Match Scoring Guide

Match scoring decides how broadly or narrowly Minerva keeps potential screening matches as they pass through candidate retrieval, entity resolution, and final result filtering. Raise the bar and fewer potential matches reach analysts in Screening. Lower it and more borderline candidates stay visible for review. Access: Requires the Admin role or above. In the sidebar, go to Administration > Configuration, then open Match scoring under Screening Behaviours. Configuration is workflow-specific and workspace-scoped, so every profile screened in a workspace shares the same defaults unless a request overrides them. A Calibration workspace lets you test settings without changing Live until a workspace preset is promoted. Profile-level segmentation is handled separately by profile groups.

How match scoring works

Every screening result passes through three stages:

Candidate retrieval: Minerva gathers source records from the selected feeds.
Pre-resolution scoring: each source candidate is scored against the searched subject before entity resolution.
Post-resolution scoring: Minerva clusters likely duplicate source records, scores the resolved candidate, and filters the final returned result set.

The Match Scoring configuration controls the second and third stages through eight groups of settings:

Decision thresholds: feed-specific pre-resolution thresholds and one post-resolution threshold.
Candidate score weights: how much each evidence category contributes when it is present.
Candidate gates: minimum name and Sanctions/PEP cutoffs that can stop candidates before other evidence helps.
Name matching behavior: tolerance for transposed names, initials, phonetics, hyphenation, extra name parts, and organization base/full names.
DOB and date matching behavior: soft date falloff, hard date cutoffs, and inferred-date confidence handling.
Location matching behavior: default strictness, component cutoff, and whether nationality can satisfy country matching.
Entity-resolution merge behavior: merge thresholds and strong organization merge shortcut settings.
News controls: geography filter, name filter, article cap, and article-location inference before News match scoring.

Scoring compares the candidate with the searched subject across the feeds the search selected: sanctions lists, PEP data, and News adverse media, plus identity, profile context, and supporting evidence such as email, phone, and notes. PEP, Criminal, and High Risk Industry signals are derived separately by risk inference. The same settings usually call for different strictness in each workflow:

Workflow	What it covers	Typical tuning posture
Onboarding	Profile creation and first-pass screening before a customer relationship is active.	Usually stricter because analysts are deciding whether to accept a new profile.
Ongoing monitoring	Repeated monitoring screens for existing profiles.	Usually conservative because changes can affect recurring alert volume across many profiles.
Direct API calls	API-driven screening submissions.	Often tuned for partner or reseller integrations. Workspace defaults are preferred, but request-level overrides remain supported.
Risk assessments	One-off risk-assessment searches in the app.	Often broader because analysts expect to inspect evidence directly in an investigation flow.

What a score means

Minerva uses normalized match scores from 0.00 to 1.00.

1.00 means the compared values are effectively exact matches.
Values closer to 1.00 are stronger matches.
Values closer to 0.00 are weaker matches.
A threshold is the minimum score required to keep the candidate at that stage.

In result details and API responses, field-level comparisons carry these labels:

Label	Score range	Meaning
Exact	0.98 and above	The field is effectively exact.
Close	0.85 to below 0.98	The field is strongly similar, but not exact.
Loose	0.75 to below 0.85	The field is a weaker fuzzy match.
None	Below 0.75	The field does not provide meaningful match evidence.

The overall candidate score is not a simple average of the visible field labels. The engine scores the candidate using weighted evidence, candidate gates, missing-evidence behavior, and stage-specific thresholds. Once a potential match exists, Automatic Disposition can predict whether it should be cleared or escalated without analyst review.

Start from a preset

Minerva includes three built-in presets. Built-in presets stay fixed, and workspace presets can be created from a custom draft.

Preset	Best for	Main behavior
Balanced	Default production posture.	Preserves legacy defaults for customers that do not change match scoring.
Narrow	Reducing borderline matches and review volume.	Raises thresholds, strengthens identity/date/location evidence, and applies stricter date/location/merge behavior.
Wide	Calibration reviews, higher-risk cohorts, or investigations where recall matters more than review volume.	Lowers thresholds, broadens News filtering, makes name/date/location matching more tolerant, and lowers merge thresholds.

Balanced is the default for every workspace until an administrator changes the settings.

The exact numbers behind each preset are listed under decision thresholds, candidate score weights, and News and merge values.

Change one workflow at a time

Each workflow is configured on its own tab, so you can tune one without changing the others. The tab navigation stays visible as you scroll, so you can see which workflow you are changing. Use these actions when you need to keep workflows aligned:

Action	What it does	When to use it
Copy from another channel	Replaces the current workflow with the selected source workflow.	Use when onboarding and monitoring should share a posture, or when direct API should start from risk assessment settings before a small adjustment.
Apply this channel to all	Copies the active workflow to every other workflow.	Use after validating a complete workflow configuration in Calibration and deciding the same settings should apply everywhere.
Workspace presets	Save a custom draft as a reusable workspace preset, update a workspace preset, or promote it between workspaces.	Use for repeatable operating modes, such as a stricter monitoring preset or a broader investigation preset.

Compare representative cases in a Calibration workspace before copying a workflow’s settings to the others.

Find the setting you need

The page includes a setting finder and the Minerva command palette also indexes the Match Scoring controls.

