Changelog

Methodology updates, new features, scoring changes, and registry milestones.

2026-05-31
Methodology v3.1 — fair scoring & honest presentation methodology
  • Adoption weight doubled (10 → 20). Real-world usage is a meaningful trust signal — widely-used software receives more scrutiny. Safety 30→25, Identity 20→18, Transparency 20→17, Maintenance stays at 20.
  • License type detection: every project is now classified as Open, Source-available, Proprietary, or Unlicensed. Proprietary tools show an explanatory note rather than a misleading low-trust framing.
  • Download sources expanded: Docker Hub pulls and VS Code Marketplace installs now count toward adoption alongside npm/PyPI/crates.io.
  • Evidence grade reworked: now reflects trust score bands (A ≥ 80, B ≥ 65, C ≥ 50, D < 50) instead of signal-type count. Grade and rank now tell the same story.
  • License filter on leaderboard: filter by Open source / Source-available / Proprietary to compare like with like.
2026-05-26
Phase 3: Badges, comparison, changelog feature
  • SVG badges for every agent — embed ![HVTrust](https://hvtracker.net/badge/{slug}.svg) in READMEs
  • Agent comparison page — compare 2–3 agents side-by-side at /compare?a=langchain,crewai
  • This changelog page for tracking methodology and registry changes
2026-05-26
Phase 2: Trust trends & API events feature
  • 7-day trust score trend indicator (▲/▼) on the leaderboard
  • recent_events field added to /data/latest.json per agent
  • trust_score and evidence_grade added to history data points
  • Fixed push failures in scorecard-scan and discover-agents workflows
2026-05-26
v1 Launch feature data
  • 171 active agents across 14 categories
  • HVTrust score: 5-dimension composite (Activity 25 + Adoption 20 + Transparency 20 + Safety 20 + Identity 15 = 100)
  • Evidence grades A/B/C/D based on independent signal coverage
  • Staggered batch builds — 6 batches every 4 hours, full refresh in 24h
  • Build integrity report at /data/build_report.json
  • Agent submission via GitHub issue template
  • 7 formal specifications published
  • Open data API under CC BY 4.0
2026-05-24
Methodology v2.2 — HVTrust scoring methodology
  • Replaced single health score with 5-dimension HVTrust composite
  • Added transparency dimension (license + docs + OSSF Scorecard)
  • Added safety dimension (OSSF + provenance + signed commits)
  • Added identity dimension (listing status + evidence grade)
  • Old health score preserved as score field for backwards compatibility
2026-05-23
Initial build — 171 agents catalogued data
  • First full signal fetch for all 171 active + 7 legacy agents
  • GitHub stats, npm/PyPI downloads, OSSF Scorecard, provenance, HN mentions
  • Reputation event derivation from history snapshots