Skip to content

Multi-Source Release Corroboration

Cross-source method — instantiates Transitive Trust Boundary Hardening

Accepts a release only when independent observers agree on the same artifact digest, so no single compromised source, signer, or channel can define what 'the release' is.

A single source — even a signed one — is a single point of failure: whoever controls it defines what "the release" is. Multi-Source Release Corroboration refuses to let any one authority make that claim alone. It requires that independent observers, drawn through different channels and rooted in different trust, agree on the same artifact digest before the release is accepted. Trust here comes from correlation across independence, not from any one signature or mirror. Its defining move is treating disagreement as an alarm: when the sources diverge, that split is itself the detection signal that someone served different bytes to different consumers.

Example

A security team is about to roll out a specific release of a widely-used tool, pinned by its SHA-256. Before deploying, it corroborates that digest across three independent places: the hash published on the vendor's HTTPS site, the same hash recorded in a public append-only transparency log, and the hash reported by an unaffiliated package mirror. All three agree, so the release is accepted. Had the vendor site's hash differed from the transparency-log entry, the team would not merely fail the check — it would treat the divergence as evidence of a targeted, split-view attack and investigate, because a compromise able to fool one channel usually cannot rewrite all of them at once.

How it works

Collect the artifact's asserted identity — its digest — as reported by several independent, differently-trusted observers, and require consensus before acceptance. The security rests on the cost of compromising all the independent channels simultaneously, and the method deliberately elevates disagreement from "a failed check" to "a signal worth chasing." It confirms that observers concur on which bytes the release is; it does not, by itself, decide those bytes are safe.

Tuning parameters

  • Number and independence of sources — two loosely-related observers versus three genuinely independent ones. Real independence, not count, is what buys security.
  • Quorum rule — unanimous agreement versus majority. Unanimity is stricter but noisier when a benign source lags.
  • Source diversity — deliberately varying channel, operator, and trust root so a single compromise cannot move several observers at once.
  • Disagreement handling — block-and-investigate versus block-and-alert; whether a split view pages a human immediately.
  • Staleness window — how fresh each observer's record must be before it counts, closing replay of an old corroboration.

When it helps, and when it misleads

Its strength is defeating a single compromised signer, mirror, or CDN, and — uniquely among its siblings — surfacing targeted attacks where a victim is served different bytes than everyone else. Cross-observer agreement is evidence no single authority can fabricate.

Its failure mode is correlated independence: sources that look separate but share a root — the same upstream CDN, the same publishing pipeline — give false comfort, because one compromise moves them together.[1] And corroboration confirms identity, not safety: every observer can honestly agree on the digest of a legitimately-signed-but-malicious release. The classic misuse is inflating the source count with observers that all derive from the same origin. The discipline: engineer genuine independence into the source set, and always treat disagreement as signal rather than noise to be smoothed over.

How it implements the components

  • independent_channel_cross_check — the mechanism's core: cross-check the artifact's digest across independent channels and require agreement, so no one channel is trusted alone.
  • origin_and_integrity_evidence — corroborated agreement on the digest is origin-and-integrity evidence, produced by cross-source consensus rather than by a single cryptographic signature.

It does not itself produce or verify the cryptographic signature (consumer_side_verification_gate, Artifact Signature Verification); it does not check the build lineage (transformation_and_custody_record, Provenance Attestation Check); and it does not maintain the registry of legitimate publishers (intermediary_role_registry, Package Namespace Confusion Guard).

Notes

Independence is the entire mechanism: five observers behind one CDN are, for security purposes, one observer. When adding a source, the question is never "does it agree?" but "could the same compromise that fools the others also fool this one?" — if yes, it adds confidence without adding security.

References

[1] Certificate Transparency is the working precedent for corroboration by independent append-only logs — misissuance is caught because an entry visible to many independent monitors is far harder to hide than a single private assurance. The security depends on the monitors being genuinely independent. ↩