Method	How to use it	What happens
Page setting finder	Search within the Match Scoring page for terms such as phonetics, DOB, missing evidence, merge threshold, or News article limit.	The page filters or jumps to matching controls so an admin does not need to scan every section.
Command palette	Open the command palette and search for a setting name or common variant, such as transposition, sound alike, person name cutoff, location strictness, or score simulator.	Minerva opens Match Scoring and highlights the requested control with a brief yellow flash.

Use the command palette when you already know the setting you need. Use the page finder when you are comparing nearby settings inside the same section.

Test a change in the score simulator

The score simulator lets administrators test draft settings without creating a live search or writing screening results. Use it when you want to answer questions like:

Would this exact-name, missing-DOB candidate still pass if missing evidence is penalized more?
How much does a different city or country reduce the score?
Would a transposed or phonetic name variation pass under Narrow, Balanced, or Wide?
Does the pre-resolution score behave differently from the post-resolution score?
Which setting moved the score enough to pass or fail the selected threshold?

Simulator steps

Select the workflow tab you want to test, such as Onboarding or Direct API calls.
Adjust the draft settings or select a preset. You do not need to save first.
Go to Score simulator.
Choose Pre-resolution or Post-resolution under Score using.
Edit the Screened profile. Include the subject type, name, date, location, and screening sources that represent the request.
Edit the Potential match record. Include the candidate name, aliases, date, locations, country, nationality, occupation/role, source lists, and feed hits you want to compare.
Click Run simulation.
Review the resulting score, return/filter state, threshold comparison, and attribute comparison.

Run the same example before and after changing one setting. This makes it easier to see whether the score moved because of a threshold, a field weight, missing evidence, name behavior, date behavior, or location behavior.

Reading simulation results

Result area	How to interpret it
Overall score	The simulated candidate score under the selected workflow and stage.
Passed / failed	Whether the score meets the selected threshold and gates for that stage.
Threshold snapshot	The pre-resolution or post-resolution thresholds used for the run.
Attribute comparison	Field-level comparison scores such as name, aliases, date, city, state, and country.
Tested fields	The screened-profile and potential-match fields captured in simulator history.
Simulation history	Recent simulator runs can be reviewed so teams can connect save decisions to examples they tested. Saved changes can link back to the simulations reviewed before deployment.

Simulator results are examples, not production searches. They are best used for calibration and explanation before saving changes or promoting a workspace preset.

Calibrate in a controlled order

Match scoring changes affect which candidates analysts see. Tune in small increments, use clear change descriptions, and compare review volume, latency, and missed-risk sensitivity before moving further away from Balanced.

Use this sequence when calibrating match scoring:

Start from Balanced and collect examples of false positives, plausible missed matches, and high-volume review cohorts.
Tune one workflow at a time. Onboarding, monitoring, Direct API, and risk assessments often have different tolerance for review volume.
Use the simulator to compare representative examples before saving.
Adjust feed-specific pre-resolution thresholds when the issue is feed-specific noise, such as weak records that are obviously unrelated to the searched subject.
Adjust post-resolution thresholds when entity resolution is grouping candidates correctly but the final result set is too broad or too narrow, or when records look related before entity resolution and disappear after final scoring.
Adjust candidate score weights only when the field-level evidence balance is wrong. For example, increase location or DOB/date weights when similar names are passing despite conflicting location or date evidence.
Adjust name, DOB/date, and location behavior when the field similarity score itself is too strict or too forgiving.
Adjust merge thresholds when entity resolution is over-merging or under-merging source records.
Tune one adverse-media retrieval control at a time. Moving geography bias from Strict to Broad or None, turning on Broad name filtering, or raising max requested articles changes the upstream article pool, so the effect reaches further than the final score filter.
Enable News location inference only where the extra subject-linked geography is worth the added latency.
Save a clear change description so history shows why the calibration was deployed.
Use history and rollback if the change moves review volume, latency, or recall in the wrong direction.

General tuning guidance:

Higher thresholds are stricter and usually reduce the number of analyst-visible potential matches, while lower thresholds increase recall and review volume
Higher evidence weights make that evidence category more influential when present
Higher name tolerance settings generally make name variations easier to match, while higher penalties or floors make name matching stricter
Monitoring changes can change recurring alert volume across an existing profile population, and repeated alert suppression handles the repeat noise that ongoing monitoring creates
Direct API changes should be coordinated with API consumers, especially if those consumers already submit request-level overrides

Review and save a change

When you click Review changes, Minerva shows a grouped confirmation view before anything is saved. The review dialog shows:

The number of changes and sections affected
The selected preset change, if applicable
Each workflow with changed thresholds, weights, gates, matching behavior, merge behavior, and News controls
An optional Change Description field
Simulator examples linked to the draft, when available

Use the change description to capture the reason for the calibration, such as a false-positive review, a model calibration review, a high-risk cohort exception, or a post-launch monitoring adjustment.

