Skip to content

FAQ

Is approvo a service I run?

No. It is a library you import into your backend. There is no daemon, no HTTP server, no container to deploy, and no database that belongs to approvo. Your backend-for-frontend owns the transport, authentication, and the datastore connection; approvo is the logic in between.

Why doesn't it ship a Postgres / Mongo adapter?

Because a bundled adapter would fight your migration tool, your connection pool, and your transaction boundaries — and it would rot, because the maintainers don't run every database in anger. See Storage for what you implement instead, or ADR-0004 for the full reasoning.

Can I use it for things that aren't software releases?

Yes. kind is a free string and subject is an arbitrary JSON object. Change approvals, data-export requests, privileged-access grants, financial sign-offs, content publication — anything that needs "N of these people said yes, provably". Releases are the design target, not the limit.

What stops the server from forging an approval?

If you use the detached signing ceremony (prepare_decision + submit_signed_decision), the approver's private key never reaches your backend. The server produces the bytes to sign and only ever handles the signature plus the already-registered key material in your KeyDirectory, so a fully compromised backend still cannot manufacture a valid approval.

submit_decision (server-side signing) is offered for internal tooling where that trade-off is acceptable, and its risk is documented.

What happens if someone edits the database directly?

verify() catches it. Editing an entry breaks its entry_hash; fixing that breaks the next entry's prev_hash; deleting or reordering breaks the sequence; rebuilding the whole table diverges from the last signed Merkle checkpoint you pinned somewhere they can't reach. See Concepts and Verification.

Does approvo prevent tampering, or just detect it?

Detect. Tamper-evidence, not tamper-prevention. Pair it with revoked UPDATE/DELETE grants on the ledger table and off-site mirroring. An attacker who can destroy the whole store destroys the evidence — that's what backups are for. Security model is explicit about the boundaries.

How does it scale? Won't a hash chain serialise all writes?

Within one log_id, yes — a chain is totally ordered. So you partition: one log_id per tenant, environment, team, or service. Each is an independent chain with its own checkpoints and its own verify(). Reads are always O(decisions for this request) via the indexed events_for_request, never O(ledger). See ADR-0008.

Why can't I put floats in a subject?

Canonical serialization (ADR-0006) rejects them, because there is no cross-language, cross-version float representation stable enough to hash and sign against. Use integers, or strings for decimals ("19.99").

Can I change a policy after requests are open against it?

Not in place — the request pins policy_digest and get_status raises PolicyMismatch if it no longer matches. Add a new version (release-v2) instead. In-flight requests keep evaluating under the rules they were opened with, which is usually what an auditor wants. See Policies.

Is the audit trail confidential?

No. The ledger is plaintext by design, so it stays greppable and auditable. Keep secrets out of subject and comment fields.

How do I trust verify() if the operator runs it?

You don't — run it yourself, on infrastructure the operator doesn't control, with a key directory you obtained out of band and a checkpoint you pinned. verify() works against any EventStore, including a minimal read-only one you wrap around rows from a database dump. See Verification.

What's the relationship to in-toto / TUF / Sigstore?

approvo borrows their ideas — in-toto's step/functionary/threshold model, TUF's key validity windows, DSSE's envelope, Certificate Transparency's Merkle log — and packages them as a small embeddable library for approvals specifically. If you want hosted transparency infrastructure rather than a library, use Sigstore. See the ADR overview.

What license?

Apache-2.0.