Model operating health

Figure 01 / 12

The population that monitoring actually sees

38,220 of 39,000 source items enter this monitoring calculation. The missing 780 items are a coverage gap, not evidence of low risk. Reconcile stable identifiers and amounts where appropriate before interpreting the alert rate.

Figure data and text version

Population state	Items
Included in monitoring	38,220
Absent from this run	780

A model can retain its score distribution while its inputs, labels, or downstream actions fail. Monitoring should join technical quality with mature business outcomes.

The reference case starts with the stated population and a functioning evidence path. The owner is risk operations.

All amounts, rates, capacity limits, and outcomes in this case are synthetic. The three conditions are separate assumptions for comparison. A better result in the response condition is not measured proof that the proposed control causes that improvement. The figures expose the calculation and its limits; a real deployment needs its own evidence.

Read the result

The daily source population is 39,000 items, but 780 are outside the completed monitoring run. The included population creates 459 hits and 376 unique cases. With 95 cases already open and capacity for 460, the queue closes at 11. Coverage, duplicate work, and staffing are separate causes; reducing one number does not prove that the overall control improved.

Figure 02 / 12

From scenario hits to unique cases

459 raw hits become 376 cases after removing 83 repeated references to the same case under the stated merge rule. Deduplication should reduce duplicate work while retaining the underlying events and reasons. It must not merge unrelated activity merely because values look similar.

Figure data and text version

Stage	Count
Raw scenario hits	459
Duplicate references	83
Unique cases	376

Figure 03 / 12

The queue balance is an accounting identity

Opening backlog 95 + arrivals 376 − completed cases 460 = closing backlog 11. Completion is capped by both available work and the stated capacity. This identity is useful even when average handling times are uncertain.

Figure data and text version

Movement	Cases	Definition
Opening backlog	95	Unresolved at window start
New cases	376	Unique arrivals in this window
Completed	460	Reached a defined completion state
Closing backlog	11	Unresolved at window end

Figure 04 / 12

Backlog accumulates across uneven days

This deterministic six-day example applies a stated daily arrival multiplier and a constant completion capacity. Unused capacity does not carry forward as completed work. The series is a workload illustration; it omits variable case durations and specialist routing. Horizontal positions are the labeled observations or scenarios; equal spacing does not imply equal numerical increments.

Figure data and text version

Day	Closing backlog
D1	11
D2	0
D3	0
D4	104
D5	58
D6	0

Figure 05 / 12

Arrivals and completions need separate plots

The service line cannot exceed the work available that day. A team can complete more cases than arrive while clearing an opening backlog. Conversely, stable staffing can coexist with a growing queue when arrivals remain higher than capacity. Horizontal positions are the labeled observations or scenarios; equal spacing does not imply equal numerical increments.

Figure data and text version

Day	Arrivals	Completions
D1	376	460
D2	320	331
D3	451	451
D4	564	460
D5	414	460
D6	263	321

Figure 06 / 12

An illustrative age profile of open work

The closing backlog is 11 cases. The 50/30/remainder split is an explicit illustrative age allocation, not a distribution inferred from the arrival model. Production age buckets must come from each case’s actual receipt and status history.

Figure data and text version

Age bucket	Open cases
Under one day	6
One to three days	3
Over three days	2

Figure 07 / 12

Completion outcomes are not criminal labels

Of 460 completed review tasks, 78 are referred for further assessment, 55 need another evidence action, and 327 close under the stated procedure. These are operational outcomes. None is a probability of money laundering or a substitute for a reporting decision.

Figure data and text version

Review disposition	Tasks
Referred for assessment	78
Further evidence action	55
Closed under procedure	327

Figure 08 / 12

Known-event tests assess specific coverage

The injected test set contains 100 known test events designed to exercise monitor the model as part of a system. The system surfaces 92. That 92% detection result measures this constructed test set only; it does not establish population-wide detection of illicit activity.

Figure data and text version

Measure	Count	Interpretation
Injected known events	100	Defined test population
Detected by the scenario	92	Expected evidence reached the control
Not detected	8	Investigate data, logic, and delivery
Real-world illicit prevalence	Unknown	Not inferred from this test

Figure 09 / 12

Data quality has several independent dimensions

Each row is a different field requirement over the same source population. Completeness alone does not establish that values are accurate or current. The case uses explicit illustrative missing counts to show how data quality can affect scenario coverage.

Figure data and text version

Field requirement	Present	Absent
Party reference	38,805	195
Event time	38,610	390
Counterparty context	37,830	1,170

Figure 10 / 12

Type the relationship before drawing an inference

This evidence map distinguishes a customer relationship, a transfer, and a case reference. The links support monitor the model as part of a system; they do not imply common ownership or intent. A shared data point is a lead whose meaning depends on source, time, and context.

Figure data and text version

From	To	Relationship
Model operating health	Counterparty A	Observed transfer
Model operating health	Profile record	Declared business
Counterparty A	Case record	Evidence reference
Profile record	Case record	Context for review

Figure 11 / 12

Case clocks start from defined events

A legal deadline, an internal response target, and an evidence-expiry date can start from different events. The hours here are internal teaching targets only. They are not BSA, sanctions, consumer-protection, or other statutory deadlines.

Figure data and text version

Event	Relative time	Operational meaning
Source event	T0	Activity occurred
Data arrival	T0 + 2 hours	The monitoring system learned it
Case created	T0 + 3 hours	Work entered an owned queue
Internal review target	T0 + 27 hours	Illustrative 24-hour target from case creation
Disposition	Recorded separately	Use actual decision and reporting records

Figure 12 / 12

The end-to-end delivery contract

The scenario is incomplete until the intended evidence reaches an owned case. For monitor the model as part of a system, verify source coverage, hit creation, queue acceptance, reviewer access, and final disposition separately. A green job status proves only that a job reported completion.

Figure data and text version

Boundary	Acceptance evidence
Source to scenario	38220 included source items; 780 missing
Scenario to case	459 hits linked to 376 unique cases
Case to reviewer	Required evidence visible under the reviewer role
Reviewer to outcome	Disposition, rationale, and any separate reporting decision retained

Connect the result to the system

Monitor input coverage, action delivery, queue age, and outcome maturity together.

Check the population, evidence, permitted action, and actual effect together. A balanced calculation can still use the wrong population; a successful response can still leave an unknown financial outcome. The case’s numerical result applies only to its stated assumptions.