Check history and roll back a change

The Match Scoring page includes a quick History preview so you can review recent deployments without leaving the configuration page. Use the preview to:

See the current live deployment first
Review prior threshold, weight, matching behavior, merge behavior, and News retrieval changes
Preview an older deployment on the main page
Open the full audit history page

Rollback restores a previous deployment by writing a new history entry. It does not delete prior history. The full history page includes:

Changed: when the configuration was saved or rolled back
Action: whether the event was an update or rollback
Summary: the key scoring changes
Changed by: the user who performed the change
Actions: rollback entry points for older deployments

Decision thresholds

Pre-resolution and post-resolution thresholds answer different questions.

Threshold	Stage	What increasing does	What decreasing does
Sanctions pre-resolution	Before entity resolution	Fewer weak sanctions source records continue into resolution.	More borderline sanctions records can continue and potentially merge with other evidence.
PEP pre-resolution	Before entity resolution	Fewer weak PEP source records continue into resolution.	More borderline PEP records can continue and potentially merge with other evidence.
News pre-resolution	Before entity resolution	Fewer weak adverse-media source records continue into resolution.	More borderline News records can continue and potentially merge with other evidence.
Post-resolution	After entity resolution	Final analyst-visible results become narrower.	Lower-scoring resolved candidates can remain visible.

Preset thresholds

Decision thresholds are editable from 0.60 to 1.00.

Preset	Workflow	Sanctions pre-resolution	PEP pre-resolution	News pre-resolution	Post-resolution
Balanced	Onboarding	0.85	0.85	0.85	0.85
Balanced	Ongoing monitoring	0.85	0.85	0.85	0.85
Balanced	Direct API calls	0.75	0.75	0.70	0.75
Balanced	Risk assessments	0.75	0.75	0.70	0.70
Narrow	Onboarding	0.90	0.90	0.90	0.90
Narrow	Ongoing monitoring	0.90	0.90	0.90	0.90
Narrow	Direct API calls	0.85	0.85	0.80	0.80
Narrow	Risk assessments	0.85	0.85	0.80	0.80
Wide	Onboarding	0.75	0.75	0.75	0.75
Wide	Ongoing monitoring	0.75	0.75	0.75	0.75
Wide	Direct API calls	0.65	0.65	0.60	0.60
Wide	Risk assessments	0.65	0.65	0.60	0.60

Candidate score weights

Candidate score weights control how much each evidence category contributes when that evidence is present. The dashboard presents the main control once per setting and applies it to both stages. Use Advanced: split pre/post resolution only when a workflow needs different behavior before and after entity resolution. Most evidence-weight sliders use a customer-facing range of 0.00 to 1.00. The Missing evidence penalty uses 0 to 100.

Group	Setting	What it affects	Balanced pre	Balanced post	Tuning guidance
Identity evidence	Primary name	Main searched-subject name compared with the candidate name.	1.00	1.00	Keep high for most workflows. Lower only when other evidence is highly reliable and names vary substantially.
Identity evidence	Aliases	Known alternate names on the potential match.	0.98	0.98	Raise when aliases are reliable and important; lower if alias lists create common-name noise.
Identity evidence	Identifiers	Government, registration, and other identifier evidence.	0.05	0.05	Increase when identifiers are normalized and trusted. Keep lower when identifiers are often partial or absent.
Profile context	Location (four components)	Country, state / province, city, and address evidence, split across four component weights. See Location component weights.	0.10	0.75	Higher values separate similar names by geography. Lower values are broader when locations are incomplete or inconsistent.
Profile context	DOB / date	Birth date, incorporation date, or formation date.	0.10	0.50	Higher values make date agreement more decisive. Lower values are broader when dates are often missing, partial, or inferred.
Profile context	Occupation	Role, job title, or organization relationship evidence.	0.10	0.01	Increase when role data reliably distinguishes matches. Keep lower when role data is sparse or inconsistently sourced.
Profile context	Gender	Gender evidence when present on both records.	0.05	0.05	Use carefully because gender data can be missing or inconsistently reported.
Supporting evidence	Email	Email address overlap between candidate and match.	0.05	0.05	Increase when email is a trusted identifier for the workflow.
Supporting evidence	Phone	Phone number overlap between candidate and match.	0.05	0.05	Increase when phone data is trusted and not commonly shared or recycled.
Supporting evidence	Notes	Customer-supplied notes and narrative evidence.	0.05	0.05	Increase only when notes are consistently structured and reliable.
Score behavior	Candidate score pass threshold	Internal candidate-score pass threshold for the selected stage.	0.70	0.70	Higher values filter candidates earlier in the scoring stage. This is separate from public result thresholds.
Score behavior	Missing evidence penalty	How strongly unavailable optional evidence lowers the score.	5	10	A value of 0 disables the missing-evidence penalty. Larger values penalize records with unavailable optional evidence more strongly.

Missing evidence penalty is safe to turn off. A value of 0 means missing optional fields do not reduce the denominator. For example, if the request includes name and DOB, and the candidate matches the name but has no DOB to compare, the DOB absence does not reduce the score when the missing-evidence penalty is 0.

Location component weights

The Location card holds four separate component weights: Country, State / province, City, and Address. Each uses the 0.00 to 1.00 range, and the four add up to the location total. Whenever at least one component is compared, the location block contributes the whole location total, not just the compared components’ shares. A country-only comparison on Balanced still carries the full 0.75 post-resolution block, scaled by the country score. Because each component uses the 0.00 to 1.00 range, a custom split can also make the total larger than the single weight it replaced.

