Trust Meter methodology
The Trust Meter measures, from 0 to 10, the risk of using a project. Each component is scored out of 10; the details of each score are shown on the profile.
Components and weights
| Component | Weight | Computation |
|---|---|---|
| Security | 35% | 10 minus a penalty per open vulnerability on the latest release (minimum 0). |
| Supply chain | 20% | 10 minus 1 per dependency with an open vulnerability (minimum 0). Absent without a dependency graph. |
| Maintenance | 20% | Share of signals present: push within 90 days, release within 180 days, at least 2 people make 80% of commits, CVEs fixed within 30 days (when known). |
| Provenance | 15% | OpenSSF Scorecard when available; otherwise share of signals: older than a year, not archived, owned by an organization, license detected. |
| Quality and performance | 10% | Build, tests and performance measurements in our isolated environment (coming soon). |
A component without data is not counted: the other weights are scaled back to 100%. The score is rounded to a tenth.
Security penalties
- Per open vulnerability by CVSS severity: critical 3, high 2, medium 1, low 0.3 (no score: medium).
- ×1.5 when actively exploited (CISA KEV catalog).
- +0.5 when its EPSS exploitation probability is at least 0.1.
Method v2 (after our scan)
Once a scan of the repository is done, method v2 applies: same components and weights, plus the results of our tools. A project never scanned keeps method v1.
- Security: on top of CVEs, 2 per secret (cap 6, gitleaks); static analysis (semgrep, gosec, bandit): 1 per high finding, 0.3 per medium (cap 4).
- Supply chain: 1 per vulnerable dependency (cap 10, deduplicated sources: GitUP, osv-scanner, grype; syft graph); suspicious binary 1 (cap 3).
- Install scripts: curl_pipe_sh 2, postinstall 1, setup_py_network 1
- When a tool fails, times out or is unknown, its component is marked "partial" and capped at 5 (a failure never raises a score); only the failed tool's penalties are ignored, those of tools that succeeded count. The overall score is then marked "partial" (~, also on the badge).
- Without packages in the syft SBOM (failure, or zero packages), dependencies are "no data": supply chain is scored only when binary or install-script penalties exist (10 minus those), otherwise it is left out as in v1.
Sources
GitHub API (metadata, releases, contributors, dependency graph), OpenSSF Scorecard, and our local copy of the OSV, CVE List V5, NVD, CISA KEV and FIRST EPSS databases, updated daily.