- Runtime vs build
- Runtime components ship inside your product (the CRA's concern). Build components (GitHub Actions, base images) only run in your pipeline — but a compromised one can inject anything into what ships.
- Direct vs transitive
- Direct = you chose it. Transitive = a dependency of a dependency; it enters "via" a direct one. You fix transitive risk by replacing/vendoring the direct entry point.
- Jurisdiction (confidence-scored)
- Where the component's maintainers/governance plausibly sit legally. It is inferred intelligence, never asserted fact — the confidence shown is the engine's. Other is a known country outside the EU, US, UK/adequacy and China (rest of the world). Unknown means no signal at all.
- Maintainers
- two separate facts, shown together: how many people contribute (headcount, capped and shown as 99+), and how concentrated that work is — solo (only one contributor exists), dominant (one person does nearly all the work even though others contribute), or spread (no single dominant contributor). "1 of 40 · dominant" and "1 · solo" are different risk profiles that used to collapse into the same number. API field bus_factor is unchanged, kept for back-compat. how it's derived →
- Activity
- the component's last known release date (when available — otherwise honestly "unknown", never guessed) plus an age label: active, slowing or stale (inactive for a long time).
- single
- active maintainers ≤ 1 (either the only contributor, or one person doing effectively all the work even if others are listed). Compromise or abandonment of that one person breaks you (the xz-utils backdoor pattern).
- stale
- no maintainer activity for a long time. Security fixes are unlikely to arrive.
- abandonment-risk
- a solo/dominant maintainer and slowing-or-stale activity together: concentrated and cooling off at the same time, the combination "1 · critical" used to hide.
- relicense
- a license with restrictive-relicensing history (Elastic, BUSL, SSPL). Redistribution terms can shift under you; review for CRA.
- handoff
- documented maintainer handoff or compromise history for this component (the tj-actions compromise pattern).
- The jurisdiction vector
- One package, four answers: legal control (who can be compelled — the main score), distribution (which registry it transits — npm/PyPI/Go proxy are all US-controlled), hosting (where the source lives), and maintainer geography. They routinely diverge: EU-authored code, US-distributed.
- Floor vs ceiling
- The US share is the *confirmed* figure; the ceiling shows what it becomes if every Unknown resolves to US. Honest bounds instead of false precision.
- AI toolchain (declared tier)
- Agent config files (CLAUDE.md, copilot-instructions…) and AI co-author commit trailers make the model provider a build-chain supplier with its own jurisdiction. Declared evidence only — no statistical guessing about who wrote which line.
- disputed
- strong provenance signals point to different jurisdictions (e.g. US entity, EU maintainers). Reported as a finding with reduced confidence, never silently averaged.
Full methodology and definitions: open the documentation →