ReviewDesk guide · By the Ingenious AI team · Published 14 November 2026 · Reviewed 14 November 2026
What is a hash-chained decision record — and why does immutability matter as evidence?
Generate a free adverse-outcome letter — no sign-up Build a free pre-use notice See pricing
The short answer. A hash chain links each record to the previous one with a cryptographic fingerprint: change any earlier record and every later fingerprint breaks, making tampering detectable. ReviewDesk's decision traces are immutable and hash-chained for exactly that reason — the evidence pack you show a reviewer is the record as it was written, and corrections arrive as new records that reference the originals, never as silent edits.
What is a hash-chained decision record?
A hash-chained record is a log entry that contains a cryptographic hash of the entry before it. Because a hash changes completely when its input changes by even one character, any alteration to history breaks the chain at that point and is visible to verification. In ReviewDesk, decision traces form such a chain, giving the audit trail tamper-evidence rather than mere storage.
How does a hash chain actually work, in plain terms?
Imagine each record sealed in an envelope, and the seal stamped with a summary of the previous envelope's contents — a summary that is trivial to check and practically impossible to fake. Open any envelope and reseal it with altered contents, and its summary no longer matches the stamp on the next envelope. You do not need to know what was changed to know that something was. That is the whole mechanism: hashes are the summaries, chaining is the sequence of stamps, and verification is walking the chain checking each link.
The mathematics underneath — a hash function producing a fixed-length fingerprint from arbitrary input — has one property doing all the work: it is one-way and avalanche-prone. One-way, because the fingerprint reveals nothing about the content to anyone who does not already hold it. Avalanche-prone, because changing a comma changes the fingerprint beyond recognition. Those two properties together give you a log whose integrity can be proven to a third party without exposing the log's contents in advance.
Why does immutability matter for decision evidence specifically?
Because decision records are most valuable exactly when someone has an incentive to improve them retrospectively. A complaint lands in June about a decision made in February; the file as it exists in June is the evidence — and if the system permits quiet editing, the file proves only what the organisation currently wishes it said. Immutability converts the record from a claim into an exhibit. The organisation cannot fix the past, which is precisely why its account of the past can be believed — including by itself, six months later, when staff have changed and memories have politely rearranged themselves.
Immutability also protects the innocent explanation. Most record discrepancies are accidents — a sync failure, a misfiled correction — and a chain that shows where the sequence broke lets you localise the accident instead of impugning the whole archive. The evidence engine's pack discipline, described in the evidence-pack guide, assumes chained traces as its foundation: projections are only as trustworthy as the immutability of what they project.
How are corrections handled if nothing can be edited?
By appending, with reference. The original trace stands. A correction trace is filed containing the corrected element, the reason, and a pointer to the original it amends. Readers see both — the original decision and its repair, in chronological order, each timestamped. This mirrors how ReviewDesk treats data correction for the person affected, in the correction guide: the history of an error, honestly kept, is more persuasive than a history with the error airbrushed out, because real operations have errors and only fictional ones do not.
The same pattern governs every layer: audit-trail entries correct by superseding entries; letters are never re-issued under the same number; a re-run assessment produces a new decision record linked to the old. New staff sometimes experience this as rigidity. It is better understood as the system remembering so that people do not have to perform memory under pressure.
What can a hash chain not do?
It cannot make a false record true. If a trace files factors the tool never used, the chain will preserve that falsehood immutably and detectably — garbage in, gospel out is the failure mode of every integrity system. The chain proves the record was not altered after filing; the truth of the filing itself comes from instrumentation discipline: traces generated by the decision pipeline at decision time, not typed up afterwards. It also cannot prove completeness on its own — a decision never filed leaves no gap in the chain — which is why daily reconciliation between your decision log and the trace count, covered in the integration guide, is part of the design rather than an optional extra.
And it cannot substitute for access control or retention. A chain everyone can rewrite wholesale — by restarting it — is protected only if the chain's head and history are held somewhere rewriting is itself visible. ReviewDesk holds traces on the platform side, sealed per pack, so the customer's own administrators cannot quietly restart history either. Integrity is a system property, not a single algorithm.
