Identity evidence resolution

Figure 01 / 12

A match is an identity hypothesis

The synthetic population contains 42 known identity matches to fictional list records and 12458 nonmatches. This ground truth is supplied for the example. A real list match still requires correct identity resolution and a separate analysis of the applicable restriction.

Figure data and text version

Known synthetic status	Records
Same subject as fictional list record	42
Different subject	12,458

An applicant’s submitted identity must be connected to reliable evidence about the same subject. Similar text can refer to different people, while different text can refer to the same person.

The product treats a name similarity result as a verified identity.

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 matcher returns 376 candidates from 12,500 records. It identifies 27 of 42 known fictional identity matches and misses 15. After scoped suppressions and stale-evidence returns, review demand is 258. The example keeps identity resolution, control availability, and the final legal disposition separate.

Figure 02 / 12

Candidate generation changes reviewer workload

The matcher returns 376 candidates: 27 known synthetic matches and 349 unrelated records. It misses 15 known matches. A low candidate count can mean precision improved or coverage failed; the truth set and data pipeline are needed to distinguish those explanations.

Figure data and text version

Measure	Records
Candidates	376
Known matches found	27
Unrelated candidates	349
Known matches missed	15

Figure 03 / 12

Evaluate matching against a defined truth set

The four cells sum to 12,500. Candidate precision is 7.18% and matching recall is 64.29%. The labels refer only to fictional identity matches in this test set, not to whether a real payment is lawful or a customer is suspicious.

Figure data and text version

Actual identity	Candidate	No candidate
Same fictional subject	27	15
Different subject	349	12,109

Figure 04 / 12

A toy token score exposes a limitation

The example query is “Marin Trade”. The score is the Jaccard overlap of lowercase space-separated token sets, multiplied by 100. Reversing token order leaves this toy score unchanged. It ignores spelling distance, transliteration, dates, addresses, and legal identity, so it is not a production screening method.

Figure data and text version

Candidate string	Token overlap %
Marin Trade	100
MARIN TRADE	100
Trade Marin	100
Marin Trade Holdings	66.67
Marin Services	33.33
North Harbor Logistics	0

Figure 05 / 12

More attributes can change the interpretation

A name score is a lead. A reviewer compares reliable identifiers and relevant context, while preserving uncertainty when attributes are missing. A disagreement can support exclusion only under the approved method and facts; missing information is not a reliable contradiction.

Figure data and text version

Attribute	Illustrative evidence	Interpretation
Name	Marin Trade	Candidate-generation input
Registration identifier	Case-specific identifier	Potentially strong identity evidence
Country	Declared and document values	Context, not a universal exclusion
Effective date	List and customer record dates	Facts must refer to the relevant period
Ownership	Separate graph record	Not inferred from text similarity

Figure 06 / 12

List freshness is an operational dependency

The stated ingestion lag is 180 minutes. A download, successful parse, validated release, activation, and rescreen are separate events. “Latest file received” is not proof that the production matcher used it for the relevant decision.

Figure data and text version

Event	Relative minute	Evidence
Publisher release	0	Source release identifier
Ingestion	180	Retrieved bytes and checksum
Validation	182	Schema and population checks
Activation	184	Version used by production decisions
Rescreen batch	194	Affected population and outcomes

Figure 07 / 12

A clearance has scope and expiry

169 candidate references meet a prior suppression condition, but 51 no longer meet its evidence requirements and return to review. Suppression applies to a defined subject, list record, and evidence basis; it is not a general exemption for similar names.

Figure data and text version

State	Candidate references
Raw candidates	376
Suppression matches	169
Stale suppression evidence	51
Net review demand	258

Figure 08 / 12

Review demand must reach an owned queue

Net review demand is 258 cases and the illustrative capacity is 180. That leaves 78 pending. Staffing can change waiting time, but it does not change the applicable disposition or justify an automatic release when a required control is unresolved.

Figure data and text version

Measure	Cases
Net review demand	258
Available capacity	180
Completed within capacity	180
Pending	78

Figure 09 / 12

Unavailable screening is not a clear result

2250 records enter a pending-control state in this scenario, while 10250 have a recorded screening result. The status distinction prevents an unavailable dependency from being represented as no match. The actual action follows approved scope-specific policy and applicable duties.

Figure data and text version

Control state	Records
Screening result recorded	10,250
Required result pending	2,250

Figure 10 / 12

Match resolution and legal disposition differ

The workflow first establishes whether the candidate is the relevant subject, then applies the appropriate legal analysis and disposition. A reviewer can document a false identity match without deciding that every possible restriction on the transaction has been resolved.

Figure data and text version

Stage	Decision scope
Candidate	Potential similarity or identifier relationship
Identity resolution	Same subject, different subject, or unresolved
Scope analysis	Relevant jurisdiction, activity, program, and authority
Disposition	Approved handling and any required records or reporting
Release control	Authorized evidence-linked transition

Figure 11 / 12

Keep the list, matcher, and evidence versions

The record explains which inputs supported resolve the person behind the claim. Replaying an old payment against a current list answers a current-screening question. Reconstructing the original decision requires the list and logic versions actually available at that time.

Figure data and text version

Field	Illustrative value
entity_ref	Identity evidence resolution
list_release	fictional-release-2026-09-18
matcher_version	teaching-v2
observed_lag_minutes	180
candidate_count	376
review_case_count	258
evidence_owner	risk operations

Figure 12 / 12

False clears and false alerts have different costs

A complete assessment includes missed identity matches, unnecessary review, stale clearances, and unavailable controls. These observations cannot be collapsed into one accuracy percentage without losing their different operational and legal implications.

Figure data and text version

Failure mode	Observed in this example	Required response
Missed known match	15	Investigate matching and data coverage
Unrelated candidate	349	Improve evidence and resolution
Stale suppression	51	Reassess the clearance basis
Pending required result	2,250	Apply the approved failure policy

Connect the result to the system

Use scoped attributes, source quality, and an exception path before granting the intended authority.

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.