Component	What it compares	Balanced pre	Balanced post
Country	Country evidence. Nationality can satisfy country matching when enabled.	0.025	0.1875
State / province	State, province, or region evidence.	0.025	0.1875
City	City or locality evidence.	0.025	0.1875
Address	Street-level address evidence. Usually absent on screening records.	0.025	0.1875
Total location weight	The sum of the four component weights.	0.10	0.75

Default and backward compatibility. When no component weight is set, the previous single location weight is split evenly across the four components. Balanced defaults are 0.025 for each component before resolution and 0.1875 for each component after resolution. A workspace that never configured location components scores exactly as before, and an existing location weight needs no rewrite.

Setting any component changes the shape. The components you leave out resolve to 0, and the location total becomes the sum of the components you set. location is not consulted once a component is present, so set all four explicitly to keep a mixed posture, which the dashboard does. Address is usually absent. Set Address to 0 when street-level evidence is unreliable or incomplete. A zero-weight component does not contribute to the score, adds no absence charge, and does not trigger strict location gating. A value the request supplies can still appear in the result’s field-level comparison detail. While Address is nonzero, a request address the candidate lacks still adds to the missing-evidence denominator, scaled by the address weight. A component weight of 0 disables that component for scoring and for strict location gating: a strictness level cannot hard-reject on a component whose weight is 0. Missing evidence is not symmetric between the two sides:

Request-side omission is free. A component the request does not supply adds nothing to the numerator or the denominator, so a country-only request is scored on country alone.
Candidate-side absence is charged. When the request supplies a component and the candidate has none, the block adds that component’s missing-evidence charge to the denominator, scaled by the component’s weight. Under the even default split this is the same per-component charge the previous single location weight produced.
A zero-weight component adds no charge on either side.

Country is the exception to the two-sided rule: when Allow nationality for country match is enabled, nationality supplies the candidate side even if the candidate has no country field. Component similarity still scores partially. An address, city, state, or country value that is similar rather than exact contributes its own partial score to the average, weighted by that component’s weight. For example, a candidate recorded as Lagos, Nigeria still scores against a request that supplies only Nigeria. Country is compared because both sides have it, city, state, and address are skipped because the request has no value for them, and the location average is the country score alone. The block still contributes its full total for that comparison: a country-only match on Balanced carries 0.75, not the country share. The request is not charged for the components it never sent. Downweight city and state without discounting country. Lower City and State / province while leaving Country and Address where they are when a lone city/state disagreement should not drop an otherwise strong match.

Component	Balanced post-resolution	Lower city and state	Same total, country raised
Country	0.1875	0.1875	0.4375
State / province	0.1875	0.0625	0.0625
City	0.1875	0.0625	0.0625
Address	0.1875	0.1875	0.1875
Total location weight	0.75	0.50	0.75

Lowering city and state reduces geographic discrimination. Candidates that differ only by city keep more of their location contribution, so expect more borderline matches and additional review volume. The location total drops by exactly what you remove from the components, so location contributes less to every candidate score. Raise Country by the amount you removed when the overall location contribution must stay the same.

Retrieval geography is separate from scoring geography. News and Open Source retrieval geography (Strict, Broad, or None) decides which articles are retrieved before scoring, and it does not read the location component weights. If a subject’s address is current but their adverse coverage places them elsewhere historically, lowering city and state fixes scoring but not retrieval. Set the News geography filter to None or Broad as well.

Component weights change how much each level contributes, not how strict the comparison is. Default location strictness, the strict location cutoff, and the Sanctions/PEP location weight override work as before. The override still sets the total location contribution, and Minerva scales the component weights proportionally so their ratio is preserved.

Preset weight values

Setting	Balanced pre/post	Narrow pre/post	Wide pre/post
Primary name	1.00 / 1.00	1.00 / 1.00	1.00 / 1.00
Aliases	0.98 / 0.98	0.98 / 0.98	0.98 / 0.98
Location: Country	0.025 / 0.1875	0.0625 / 0.25	0.0125 / 0.125
Location: State / province	0.025 / 0.1875	0.0625 / 0.25	0.0125 / 0.125
Location: City	0.025 / 0.1875	0.0625 / 0.25	0.0125 / 0.125
Location: Address	0.025 / 0.1875	0.0625 / 0.25	0.0125 / 0.125
Location total	0.10 / 0.75	0.25 / 1.00	0.05 / 0.50
DOB / date	0.10 / 0.50	0.20 / 0.80	0.05 / 0.30
Occupation	0.10 / 0.01	0.10 / 0.01	0.10 / 0.01
Gender	0.05 / 0.05	0.05 / 0.05	0.05 / 0.05
Notes	0.05 / 0.05	0.05 / 0.05	0.05 / 0.05
Email	0.05 / 0.05	0.05 / 0.05	0.05 / 0.05
Phone	0.05 / 0.05	0.05 / 0.05	0.05 / 0.05
Identifiers	0.05 / 0.05	0.05 / 0.05	0.05 / 0.05
Candidate score pass threshold	0.70 / 0.70	0.78 / 0.80	0.60 / 0.60
Missing evidence penalty	5 / 10	10 / 20	2.5 / 4

Each location component is exactly one quarter of the preset’s previous single location weight: 0.10 / 4 = 0.025 before resolution and 0.75 / 4 = 0.1875 after resolution under Balanced.

