← Back

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

← All guides

The ADMT Quarterly Review: Running the Register, Audits, and Re-Tests on a Cadence

The short answer: Every artefact in this series has a decay rate: registers stale as tools change, negative findings expire as roles drift, exception conditions lapse as rotas change, assessments age as vendors ship updates, templates drift as flows are redesigned. The quarterly review is the countermeasure — a fixed cadence that re-runs the checker on changed combinations, re-tests role and exception findings, verifies the hash chain, samples exports with the stranger test, reviews the monthly audit numbers in aggregate, refreshes assessments where triggers fired, and re-reads the forward calendar. This guide assembles the cadence from the individual practices the earlier guides prescribe, into one repeatable quarterly sitting with a written output.

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

What is the quarterly review? It is the programme's maintenance sitting — distinct from the monthly audits (operational numbers) and from incident response (event-driven). Its output is a dated review record: what was re-tested, what changed, what converted (findings whose conditions failed and duties that re-activated), and what is diarised next. The register it maintains is the one built in How to Inventory Automated Decision Tools.

Agenda item one: the register diff

Open with what changed in the estate: tools added, removed, updated (vendor material-update notices received are the input), decision types launched or retired, states entered or exited, flows redesigned (placement re-tests owed). Each change maps to register rows created, updated, or closed — with checker runs for new or changed combinations filed the same day. A quarter with 'no changes' should still produce the diff document saying so, because the claim was checked; silence is not a diff.

The diff is also where shadow tools get their recurring chance to surface: the standing amnesty question to team leads, repeated quarterly, catches the tool adopted in month two of the quarter while it is still cheap to govern.

Agenda item two: findings re-tests (roles, exceptions, thresholds, lists)

Work the expiry list: every negative finding and exception position carries a re-test date or tripwire, and the quarterly sitting processes the due ones. Role findings re-test against fresh override data; exception conditions are verified live (is the rota staffed today? does the fraud score still feed one decision? is the evaluation younger than the tool's last update?); threshold positions re-measure against the quarter's numbers; listing boundaries re-check output mappings.

Record conversions without drama: a finding whose conditions failed converts to the duty set, the conversion is filed with its date and cause, and the build tasks are scheduled. The conversion rule working as designed is the cadence's success case — the Pioneer Valley example in the negatives guide shows one conversion and re-conversion filed as routine events. A programme whose findings never convert is either very stable or not really re-testing.

Worked example: Fairhaven Consumer Finance

Example — fictional business, real duty structure. Fairhaven, the fictional California lender from the opt-out and checker guides, runs its quarterly review in a half-day sitting. One quarter's record, in summary:

Agenda item three: record integrity and the export sample

Verify the chain over the quarter's entries (full recomputation where volume allows; the expected result is total verification, and any break is an incident with a location). Draw the export sample — the pattern in these guides is 25 cases from the quarter — run the stranger test, and file the narration results with any template fixes they generate. Check retention housekeeping: cases reaching their 3-year horizon are handled per the retention position, neither early-deleted nor silently immortal.

This item also confirms the one-hour test still passes on a fresh case. Programmes pass it at launch and fail it two quarters later as fields quietly stop being captured; the quarterly re-test is what makes the failure a finding rather than a discovery during a person's case.

Agenda item four: the numbers, in aggregate

Review the quarter's monthly audits together: explanation clock performance (sent inside 30 days, median, exceptions with causes), opt-out reconciliation exceptions, review queue metrics (requests, median days, changed-outcome rate as a diagnostic), correction patterns by kind and source, access answer completeness sampling. Aggregated quarterly, these stop being operational noise and become the programme's quality evidence — and its improvement list, since every recurring pattern names an upstream fix (a data source, a template sentence, a rota gap).

Feed the aggregates upward: the quarterly review record should be readable by leadership in five minutes — state of the duties, incidents and fixes, conversions, next quarter's dates. The enforcement frames in the kit's data reward exactly this artefact: a programme that examines itself, on paper, on a cadence.

Agenda item five: assessments, templates, calendar

Refresh what the triggers touched: risk and data-protection assessments re-issued where purposes, tools, or data changed; templates re-passed through the plain-language tests where they were edited; placement phone tests re-run where flows changed; the vendor files updated with the quarter's answers and update notices. Then re-read the forward calendar aloud: the next fixed dates (state starts, the EU gate), the pending-items watch (Colorado's AG rules until they land), and every re-test and renewal falling in the next quarter, each with its owner confirmed still in post.

Close by issuing the review record and setting the next quarter's date before the room disperses. Cadences die by postponement, not by decision; a dated next sitting, owned, is the difference between a programme and a project. The cadence sits on top of the operating system the plans support — the record, tracker, and queue whose outputs the review consumes.

  1. Diff the register against the quarter's reality — new, changed, retired tools and flows — and file checker runs for every changed combination
  2. Re-test every due finding: roles against fresh override data, exception conditions live, thresholds re-measured, boundaries re-mapped
  3. Verify the chain, sample exports through the stranger test, and re-pass the one-hour test on a fresh case
  4. Read the quarter's audit numbers in aggregate and convert recurring patterns into named upstream fixes
  5. Refresh assessments and templates where triggers fired, re-read the forward calendar with owners confirmed, and date the next sitting before dispersing

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 happens at the quarterly ADMT review?

Five things: the register diff, findings re-tests (with conversions recorded), chain verification and the export sample, the quarter's numbers in aggregate, and assessment/template/calendar refresh — closing with a dated review record and the next sitting scheduled.

How is this different from the monthly audits?

Monthly audits are operational: opt-out reconciliation, clock performance, queue numbers, correction patterns. The quarterly review maintains the programme's foundations — scope findings, assessments, integrity, and the calendar — and reads the monthly numbers in aggregate.

What is a conversion?

A finding whose conditions have failed, converting back into the duty set: a lapsed exception position re-activates the opt-out; a drifted role re-activates the full duty set for that tool. Conversions are filed with dates and causes — they are the cadence working, not failing.

Who should attend?

The register owner, the review-queue owner, whoever owns vendor relationships, and someone able to commit fixes. Leadership receives the five-minute review record rather than attending the working sitting.

What if a quarter shows no changes?

File the diff saying so, with the checks that support it. A verified no-change quarter is a result; an unexamined quarter is a gap wearing the same clothes.

Build your first notice — free