SeenRelay
VERIFY BEFORE YOU TRUST

Do not take SeenRelay on faith.

A freshness preflight sits close to validation, provider spend and agent decisions. That deserves scrutiny. SeenRelay is designed so you can begin with almost no trust: inspect the client source, install a zero-runtime-dependency package, run in shadow mode, keep your original validation path, and verify the live deployment and release evidence yourself.

Inspect the sourceMachine-readable postureData practices
MINIMAL BLAST RADIUS

Start with a wrapper that is easy to remove.

Trust should be earned after integration, not required before it.

Zero runtime dependenciesThe public JavaScript/TypeScript and Python client libraries do not pull a dependency tree into your application.
Shadow mode firstCHECK can run while every original validation still executes. No reuse is required until your own evidence and policy justify it.
Fail openIf SeenRelay is unavailable, the wrapper falls back to the validation your application already intended to perform.
No lock-in pathRemove the preflight and keep the original validation function. Leaving does not require exporting SeenRelay-owned application state.
SOFTWARE SUPPLY CHAIN

Verify what you install and how it was released.

The client source is public and MIT-licensed. Automated package publishing is configured for registry Trusted Publishing/OIDC instead of long-lived publication tokens. Registry provenance or attestations should be checked for the exact version you install; a release process is evidence, not a substitute for reviewing the package.

Current client: v0.1.0 · 0 runtime dependencies.

npm install seenrelay

# or

pip install seenrelay

Production commit:
GET https://seenrelay.com/healthz
RELEASE BOUNDARY

Production should be reproducible from enforced evidence.

Pull request -> required CI verify -> required isolated Preview Release Gate -> main -> Production.

Required merge gatesThe active main-branch ruleset requires a pull request, the verify check and the isolated preview-release-gate. Inspect the linked ruleset instead of trusting this sentence.
Dependency securityRequired CI audits the complete locked npm dependency set at high severity. CodeQL provides additional analysis. GitHub dependency-delta review is attempted on pull requests when its API is available; an unavailable API is surfaced explicitly, not called a completed review.
Live commit/healthz exposes the Production deployment SHA so an integrator can compare what is live with the public repository.
Security reportingSecurity-sensitive reports use the private vulnerability path described in SECURITY.md, not public exploit disclosure.
SERVICE BOUNDARY

SeenRelay reports observations. It does not ask you to outsource truth.

The service does not browse the fact source or use an LLM to decide truth. SAME_OBSERVED is evidence that matching observations exist within the requested window; whether that evidence is reusable remains your policy decision.

Billing is disabled in the current deployment. Current public access is free and requires no account. The current client cannot silently create a charge on your behalf.

Exactly two operationsCHECK and OBSERVE.
No source credentialsCredential-bearing and signature-bearing source URLs are rejected from shared fact identity.
Bounded evidenceObserver proofs establish key possession/continuity, not truth or one-key-one-organization identity.
Emergency controlsRuntime modes can reduce capacity, stop new observations, or freeze CHECK/OBSERVE at the application layer.
WHAT WE DO NOT CLAIM

Transparency includes the missing assurances.

No third-party security certification is claimed. SeenRelay currently exposes self-attested, externally inspectable technical controls and public release evidence. A clean automated scan or provenance record does not prove that software is vulnerability-free. Evaluate it under your own risk model and begin in shadow mode.
VERIFY IT YOURSELF

Inspect first. Measure in shadow mode. Reuse only when your policy accepts the evidence.

Run a shadow pilotSecurity policyThreat model