Candidate gates

Candidate gates are cutoffs that can stop a candidate before weighted evidence alone determines the score.

Setting	Range	Balanced	Narrow	Wide	What increasing does
Person name cutoff	0.00 to 1.00	0.50	0.60	0.35	Person candidates need stronger name evidence before other signals can help.
Organization name cutoff	0.00 to 1.00	0.80	0.88	0.70	Organization candidates need closer name similarity before continuing.
Sanctions/PEP cutoff	0.00 to 1.00	0.75	0.82	0.65	Sanctions and PEP candidates need stronger total evidence before continuing.
Sanctions/PEP location weight	0.00 to 1.00	0.05	0.10	0.02	Location agreement has more influence for Sanctions and PEP candidate scoring.

Some legacy backend fields remain available for API compatibility but are not shown in the dashboard. Customer-facing configuration should use the settings above.

Name matching behavior

Name matching behavior controls how the engine converts compared names into name similarity scores before the name weight is applied.

Person names

Setting	Range	Balanced	Narrow	Wide	What it controls
Transposition tolerance	0.00 to 1.00	0.9425	0.90	0.97	How much name-part transposition affects similarity, such as given and family names appearing in reverse order.
Initials tolerance	0.00 to 1.00	0.80	0.70	0.90	How initials are treated when compared with full given, middle, or family-name parts.
Extra name penalty	0.00 to 1.00	0.05	0.08	0.02	How strongly unmatched extra name parts reduce similarity.
Text similarity floor	0.00 to 1.00	0.05	0.08	0.02	Minimum spelling similarity needed before phonetic matching can help.
Phonetics tolerance	0.00 to 1.00	0.815	0.75	0.90	How much Double Metaphone-style sound-alike similarity can preserve score when spellings differ.
Length difference weight	0.00 to 1.00	0.70	0.80	0.50	How strongly large name-length differences reduce similarity.
Hyphenated name tolerance	0.00 to 1.00	0.98	0.95	1.00	How hyphenated names and expanded variants compare, such as Smith-Jones and Smith Jones.

Use higher tolerance values when the workflow should accept common formatting, transliteration, phonetic, and ordering variation. Use lower tolerance values when name precision matters more and broader recall is creating too many same-name matches.

Organization names

Setting	Range	Balanced	Narrow	Wide	What it controls
Base name weight	0.00 to 1.00	0.75	0.80	0.65	How much the normalized base organization name contributes after common suffixes and descriptors are reduced.
Full name weight	0.00 to 1.00	0.25	0.20	0.35	How much the full organization name contributes, including suffixes, descriptors, and trading-name text.

Raise base name weight when legal suffixes, punctuation, and descriptors vary frequently. Raise full name weight when exact full-name agreement should matter more.

DOB and date matching behavior

DOB/date behavior controls how date evidence is scored before the DOB/date weight is applied.

Setting	Range	Balanced	Narrow	Wide	What it controls
Person DOB soft match falloff days	1 to 36500 days	730	365	1825	How quickly person DOB similarity decays as dates move apart.
Person DOB hard range cutoff years	0 to 500 years	500	10	500	Maximum person DOB gap that can receive any date similarity. Beyond this range, date similarity is forced to 0.
Organization date soft match falloff days	1 to 36500 days	3650	1825	7300	How quickly organization date similarity decays as incorporation or formation dates move apart.
Organization date hard range cutoff years	0 to 500 years	500	20	500	Maximum organization date gap that can receive any date similarity.
High inferred date confidence	0.00 to 1.00	0.95	0.95	0.95	Contribution for high-confidence inferred date evidence when exact date is unavailable.
Low inferred date confidence	0.00 to 1.00	0.30	0.40	0.20	Contribution floor for low-confidence inferred date evidence.
Normal inferred date range years	0.10 to 100 years	2	1	5	Year range allowed around normally inferred dates.
Low confidence date range years	0.10 to 100 years	20	10	30	Year range allowed around low-confidence inferred dates.
Apply inferred date confidence weighting	Enabled or disabled	Enabled	Enabled	Enabled	Whether lower-confidence inferred dates contribute less than higher-confidence inferred dates.

The soft falloff is not a pass/fail boundary. It changes the shape of date-score decay for dates within the hard range cutoff. The hard cutoff is the boundary where date similarity becomes 0 before weighting.

Location matching behavior

Location matching behavior controls how location evidence is evaluated before the location component weights are applied.

Setting	Values / range	Balanced	Narrow	Wide	What it controls
Default location strictness	No default, Country, State / province, City, Address	No default	City	No default	Minimum location level required when a request does not provide its own strictness.
Strict location cutoff	0.00 to 1.00	0.80	0.90	0.65	Minimum component score when strict location matching is active.
Allow nationality for country match	Enabled or disabled	Enabled	Disabled	Enabled	Whether nationality can satisfy country-level matching when country evidence is missing or incomplete.

