← Back

Guide · Published November 14, 2026 · Reviewed November 14, 2026 · Ingenious AI team · Not legal advice

← All guides

Hash-Chained Decision Records: Why Tamper-Evidence Is the Point of the Record

The short answer: The kit keeps its decision record as a SHA-256 hash chain: every entry — a notice, a decision, an explanation, a correction, a re-run, a review — carries a fingerprint computed from its own contents and the previous entry's fingerprint. Change any earlier entry and every fingerprint after it stops matching: the alteration is detectable, by anyone who can recompute the chain, without trusting the business's say-so. That property — tamper-evidence — is what turns a log into a record that can be shown to an examiner, produced in an export, and believed a year later. It does not make entries true; it makes them unchangeably what they were when they were written.

Check your duties and dates — free Build your first notice — free

What is a hash chain? A sequence of records linked by cryptographic fingerprints (SHA-256 hashes), each computed over its entry plus the link before it — the same family of construction used in version-control histories and audit ledgers. The kit applies it to the ADMT decision record described in How to Keep ADMT Records for Three Years; this guide explains the mechanism and its evidentiary value.

The mechanism in one paragraph, honestly

When an entry is written, the system computes a SHA-256 hash — a fixed-length fingerprint that changes completely if any input changes — over the entry's contents together with the previous entry's hash, and stores that fingerprint with the entry. Entry 50 therefore commits to entry 49, which commits to 48, and so on to the start. Editing entry 12 (a date, a reason, a name) changes entry 12's computed fingerprint; since entry 13 was computed from the original 12, the sequence no longer verifies — and it fails at exactly the edited point, which localises the tampering.

Two properties follow. Detection is independent: verification needs only the entries and the algorithm, so an export can be checked outside the system that produced it. Backdating is visible: inserting a 'missed' notice months later, with an old date, breaks the chain at the insertion point, because the insertion was not there when its neighbours were fingerprinted. The chain thus polices the two edits a compliance record is most tempted by — the quiet correction and the retrospective insertion.

What the chain proves, and the sentence it lets you say

With a verified chain you can say a specific, strong sentence: 'this is the record as it was written, entry by entry, at the times recorded — and here is the verification.' For a record whose purpose is to show compliance to an Attorney-General enforcement frame, to answer a person's question about their own case, or to survive an internal audit, that sentence is the difference between evidence and assertion.

The chain also disciplines prospectively. A team that knows its record is tamper-evident writes entries at the time of the event — because later reconstruction cannot be backfilled invisibly. The technology thus buys the behaviour the duties actually want: notices logged when given, sent dates logged when sent, corrections appended rather than overwritten. Systems shape behaviour; a chain shapes it toward contemporaneity.

Worked example: Ozark Ridge Health Staffing

Example — fictional business, real duty structure. Ozark Ridge, the fictional staffing firm from the records guide, demonstrates what its chain does across one quarter of operation:

What the chain does not prove: truth at entry

Tamper-evidence is not truth. A false entry, hashed promptly, verifies perfectly: the chain guarantees the entry has not changed since it was written, not that it described reality when written. A notice logged as sent that was never sent is a false entry in a perfect chain. This limit is not a footnote — it is the reason the guides pair the chain with capture discipline (generation at the event, per-person logging, reconciliation audits like the opt-out run-log check) and with sampling (the stranger test, the override-rate reviews).

Hold both halves at once: capture discipline makes entries true; the chain keeps them that way. A business with only the chain has a well-preserved account it cannot fully warrant; a business with only discipline has honest records it cannot prove were not later edited. The duties' design — a record that shows compliance — wants the pair.

The chain in the workflows you already run

Notice where the chain quietly does work across this series. Corrections and re-runs: both versions kept, appended in order — the chain is what makes 'keep both versions' verifiable rather than a claim. Review records: a changed outcome appends to the case; the original decision remains untouched beneath it, which is exactly the sequence a reviewer wants to read. Exports: the export carries chain references so the produced case verifies against the live record at production time. Quarterly integrity checks: recompute the chain over the period and file the result — the worked example in the records guide verifies 312 of 312 entries, which is the check's expected shape.

The plans all include the core record log: the chain is not a premium feature, because a record without tamper-evidence is a lesser product category — a log. The comparison with spreadsheet tracking turns substantially on this point, as that guide sets out.

Verifying a chain: the practice, not just the property

A property nobody exercises is a claim, not a control. Build verification into the calendar: recompute the chain on a sample of cases quarterly (or the full period — recomputation is cheap), file the verification result with its date and scope, and include chain verification in every export acceptance test. When verification fails, the failure is an incident with a known location — the break point — and the response is investigation and honest annotation appended to the chain, never a silent rebuild.

If you evaluate any records system — the kit or another — ask the verification question in its operational form: show me a case export, and show me how I check it against the live record, in this room, without your engineers. The answer's quality is the product's quality, stated in one demonstration.

  1. Understand the property precisely: tamper-evidence — alteration and insertion are detectable; entry truth still depends on capture discipline
  2. Write entries at the event (generation, sending, deciding) — the chain exists to make contemporaneity the only available behaviour
  3. Append, never overwrite: corrections, re-runs, and review outcomes join the chain in order, with originals untouched
  4. Verify on a schedule — sampled or full recomputation quarterly, result filed — and verify every export at production
  5. Ask any records vendor the demonstration question: show me an export, and show me how I check it against the live record, unaided

Sources, as carried in the kit's own data: the duty summaries in app/rules.py and the checker data in app/checker_data.py, verified against primary or authoritative sources on October 28, 2026. Run the free duties checker for your jurisdiction, decision type, and the tool's role, and compare the kit's plans for the record, builder, and review queue. More from Ingenious AI is on the hub.

Build your first notice — free

Related guides

See all ADMT guides for the full set.

Frequently asked questions

What is a hash-chained record?

A sequence of entries linked by SHA-256 fingerprints, each computed over the entry's contents and the previous entry's fingerprint, so altering or inserting an earlier entry breaks the sequence detectably.

Does the chain prove our records are true?

No. It proves they have not changed since they were written. Truth comes from capture discipline — entries written at the event, reconciliations and sampling — which is why the kit pairs the chain with per-person generation and logging.

Can a missed notice be backdated into the chain?

Not invisibly. An entry added later carries its position at the end of the sequence; inserting it with an old date breaks verification at that point. The honest practice is to append it now and let the record show the gap and its fix.

How often should the chain be verified?

On a cadence you will actually keep — quarterly sampled or full recomputation is the pattern in these guides — and at every export's production, as part of its acceptance test.

Is the hash chain an extra-cost feature?

No. The core record log is included in every plan, including Free, because the chain is what makes the record a record rather than a log.

Build your first notice — free