Verified ReviewsTrust ledger

Trust ledger methodology: how we anchor review evidence

The full methodology behind the Trust Ledger — what is hashed, what is anchored, how independent parties can verify a review's provenance, and what is deliberately kept off-chain.

Priyanka Rao · Principal Engineer, Trust Infrastructure Published 26 March 2026 14 min read Verified Reviews
Discuss

A trust ledger is only useful if outsiders can verify it without the platform's cooperation. This paper documents the cryptographic scheme behind Trust Ledger v2, the exact fields that are hashed, the anchoring cadence, the verification recipe, and the deliberate omissions that keep personal data off the ledger. It is written for security researchers, procurement teams, insurers and journalists who want to audit the system end to end.

The design goal in one paragraph

The Trust Ledger exists to make one specific claim testable: 'at time T, ScoreReview UK held evidence E for review R, and the moderation state was S'. Nothing more, nothing less. It does not attempt to prove the review is truthful. It proves only that the platform cannot silently rewrite history.

This is a narrow goal on purpose. Trust systems that try to prove too much end up proving nothing because the assumptions collapse. Anchoring a hash of the evidence and the state is a claim we can defend cryptographically and audit publicly.

The design borrows from certificate transparency: append-only log, third-party verifiers, published inclusion proofs. Nothing in the scheme is novel, which is a feature — it means the properties are well understood.

What is hashed and what is anchored

Every verification event produces a leaf hash. Leaves are batched into a Merkle tree every 60 seconds, and the tree head is anchored to two public transparency services on a five-minute cadence.

The consistent pattern is: identifiers and outcomes are hashed; personal data and free text are held internally, salted, and only their hashes appear on the log. This satisfies GDPR's data-minimisation principle while still allowing third parties to verify that no event has been added, removed or altered after the fact.

Salts are per-review and rotated only when a review is deleted for legal reasons (right to erasure). Rotation invalidates the previous inclusion proof but leaves the log structure intact, and the rotation itself is a logged event.

EventFields hashedFields excluded
Invitation issuedreview_id, tier_at_issue, invited_at, saltreviewer email, business email
Evidence uploadedreview_id, evidence_class, sha256(evidence_blob), uploaded_atthe evidence blob itself
Moderation decisionreview_id, decision, moderator_role, rationale_hash, decided_atmoderator identity, free-text rationale
Dispute openedreview_id, dispute_id, opened_by_role, opened_atdisputant identity, matter details
Dispute resolvedreview_id, dispute_id, outcome, resolved_atadjudicator identity
Tier changereview_id, from_tier, to_tier, reason_code, changed_atoperator identity
Fields included in the leaf hash for each event class.

How anyone can verify a review

The verification recipe is public and language-agnostic. It requires the review ID, the ledger anchor timestamp, and access to the two public transparency services — all of which are free.

  1. 1Fetch the review record from GET /api/public/ledger/{reviewId}. The response includes the event list, the leaf hashes and the tree-head anchors.
  2. 2Recompute each leaf hash locally from the disclosed fields using SHA-256. Compare byte-for-byte against the returned leaf hashes.
  3. 3Reconstruct the Merkle inclusion path for each leaf and confirm it produces the disclosed tree head.
  4. 4Fetch the same tree head from the two independent transparency services using the anchor timestamp; confirm both return the same head hash.
  5. 5If any step disagrees, the record has been altered. If every step agrees, the record has not been touched since the anchor time.

What the ledger is deliberately not designed to do

Being explicit about the limits is as important as the guarantees. Confusing the two is where most 'blockchain for trust' projects lose their footing.

The ledger does not prove the reviewer is a real person. That is a job for the verification tier system (see the tiers whitepaper) — the ledger only proves that whatever the platform held as evidence, it held at the claimed time.

The ledger does not prevent the platform from taking a review down for legal reasons. GDPR erasure and DMCC compliance sometimes require removal. What the ledger prevents is a silent removal: every deletion is itself an anchored event.

The ledger does not distribute trust across untrusted validators — that would be a full blockchain, with different trade-offs. It publishes anchors to two independent transparency services, which is enough for the specific claim being made and dramatically simpler to operate.

The ledger does not touch personal data. Nothing on the log can be reversed to expose a reviewer, a moderator or a business identity, even under compromise of the platform database, because those fields are never included in the hashed payload.

The right question to ask a review platform is not 'do you use blockchain' but 'can I, without your cooperation, prove that no review has been altered since publication'. Trust Ledger v2 answers that question in the affirmative.

Prof James Kellner, Applied cryptography, Imperial College London

Summary and next steps

A trust ledger is worth exactly as much as the ability to audit it. Published anchors, a published recipe and an open-source verifier are the minimum bar; anything less is marketing.

For consumers, no action is required — the badge on every verified review already links to the ledger anchor. Clicking it and seeing a tick from the verifier is the whole ceremony.

For enterprise buyers and insurers, the ledger is now a procurement-grade evidence artefact. We supply signed export bundles on request under the standard DPA, and Q3 2026 will bring a self-service export endpoint.

Key takeaways
  • The ledger proves one narrow claim: no event has been silently altered since anchoring.
  • Personal data is never hashed into the log — only identifiers, outcomes and evidence hashes are.
  • Anyone can verify any review in under two seconds with a public open-source verifier.
  • GDPR erasure is preserved through per-review salt rotation.
Recommendations
  • Bookmark /trust/ledger and click through from any verified badge to see the anchor.
  • For procurement, request a signed export bundle under the standard DPA.
  • For research, cite the specific anchor timestamp — it is the durable identifier.
  • For auditors, run the open-source verifier as part of your evidence workflow.

FAQ

Can I run my own verifier?

Yes — the open-source verifier at github.com/scorereview-uk/ledger-verifier reproduces the full recipe. There is no proprietary step.

What happens if one of the transparency services goes offline?

The ledger continues operating on the surviving service, and a third service is provisioned within 24 hours. The recipe requires two independent confirmations, and we publish which two were active for each anchor.

Is personal data ever included in the ledger?

No. Personal data — reviewer email, business email, moderator identity, free-text rationale — is held internally and only its salted hash is included in the leaf. GDPR erasure remains fully possible via salt rotation.

Does the ledger cover reviews from before v2?

Yes. Every v1 review was migrated with a back-dated anchor computed from the evidence stored at v1 verification time. The migration itself is a logged event, with the full set of migrated review IDs anchored on 16 December 2025.

Tags:#trust ledger#methodology#cryptography#transparency#audit
Key focus:trust ledger methodology
Secondary keywords:review transparency, merkle log, review audit, verified reviews cryptography, review provenance, trust ledger v2, certificate transparency, review platform audit

Keep reading

Trending#DMCC#CMA#compliance#UK law#moderation#GDPR#data protection#retention#AI#reply automation#brand voice#sentiment#analytics#hospitality

Discussion (2)

Comments are stored locally on your device for this demo. Be respectful — no spam, no personal attacks.

  • Priya S.· 2 days ago

    Really practical breakdown — the four-part reply structure is now on our till-side crib sheet. Thank you.

  • Dan (Cannock Plumbing)· 5 days ago

    Went from 12 reviews to 47 in three months following almost exactly this playbook. It works.