Strictness levels are progressive:

Strictness	Matching behavior
No default	Location details are scored, but no minimum location level is required unless the request sets one.
Country	Requires country agreement when both sides provide country evidence. Nationality can count when enabled.
State / province	Requires country and state/province agreement when both sides provide those details.
City	Requires country, state/province, and city agreement when both sides provide those details.
Address	Requires all available location levels, including street address, to agree closely enough.

Entity-resolution merge behavior

Entity resolution combines likely duplicate source records before post-resolution scoring. Merge settings affect candidate clustering, not the final score threshold directly.

Setting	Range	Balanced	Narrow	Wide	What increasing does
AI-assisted merge threshold	0.00 to 1.00	0.66	0.72	0.60	Automatic AI-assisted merges become stricter, reducing accidental merges.
Exact-evidence merge threshold	0.00 to 1.00	0.66	0.72	0.60	Rule-based or exact-evidence merges require stronger agreement.
Enable strong organization merges	Enabled or disabled	Enabled	Enabled	Enabled	Allows highly similar organization records to use the strong organization shortcut.
Strong organization name cutoff	0.00 to 1.00	0.88	0.92	0.82	Strong organization shortcut merges require closer organization-name agreement.
Strong organization date days	0 to 36500 days	31	14	90	Strong organization shortcut merges tolerate larger incorporation or formation-date differences.
Strong organization confidence	0.00 to 1.00	1.00	1.00	0.90	Shortcut merges are recorded with higher confidence.

Raise merge thresholds when unrelated source records are being combined. Lower them when duplicate records are staying separate and creating repeated analyst review.

News controls

Each screening workflow has its own News controls. These settings sit before final match scoring. They change which articles are retrieved, which retrieved articles are admitted for adverse-media analysis, and whether article-linked locations are added before geography scoring. For how articles are found and categorized, see adverse media.

Broadening News retrieval can increase adverse-media volume and processing time. Avoid changing geography bias, name filtering, article caps, and score thresholds all at once unless you are running a controlled calibration pass.

Setting	Values / range	Balanced	Narrow	Wide	Main effect
News geography filter	Strict, Broad, None	Strict	Strict	Broad	Controls how subject geography is applied by the News retrieval filter before match scoring.
News name filter	Strict, Broad	Strict	Strict	Broad	Controls how subject names are matched by the News retrieval filter before match scoring.
Max requested articles	100 to 750	300	200	600	Caps how many News article candidates are reviewed for article matching, sentiment scoring, and risk classification.
Infer article locations from News	Enabled or disabled	Disabled	Disabled	Disabled	Reviews matching News articles for subject-linked geography before News geography scoring.

News geography filter

Setting	Retrieval behavior	When to use it
Strict	Uses exact subject-name retrieval and applies city/state terms when available. Country-only searches do not add a geography clause.	Use as the default for common names and high-volume workflows.
Broad	Uses broader city/state matching with optional country expansion when location evidence is available.	Use when relevant articles often mention only a wider region or country.
None	Searches without geography terms.	Use for targeted investigations where location evidence is weak or misleading and analysts expect more review volume.

News name filter

Setting	Matching behavior	When to use it
Strict	Requires exact-name evidence in trusted article fields before News records continue to scoring.	Use for common names, high-volume monitoring, and workflows where precision matters more than recall.
Broad	Allows bounded token-proximity matches, minor name variation, middle-name expansion, hyphenation differences, and trusted snippet or quote matches.	Use when subjects are often reported with middle names, transliteration variants, entity suffix differences, or reliable partial snippets.

Max requested articles

Raise this value when relevant adverse-media articles may appear later in provider results. Lower it when common-name searches are creating too much latency or when a workflow does not need broad article recall.

Higher article caps can add processing time because Minerva may retrieve and review more News candidates. High-volume subjects may take a few additional seconds for each extra 100 articles, especially when retrieval has also been broadened.

News location inference

News location inference reviews matching risk-classified News articles and extracts locations that appear tied to the screened subject. Those inferred article locations can then provide more geographic evidence before News match scoring. Turn it on when article text often identifies the subject’s relevant city, state, region, country, operations, arrest, investigation, or other subject-linked geography more clearly than the original search request.

News location inference uses model-backed article analysis and can add seconds to News searches. Enable it first in Calibration, direct API, or risk-assessment workflows where the extra geographic evidence is worth the added latency before considering it for high-volume monitoring.

Preset retrieval and merge values

The built-in presets also set News retrieval, location strictness, and merge behavior:

Preset	News geography	News name filter	Max requested articles	Location inference	Default location strictness	Merge thresholds
Balanced	Strict	Strict	300	Off	No default	AI-assisted 0.66, exact-evidence 0.66
Narrow	Strict	Strict	200	Off	City	AI-assisted 0.72, exact-evidence 0.72
Wide	Broad	Broad	600	Off	No default	AI-assisted 0.60, exact-evidence 0.60

Request-level overrides

Match scoring is the workspace default. It applies when the screening request does not provide an explicit override. This is the preferred operating model, especially for Direct API customers, because it keeps defaults visible in the dashboard and audit history. Direct API requests can still override settings for partnership, reseller, legacy, or investigation-specific flows. Core screening logic gives request-level values precedence over workspace configuration. For new integrations, use the same customer-facing ranges shown in the dashboard tables. Some backend fields accept broader legacy ranges for backward compatibility, but dashboard-equivalent values are easier to reason about, audit, and compare with simulator runs.

