Product line 02

A chain of custody that does not depend on anyone’s word.

A custody form records that a transfer was claimed. A hash chain records what the artefact was. Our ledger makes the second kind of statement: append-only, Merkle-batched, anchored to a permissioned blockchain, and verifiable by a party who has no access to the case and no reason to trust the laboratory that produced it.

Why a ledger at all

Tamper-proof is a marketing word. Tamper-evident is an engineering one.

No storage system can stop a sufficiently privileged administrator from altering bytes. What a properly constructed ledger does is make that alteration impossible to hide: the altered record no longer hashes to the value its successor committed to, and that successor was witnessed outside the system before the alteration happened.

That distinction is the whole design. We do not claim evidence cannot be changed. We claim that if it is changed, the arithmetic will say so — in a way that can be demonstrated to a court, on a laptop, by someone else.

The defensible claim is not “nobody touched this”. It is “here is the computation that shows nobody could have touched this without leaving a mark”.

Construction

How the chain is built.

Four layers, each doing one job. The chain gives ordering, the Merkle tree gives efficient batching, the anchor gives external witness, and the signature gives attribution.

  1. Per-event hashing

    Every custody event — intake, transfer, access, analysis, export, verification — is canonicalised and hashed with SHA-256. The record includes the digest of its predecessor, so the sequence is fixed at the moment it is written.

  2. Append-only chaining

    Each record commits to the one before it. Removing or reordering any record invalidates every record after it. The chain is offline-capable: it continues to extend and self-verify without any network.

  3. Merkle batching and anchoring

    Records are batched into a Merkle tree and the root is committed to a permissioned ledger. One commitment witnesses an arbitrary number of records, and each record retains an inclusion proof against that root.

  4. Signature and attribution

    Packages are signed with detached signatures verifiable offline using standard tooling, and the platform implements NIST-standardised post-quantum signature and key-encapsulation schemes alongside classical ones.

  5. Continuous re-verification

    Independently of any user action, stored bytes are pulled back from both backends, re-hashed from scratch, and compared against the sealed digest and the anchored record. Divergent artefacts are quarantined rather than silently served.

SEQ 0146SEQ 0147 SEQ 0148SEQ 0149 h(prev|rec)h(prev|rec) h(prev|rec)h(prev|rec) h(0|1)h(2|3)MERKLE ROOT ANCHOR → PERMISSIONED LEDGER root committed · per-record inclusion proof retained PUBLIC VERIFIER recompute digest → check inclusion proof → compare anchor no account · no case data exposed · rate-limited

Fig. 04 — chain, tree, anchor, verifier

SEQ 0147 Evidence intakesha256 342b0a4212e73014b7654e20236b4b4fcdd65866608bbe6007e11f155d56a269 SEALED
SEQ 0148 Custody transfer — examinerprev 342b0a4212e73014b7654e20236b4b4fcdd65866608bbe6007e11f155d56a269 CHAINED
SEQ 0149 Verification — recomputed, matchprev 181555434a1652c62314ff6e2577e1b6b9568051e873e76afbe4e7ac381180f3 VERIFIED

Fig. 05 — three consecutive records

Cryptographic primitives

What is actually computed.

Specific, checkable, and chosen so that the verification path needs no proprietary software.

Primitive selection by function
FunctionPrimitiveWhy this one
Artefact digest SHA-256, SHA3-256, MD5, SHA-1 Four digests at intake. The modern two carry the integrity claim; the legacy two are retained because existing case records, hash sets and third-party tooling still speak them.
Custody chaining SHA-256 over canonical records Deterministic canonicalisation means an independent implementation can reproduce the digest byte-for-byte from the published record.
At-rest encryption AES-256-GCM Authenticated encryption with a unique per-file data key, wrapped by a key-management service; plaintext keys are zeroed after use.
Post-quantum KEM ML-KEM-768 NIST-standardised key encapsulation, so material sealed today survives a future capable adversary rather than relying on classical assumptions alone.
Post-quantum signature ML-DSA-65 NIST-standardised lattice signature. Evidence outlives litigation timelines; signatures on it should outlive the cryptography they were made with.
Package signature RSA-2048 / SHA-256, detached Court packages carry a detached signature verifiable offline with standard command-line tooling and no product installed.
External witness Merkle root on a permissioned ledger Batched commitment with per-record inclusion proofs, so one on-chain write witnesses many records without publishing any of them.

The part that matters

Don’t take our word for it.

A verification system whose verification step requires our cooperation is not a verification system. The check has to be performable by the other side.

Public, account-free

Scanning an evidence QR sticker — or opening the verifier directly — confirms an item’s digest and custody timeline with no login. It is rate-limited and returns integrity status without exposing case-internal content.

Offline-verifiable packages

Court packages ship with detached signatures that verify using standard open-source tooling. Opposing counsel does not need our software, our servers, or our permission.

Reproducible by third parties

Canonical record formats and published primitive choices mean an independent expert can write their own verifier and get the same answer — which is the only kind of verification worth arguing in front of a judge.

Beyond the case file

The same ledger, other kinds of record.

Audit and inspection records

Assessment findings, configuration snapshots and remediation evidence sealed at capture, so a later dispute about what a system looked like on a given date has an arithmetic answer.

Regulated document lineage

Submissions, approvals and revisions chained in order, with each version’s digest witnessed externally — useful wherever “which version was in force?” is a contested question.

Sensor and telemetry provenance

Field-captured streams sealed at the point of capture and anchored in batch, so a reconstruction built from them later can prove which readings it was built on.

Model and dataset provenance

Training inputs, transformations and outputs recorded as a chain, so an analytical conclusion can be traced back to the exact material that produced it.

Have a record that will be contested?

We are happy to walk through the construction in detail, including where it is weaker than the marketing of comparable systems suggests theirs is.