Prove an observation is authentic - without an account or database access.
A file on disk or a row in a database carries no built-in proof that it was not edited, and no way for an outside party - a customer, coalition partner, funder, auditor or agency evaluator who was not there - to confirm who produced it. "Tamper-proof" is an assertion; a forger can even re-sign a doctored bundle with their own key so it looks internally valid. Disputes then become he-said-she-said and evidence gets discounted.
Level Zero closes that gap. Every observation - an EO/SAR scene, a downlink frame, a live track state, an operator report - is normalised into one canonical JSON product, cryptographically signed over its canonical hash and appended to a hash-chained, append-only chain of custody enforced at the database layer, so it can't be silently altered. Anyone can then re-check any bundle themselves through the independent public verifier below - no account, no session, off your database and console. The published JSON Schema is the contract; an OGC SensorThings 1.1 alignment crosswalk maps the canonical fields for interop tooling - a mapping table for review, not a certified SensorThings server.
OGC SensorThings alignment crosswalk
Timestamp
UTC timestamp for every canonical product.
Source
Source type and how it was sourced (synthetic, replayed or live).
Geometry
GeoJSON Point/Polygon, spatially queryable.
Quality flags
Machine-readable flags, e.g. flagged as spoofable.
Confidence
0.0-1.0 confidence on the record.
SensorThings
Crosswalk mapping the canonical fields to OGC SensorThings 1.1 (not a certified server).
Verify it yourself - no account, no trust required
When you hand an evidence bundle to someone who does not trust your word and has no login on your platform, they need to check it themselves. The independent public verifier is a no-account HTTP endpoint pinned to a published trust anchor. It returns exactly one of verified, tampered or untrusted_key. The trusted key is loaded server-side from the platform's published anchor and is never taken from the request, so a bundle re-signed with an attacker's own fresh key is internally self-consistent yet returns untrusted_key - rejected, not trusted. That case is locked by a unit test. The trust anchor itself is published as its public key plus a short fingerprint, so a third party can pin the check independently, off your database and console.
POST /api/level-zero/public/verify
{
"product": { "...": "a canonical lz.canonical.v1 record" },
"signature": "ed25519-signature",
"recordHash": "..."
}
200 OK
{ "status": "verified | tampered | untrusted_key" }
Ingest & transactions
Inputs are ingested in one step and normalised into canonical products. The chain of custody is append-only: updates and deletes are blocked at the database layer, so the catalog cannot be silently rewritten.
Performance
Each tenant's data lives in its own isolated database schema - a physical boundary, not a shared table filtered by a query that one bug could leak past. The store is PostgreSQL + PostGIS, self-hosted on a single server you control (designed to run France-hosted, with no hyperscaler dependency), with role-based access control, audit logging and single sign-on (SSO). Corroboration candidates come from a real spatial gate: within 75 km and a 30-minute window.