Override type	Preferred request field	Notes
Canonical runtime object	`global_filters.match_scoring_runtime_config`	Nested object that mirrors the workspace runtime sections: `candidateScoring`, `nameMatching`, `dobMatching`, `locationMatching`, and `entityResolution`.
Feed pre-resolution threshold	`feed_filters.Sanctions.pre_resolution_match_threshold`, `feed_filters.PEP.pre_resolution_match_threshold`, `feed_filters.News.pre_resolution_match_threshold`, `feed_filters.Internal.pre_resolution_match_threshold`	Feed-specific overrides take precedence for that feed.
Feed post-resolution threshold	`feed_filters.Sanctions.post_resolution_match_threshold`, `feed_filters.PEP.post_resolution_match_threshold`, `feed_filters.News.post_resolution_match_threshold`, `feed_filters.Internal.post_resolution_match_threshold`	Per-feed stage threshold: it decides which candidates survive that scoring stage for that feed, and the final result set is additionally filtered by the global output threshold. The legacy `feed_filters.<Feed>.match_threshold` still outranks the stage-specific key for the same feed. `Internal` is the Internal Risk feed.
Feed threshold aliases	`internal_pre_resolution_match_threshold`, `sanctions_pre_resolution_match_threshold`, `pep_pre_resolution_match_threshold`, `news_pre_resolution_match_threshold`	Top-level aliases on v1 write requests, normalized into `feed_filters.<Feed>.pre_resolution_match_threshold`. The camelCase spellings (`sanctionsPreResolutionMatchThreshold`) and the `<feed>_match_threshold` forms are also accepted. Per-feed post-resolution thresholds have no top-level alias; send them in `feed_filters`.
Global pre-resolution threshold	`global_filters.pre_resolution_match_threshold`	Applies when a feed-specific threshold is not supplied.
Post-resolution threshold	`global_filters.post_resolution_match_threshold`	Controls the final returned result set for the request.
Legacy threshold	`match_threshold`	Legacy alias. It still takes precedence where supported, but new integrations should use stage-specific thresholds.
Evidence weights	Scalar aliases such as `nameWeight`, `postResolutionNameWeight`, `locationWeight`, `dobWeight`, `identifierWeight`	Aliases normalize into `candidateScoring.preResolutionWeights` and `candidateScoring.postResolutionWeights`. Shared aliases apply to both stages; stage-specific aliases override one stage.
Location component weights	`global_filters.match_scoring_runtime_config.candidateScoring.preResolutionWeights.locationAddress`, `locationCity`, `locationState`, `locationCountry`, and the same four keys under `postResolutionWeights`	Nested only: the four component keys exist only inside the nested runtime object, and no scalar alias stands in for a component (`addressWeight` or `address_weight` is the scalar alias for the whole `location` block; bare `address` is the subject’s address field, not a weight). `location` remains the legacy whole-block weight. If no component is set, that stage’s `location` is split evenly across the four. If at least one is set, every component left out resolves to `0`, the block weighs the sum of the components present, and `location` is ignored for that block. `location: 0.25` with `locationCountry: 0.4` therefore scores country at `0.4` with city, state, and address at `0`, a block total of `0.4`, not `0.5875`. Set all four explicitly to keep a mixed split. See Location component weights.
Missing evidence penalty	`missingEvidencePenalty`, `preResolutionMissingEvidencePenalty`, `postResolutionMissingEvidencePenalty`	Customer-facing 0 to 100 scale. 0 disables the penalty. Larger values penalize missing optional evidence more.
Name behavior	`transpositionTolerance`, `initialsTolerance`, `extraNamePenalty`, `textSimilarityFloor`, `phoneticsTolerance`, `lengthDifferenceWeight`, `hyphenatedNameTolerance`	Scalar aliases normalize into `nameMatching.person`.
Organization-name behavior	`baseNameWeight`, `fullNameWeight`	Scalar aliases normalize into `nameMatching.organization`.
DOB/date behavior	`personDateWindowDays`, `personMaxDateToleranceYears`, `organizationDateWindowDays`, `organizationMaxDateToleranceYears`, inferred-date aliases	Scalar aliases normalize into `dobMatching`.
Location behavior	`defaultLocationStrictness`, `strictLocationCutoff`, `allowNationalityForCountryMatch`	Scalar aliases normalize into `locationMatching`.
Merge behavior	`aiAssistedMergeThreshold`, `exactEvidenceMergeThreshold`, `enableStrongOrganizationMerges`, `strongOrganizationNameCutoff`, `strongOrganizationDateDays`	Scalar aliases normalize into `entityResolution`.
News retrieval controls	`news_geography_bias`, `news_name_filter`, `news_name_fuzzy`, `news_subject_mention`, `news_location_inference`	Applied before scoring, so they change which articles become candidates rather than how a candidate scores. `feed_filters.News.<key>` overrides the global value for the News feed.
Article retrieval limits	`max_lookback_days`, `max_requested_articles`, `deduplicate_media_dates`, `deduplicate_media_titles`	Bounds and de-duplicates retrieved News articles before scoring. Accepted in `global_filters`, and `feed_filters.News.<key>` overrides the global value, except for the two dedupe keys: a dedupe already enabled globally stays on, because the feed value is only consulted while the global value is off. `max_lookback_days` only affects article retrieval; it does not narrow watchlist feeds.
Article search controls	`news_max_lookback_days`, `news_max_requested_articles`	Used by article search requests, not by the News feed: `news_max_lookback_days` sets the lookback window and falls back to `max_lookback_days` when absent; `news_max_requested_articles` caps returned articles at 100. Neither changes the News feed request.
Open Source retrieval geography	`open_source_geography_bias`	Open Source (web) retrieval bias for the request. The legacy spelling `open_source_serp_geography_bias` is also accepted.

