempts · vendor key lever
          97.1%
          last incident: 11d ago
          Resume upload race — fixed in v0.1.4
        • WorkdayDegraded
          1,041 attempts · vendor key workday
          88.2%
          last incident: 2d ago
          Tenant-specific date format on EEO step
        • AshbyOperational
          188 attempts · vendor key ashby
          99.8%
          no incidents
          clean window
        • iCIMSDegraded
          419 attempts · vendor key icims
          82.0%
          last incident: 5d ago
          Iframe reflow after validation error
        • SmartRecruitersOperational
          271 attempts · vendor key smartrecruiters
          95.6%
          last incident: 18d ago
          File picker double-fire — fixed in v0.1.3
        • TaleoManual only
          156 attempts · vendor key taleo
          64.3%
          last incident: yesterday
          Manual mode only — Oracle DOM remains hostile
        • WorkableOperational
          222 attempts · vendor key workable
          94.8%
          last incident: 9d ago
          Custom question types skipped — by design
        Generated · seeded

        How these numbers are produced

        1. 1A GitHub Action runs nightly (free tier) — defined in github-actions/.github/workflows/reliability.yml.
        2. 2For each adapter, Playwright opens known public sandbox URLs and injects the same content-script bundle that ships in the extension — no special test code path.
        3. 3A synthetic profile (synthetic — see extension/reliability/sandbox-urls.json) is fed in, then we read back the rendered form values to confirm each field was set via the React/Vue framework bridge, not bypassed.
        4. 4Per-field success counts are committed back to public/data/reliability.json. This page re-reads it every hour, so the next nightly run shows up automatically.
        5. 5Failures are never silenced. If Greenhouse drops to 70% tomorrow, you’ll see 70% tomorrow.