The ten tests
TEST 01Create a real incident
Produce a controlled event: a temperature excursion, a geofence exit, a door open, a hard brake.
Expect: the record is sealed on arrival, carries both the asserted event time and the receipt time, and exports as a ProofPack that passes the verifier with exit code 0. The pack's Evidence Trust Assessment lists the checks that ran, including the ones that found nothing.
TEST 02Unplug a device
Disconnect a reporting device and wait past its expected reporting interval.
Expect: the portal marks the device offline against its expected interval; the gap is visible, not papered over. A ProofPack contains exactly what arrived: absence shows up as absence. Honest boundary: per-pack expected-versus-reported completeness reporting is on our roadmap, and we will say so to your face rather than imply it exists.
TEST 03Delay a transmission
Hold an event back and deliver it late.
Expect: the record carries the asserted event time and the actual receipt time as separate facts, and the assessment flags skew or staleness beyond tolerance as an enumerated entry. Nothing pretends the event arrived when it did not.
TEST 04Backfill an old event
Send an event dated well in the past.
Expect: it seals with its honest timestamps and a staleness entry in the assessment. The chain position records when it actually arrived; backdating the chain is not possible without breaking verification.
TEST 05Send a duplicate
Deliver the same event twice, as a retrying webhook would.
Expect: the API recognizes the external event id and answers with the original record. Counts stay right; no phantom second incident.
TEST 06Tamper with a ProofPack
Export a pack, change one character anywhere inside a sealed record, and verify it.
Expect: verification fails with exit code 1 and names the failing step, both in the command-line verifier and the drag-and-drop verifier. There is no edit small enough to pass verification.
TEST 07Present a wrong or revoked key
Sign events with a key we have never seen, or revoke an enrolled key and keep signing with it.
Expect: the record is preserved with an honest status, unverifiable or signature-invalid, and is never presented as Source-signed. A revoked tenant signing key stops signing entirely, fail-closed, rather than silently borrowing another key. A forged pack under an unknown key verifies as internally consistent at best, never as a PriviNet record.
TEST 08Reconcile against your own system
Compare every record in Lumra with your source platform's log, record by record.
Expect: every record carries your external event id, so the diff is mechanical. Honest boundary: sealing what arrived does not prove nothing was omitted upstream, which is exactly why this reconciliation is in the script.
TEST 09Verify with PriviNet gone
Export your packs, block privinet.net at your firewall, and run the verifier.
Expect: every pack verifies identically. The verifier is a single file using only the Python standard library and contacts no PriviNet system; the drag-and-drop page works saved to disk. Your evidence does not depend on our existence.
TEST 10Bring your own expert
Hand a pack, our published keys, and the pack's own verification instructions to an independent forensic professional or your insurer's claims examiner.
Expect: they can check every signature, digest, and timestamp against public standards (SHA-256, Ed25519, ECDSA P-256, RFC 3161) without asking us anything. We would rather earn their skepticism than your faith.
What these tests prove, and what they do not: a passing run shows records cannot be altered, backdated, reordered, or misattributed after sealing without detection. It does not prove a sensor was accurate or calibrated, and we never claim it does. Sealing preserves what your systems reported; whether the report was true is exactly the question the preserved record lets both sides argue from the same facts.
Run the script against us.
Every free 60-day pilot includes all ten tests, on your devices, with our help setting up the hostile cases. Start the pilot, or send this page to the most skeptical engineer you have.
Start your free pilot