BY-G7 · Fraud Prevention · Foundation · 40 min

Detected is not actual

A governed response in the ai.heart canon · printable worksheet + reveal · 15 min

1 · The situation

A Commonwealth entity's finance team reports a small confirmed fraud loss in accounts payable each year, drawn from its exception report and duplicate-payment review. On that basis, resourcing for payment assurance has been held flat for three years. A new finance director commissions a fraud loss measurement over a sample of payments, examined in full regardless of whether anything flagged them. The measured estimate of combined fraud and error is several times the detected figure, with a wide range and a number of transactions marked unable to determine because records had aged out under the retention schedule. The audit committee meets in three weeks.

2 · Your decision

Take a position — or defer with reasons — before you reveal the governed response.

A · Report the measured estimate to the audit committee — as the entity's best available figure, alongside the published detected figure, with the range, scope, period and the undeterminable cases shown.
B · Hold the measured estimate back this cycle — and repeat the measurement first — a single point estimate with a wide range and a material undeterminable category cannot be reported as the figure.
C · Report the measured estimate but frame the primary finding as a records and retention problem — the undeterminable cases mean the true figure cannot be established.
Your decision and one-line reason:



AI governanceThe exception report is now AI-generated, and the full-sample review found the AI missed 3 of 4 irregular payments. Low detection is not low fraud — and AI inherits the data duty underneath every number.

3 · Governed response — reveal after deciding

AIHEART · NESTED CONTAINMENT GEOMETRY · GOVERNED RESPONSE BLOCK

Input under review: The measured fraud-loss estimate versus the detected figure — and the question of what the audit committee is told.

Classification of the event: A measurement output that must never be dressed as certainty — and a decision that must not hide behind either number alone.

LLayerCheck on this caseVerdict
L1 · Input validationSample design, method and calculation verified; the full-sample review is reproducible.PASS
L2 · Context analysisDetected fraud is not actual fraud — FRA and FLM answer different questions. The measured estimate is best available, not gospel.HOLD
L3 · Harm / lossReporting only the detected figure understates; a single point estimate overstates. Present both, with the uncertainty.PASS
L4 · Equity weightingHonest numbers protect the public purse and the staff who run the controls — no smoothing to comfort anyone.PASS
L5 · Coherence checkStay inside the measurement scope of the course — entity-level duties live in the PGPA Act and the Rule, not here.VERIFY
L6 · Integration gateCross-layer consistent: both figures, method, and the reason they differ.PASS
L7 · Audit generationMethod, sample, estimate and wide confidence bounds appended — append-only.PASS
L8 · Output certificationThe committee brief is certified by the accountable officer before release.PASS

OVERALL VERDICT  ·  HOLDVERIFYCERTIFIED RELEASE

HOLD → CERTIFIED RELEASE — both figures go to the committee, with method and wide bounds.

PERMISSION GATE · NODE 9 — RELEASE AUTHORITY
The accountable officer signs the committee brief. The estimate is released with its uncertainty — never alone, never as a verdict on any individual.
[00:00:00.000] L1 method verified · sample + full review · reproducible
[00:00:01.000] L2 framing held · detected ≠ actual · two questions
[00:00:02.000] L3 harm weighed · both numbers, both honest
[00:00:03.000] L4 equity held · no smoothing for comfort
[00:00:04.000] L5 scope kept · measurement only · entity duties elsewhere
[00:00:05.000] L6 integration passed · figures + method + reason
[00:00:06.000] L7 audit appended · bounds recorded
[00:00:07.000] L8 accountable officer certifies · EXIT = CERTIFIED ONLY

Governed answer

Report the measured estimate as the entity's best available figure alongside the published detected figure, with the method and the wide confidence bounds, and say plainly why they differ. The audit committee gets both numbers — and the reason — not one dressed as the truth. The measurement is scoped to the IPSFF measurement controls; entity-level fraud duties live in the PGPA Act regime and your fraud control plan, not in this brief.

Cited controls: IPSFF FLM (measurement-scoped) · FRA

Provenance. This governed response was drafted with AI assistance and gated against the cited source documents. It is a training artefact of the BestYou·AI scenario layer and carries no endorsement; the governed verification pipeline status of the source course (BY-G7) applies as stated on its course page. Nothing here is legal advice.

BestYou·AI · Scenario Lab · Standalone governed response — BY-G7 · Practice material — builds capability toward the frameworks cited on the course page; it does not discharge or evidence compliance.