HVTracker Listing Lifecycle Specification

Version: v0.1 Status: Published Published: 2026-05-26 Authors: HVTracker

1. Abstract

This document defines the lifecycle states for agent listings in the HVTracker registry. It specifies how agents transition between states, the evidence required at each stage, the evidence grade system, and the HVTrust composite score methodology.

The key words MUST, MUST NOT, SHOULD, SHOULD NOT, and MAY in this document are to be interpreted as described in RFC 2119.

2. Motivation

As HVTracker evolves from a leaderboard into a trust registry, agents need a formal lifecycle beyond "tracked or not." A project may be discovered but unreviewed, listed but not yet verified, or previously listed but now delisted due to archival or license changes. The listing lifecycle provides a vocabulary for these states and defines the evidence thresholds for transitions.

3. Listing States

Every agent in the HVTracker registry MUST be in exactly one of the following states:

StateDescriptionVisible on site
discoveredAgent identified but not yet reviewed for eligibility.No
candidateSubmitted for listing (via issue template or discovery). Under review.No
listedMeets eligibility criteria. Tracked with daily signal collection.Yes — full leaderboard entry
verifiedListed + additional manual verification of agent identity and capabilities.Yes — with verified badge (future)
warningListed but flagged for eligibility issues (archived, no license, stale).Yes — with warning indicator
legacyHistorically significant but no longer actively maintained (>365 days inactive).Yes — in legacy section
delistedRemoved from active tracking due to disqualification criteria.No (historical data preserved)
rejectedReviewed and determined not to meet eligibility criteria.No

4. State Transitions

State transitions are triggered either automatically by the daily build or manually by the registry owner.

FromToTriggerType
discoveredAgent identified in universe scan or external sourceManual
discoveredcandidateGitHub issue submitted or owner initiates reviewManual
candidatelistedPasses all MUST criteria in Eligibility SpecManual
candidaterejectedFails eligibility reviewManual
listedverifiedOwner manually confirms identity and capabilitiesManual
listedwarningAutomated eligibility check fires (§4.1.1, §4.2.1, §5.1)Automatic
warninglistedWarning condition resolved (e.g., license added, repo unarchived)Automatic
listedlegacyNo meaningful activity in 365+ daysManual
listeddelistedDisqualification criterion met (§5.1–5.5 in Eligibility Spec)Manual
legacylistedProject resumes activityManual
Current implementation: As of v0.1, only listed and legacy states are used in production. Other states are defined for future use. Transitions are manual; automated state changes will be added in a future version.

5. Evidence Grade

Every listed agent receives an evidence grade (A–D) based on how many independent signal types contribute data for that agent. The grade reflects data coverage, not quality.

GradeSignal types requiredInterpretation
A≥5Comprehensive coverage — GitHub + downloads + trust signals + fingerprints + HN
B4Strong coverage — multiple independent signal families
C3Moderate coverage — GitHub + two additional signal types
D1–2Limited coverage — GitHub only or GitHub + one other source

Signal types counted:

  1. GitHub — always present (stars, commits, freshness, forks, license)
  2. Package downloads — npm and/or PyPI weekly downloads are non-null
  3. Trust infrastructure — OSSF Scorecard score is non-null OR package provenance is present
  4. Behavioral fingerprints — public action tracking returns data
  5. Community signals — Hacker News mention count is non-null

6. HVTrust Score

The HVTrust Score is a composite metric (0–100) that measures trust across five dimensions. It is computed alongside the existing health score but uses different inputs and weights.

DimensionMaxInputs
Activity25Freshness (days since push, linear decay over 180d, max 15) + commit activity (log-scaled, max 10)
Adoption20Stars (log-scaled vs 100k, max 12) + weekly downloads (log-scaled vs 1M, max 8)
Transparency20License present (8 pts) + OSSF Scorecard contribution (scaled to 12 pts)
Safety20OSSF Scorecard (scaled to 10 pts) + package provenance (5 pts) + signed commits ratio (scaled to 5 pts)
Identity15Evidence grade (A=10, B=7, C=4, D=1) + listing status (listed=5, legacy=2)
trust_score = activity + adoption + transparency + safety + identity

Relationship to health score: The health score (stars + freshness + activity + community = 100) measures momentum. The HVTrust score measures trustworthiness. Both are displayed; the leaderboard can be sorted by either. The health score formula is unchanged — HVTrust is purely additive.

v0.1 status: The HVTrust score is experimental. Dimension weights and input scaling may change in future versions. The score is computed daily but does not yet affect default sort order.

7. Versioning and Changelog

VersionDateSummary
v0.12026-05-26Initial publication. Defines 8 listing states, transition triggers, evidence grade system (A–D), and HVTrust composite score (5 dimensions, 100 points).

A. State Diagram

  discovered → candidate → listed → verified
                  ↓           ↓ ↑
               rejected    warning
                              ↓
                           legacy ↔ listed
                              ↓
                           delisted

Arrows indicate valid transitions. Bidirectional arrows (↔) indicate the transition can go either way. The delisted state is terminal in practice but historical data is preserved.