How would a reviewer verify a pack?
By re-computation. The verifier takes the traces in order, recomputes each hash, and checks each link against the next record's stored reference, confirming the pack's seal matches the chain's state. Any break identifies the region of alteration. In practice the reviewer rarely needs to go that deep — the pack presents its seal, the chain verifies cleanly inside ReviewDesk, and the human work is reading the contents. But the verification must be possible, on demand, or the immutability is a marketing word rather than a property.
Keep verification boring. Evidence systems fail socially when checking them requires the vendor's professional-services team and a maintenance window. A chain that a customer's own engineer can verify from an export — formats documented, order explicit, hashes recomputed with standard tools — is evidence the customer owns. Possession matters: evidence you cannot independently check is evidence you are merely renting.
Where does the chain sit in the wider record system?
The traces are the decision layer. Around them sit the workflow's other records — the notice versions, the letters, the inbox requests and outcomes, the audit trail events — each timestamped, each part of the three-year file described in the audit-trail guide. The chain hardens the layer where retrospective temptation is strongest — what the tool decided and why — while the surrounding trail supplies the human context: what the person was told, what they asked, what a reviewer did.
Together they answer the two questions every enquiry eventually asks. What did the machine do? The chained traces answer, verifiably. What did the organisation do about it? The trail answers, timestamp by timestamp. A programme that can answer both, from records, in an afternoon, has very little to fear from being asked.
Your five actions for a hash-chained decision record
- File traces from the decision pipeline at decision time — the chain proves integrity, not truth, so the filing must be honest.
- Correct by appending referenced records; never edit, re-issue or backdate history.
- Reconcile decision counts daily — the chain cannot reveal a decision that was never filed.
- Keep verification independent: exports in documented formats your own engineer can re-check.
- Pair the chained traces with the surrounding audit trail so machine facts and human actions read as one file.
Worked example: the altered spreadsheet that a chain would have caught
Consider a fictional screening provider — not a ReviewDesk customer — keeping decision records in a shared spreadsheet. After a complaint, an operations lead 'tidies' three rows from February, aligning the recorded factors with what the vendor says the model probably used. Nobody can later prove which rows changed, when, or from what — so every row in the file becomes suspect, including the thousands that were accurate. The provider's honest records are destroyed by three dishonest minutes, and undetectably so.
Under a hash chain, the same tidying breaks every link after the first altered row, and verification names the region immediately. The three rows become a locatable incident instead of a contagion. That asymmetry — cheap to verify, expensive to fake, precise about damage — is why ReviewDesk chains its traces rather than merely storing them, and why the evidence packs can be sealed with a straight face.
Frequently asked questions
What is a hash, in one sentence?
A cryptographic fingerprint of a record — fixed-length, impossible to reverse into the content, and completely different if the content changes by even one character. Chaining records means each one stores the fingerprint of the record before it.
If records are immutable, how are mistakes fixed?
By appending a correction record that references the original and states the corrected element and the reason. Both records remain visible in order — the history shows the error and its repair.
Does a hash chain prove a record is true?
No. It proves the record was not altered after it was filed. Truth comes from filing discipline — traces generated by the decision pipeline at decision time — and completeness comes from reconciling counts against your own decision log.
Which plans include hash-chained traces?
The One-Trace, Three-Regime evidence engine, with immutable hash-chained traces, is a Pro and Team feature. The timestamped audit trail itself is on every plan including free.
Is this guide legal advice?
No. ReviewDesk is templates and workflow tooling, not legal advice. The mechanism is explained as built; consult counsel for your evidence obligations.
Keep reading: How do you run the scope test: does your system materially influence or substantially replace human decision-making? · How should you prepare EU Annex III evidence for automated decisions — before the duties apply? · How do you let someone correct wrong data after an adverse automated decision? · How do you write principal reasons for an adverse decision that a person can actually use? · What must you tell job applicants when a hiring algorithm screens them? · All ReviewDesk guides · ReviewDesk home · Try the free desk · Free letter generator · Free notice builder · Pricing
ReviewDesk is built by Ingenious AI. Templates and workflow tools — not legal advice.