Specwarden lockupspecwarden

Build progress

Engineering milestones leading to Specwarden public launch. Newest first.

Specwarden is currently in private beta. Join the waitlist to be invited when paid plans open.

Waitlist launch

Latest

2026-05-19

specwarden.ai is live. Engineering teams can sign up to be notified when paid plans open.

  • Marketing site + waitlist form live at specwarden.ai
  • Resend audience integration + transactional welcome email
  • App routes gated by env-flag, ready for one-toggle public launch

PFMEA engine + multi-standard support

2026-05-18

Process FMEA review added alongside Design FMEA. Standards selector restructured around scoring rubric, not document type.

  • PFMEA review engine with 6 PFMEA-specific rule checks (U-101, U-102, D-101, D-102, D-103, M-101)
  • PFMEA-tuned Sonnet prompt with separate Control Method handling for Prevention vs Detection
  • Standards page restructured: AIAG-VDA Handbook (2019) and AIAG-RPN (FMEA-4) are first-class rubrics
  • Detection banner in mapping UI when document type is auto-detected from headers

Reviewed Copy + dark mode

2026-05-18

Two product surface improvements requested by early reviewers.

  • Reviewed Copy export — your original FMEA spreadsheet with findings highlighted as cell comments
  • System-wide dark mode with persistent preference (Settings → Appearance)

Closed-loop validation

2026-05-17

Post-action review per AIAG-VDA §2.6.3 — verify whether your taken actions actually moved the risk score.

  • U-013: Revised ratings missing after target date passes
  • U-014: Revised RPN arithmetic incorrect after action
  • D-005: Implausible risk-reduction flag (action claims over 70% RPN drop with no S or O change)
  • D-004 expanded to 3 cases: overdue target date, empty revised values, vague action description
  • Action Effectiveness Summary table in PDF and Word exports

Production scale + delivery

2026-05-15

Engineering scale work to handle real-world FMEA sheet sizes.

  • Up to 500-row file support via item-aligned chunking
  • Prompt caching — significant cost reduction on multi-chunk reviews
  • Project history dashboard (recent reviews + re-download)
  • Column mapping templates (save and reuse mappings across reviews)

Engine validation

2026-05-11

Foundation work — review engine validated against a hand-authored fixture set.

  • 10 hand-authored DFMEA fixtures across casting, fastener, gearbox, and turbocharger domains
  • 4-gate evaluation methodology: coverage, hallucination ceiling, bloat, qualitative
  • AIAG-RPN and AIAG-VDA both auto-detected and verified
  • Tier-2 supplier-blame detection moat: 3 spontaneous catches in 10 fixtures, senior QE verified
  • 7 deterministic rule checks (ruleEngine) with 100% unit test coverage