Applies to. POST /v1/search, POST /v1/search-sync, GET /v1/search-sync (URL-encoded JSON in the query string), and POST /interface/search (the risk-assessment interface route) accept these overrides: nested settings under global_filters, per-feed settings under feed_filters, and the top-level aliases noted in the table. POST /v2/search accepts the nested filters per entity only: put them inside each entities[] object, because a body-level global_filters or feed_filters is ignored on a v2 request.

If a request sends both a nested runtime object and scalar aliases, Minerva merges them into one runtime configuration before scoring. Omitted sections fall back to the workspace setting, then to Balanced defaults when no workspace value exists.

Example Direct API override:

{
  "global_filters": {
    "post_resolution_match_threshold": 0.78,
    "match_scoring_runtime_config": {
      "candidateScoring": {
        "preResolutionWeights": {
          "name": 1,
          "location": 0.25,
          "time": 0.2,
          "threshold": 0.78
        },
        "postResolutionWeights": {
          "name": 1,
          "location": 0.85,
          "time": 0.7,
          "threshold": 0.8
        },
        "gates": {
          "individualNameMinimumContribution": 0.6,
          "organizationNameMinimumScore": 0.88,
          "sanctionsPepThresholdOverride": 0.82,
          "sanctionsPepLocationWeightOverride": 0.1
        }
      },
      "nameMatching": {
        "person": {
          "positionWeight": 0.9,
          "initialsWeight": 0.7,
          "phoneticsWeight": 0.75
        }
      },
      "dobMatching": {
        "individualDateStdevDays": 365,
        "individualMaxDateToleranceYears": 10
      },
      "locationMatching": {
        "defaultStrictness": "city",
        "strictComponentCutoff": 0.9,
        "allowCountryMatchFromNationality": false
      },
      "entityResolution": {
        "mergeDecision": {
          "modelConfidenceThreshold": 0.72,
          "ruleBasedConfidenceThreshold": 0.72
        }
      }
    }
  },
  "feed_filters": {
    "Sanctions": {
      "pre_resolution_match_threshold": 0.9
    },
    "PEP": {
      "pre_resolution_match_threshold": 0.9
    },
    "News": {
      "pre_resolution_match_threshold": 0.85,
      "news_geography_bias": "strict",
      "news_name_filter": "strict",
      "max_requested_articles": 200
    }
  }
}

Scalar aliases are useful for legacy integrations, but the nested runtime object is easier to audit because related settings stay grouped by scoring behavior.

Country-dominant matching (worked example)

A customer screens a subject by name, date of birth, country, city, and state in the onboarding workflow. The request carries West Kelowna, BC, Canada. The candidate has the same name, no date of birth, and is recorded in Maple Ridge, British Columbia, Canada. The city does not match, while the state matches through the BC to British Columbia expansion and the country matches exactly. The date is not compared because the candidate carries none, but its absence still adds a missing-evidence charge. The city mismatch is the only failed comparison. Under the Balanced even split the four components carry 0.1875 each, but this request supplies no address, so the 0.75 block is averaged over the three components it does supply. Each carries 0.25 of the block’s weight, so the city miss removes 0.25 from the numerator, about 0.14 of the final score against the 1.85 denominator, while the matching country and state earn the same 0.25 each. A second miss would remove another 0.25. The candidate scores 0.811 against a 0.85 post-resolution threshold (the Balanced onboarding and ongoing-monitoring setting), so a record that should match is dropped. On the Balanced Direct API output threshold of 0.75 the same candidate clears the threshold even under the default split. On a Balanced workspace, give country most of the post-resolution block and keep the total at 0.75:

{
  "global_filters": {
    "match_scoring_runtime_config": {
      "candidateScoring": {
        "postResolutionWeights": {
          "locationAddress": 0,
          "locationCity": 0.05,
          "locationState": 0.05,
          "locationCountry": 0.65
        }
      }
    }
  }
}

The same request and the same candidate now score 0.919 against the same 0.85 threshold. Address is 0 because screening records rarely carry street-level evidence, and the four components still sum to 0.75, so the block’s total weight is unchanged; only its distribution changed. This example sets postResolutionWeights only. Pre-resolution scoring keeps the Balanced even split of its 0.10 location block; to make country dominant before resolution as well, set the four preResolutionWeights keys to a distribution that sums to 0.10. The 0.75 total is Balanced-specific: Narrow uses 1.00 for the post-resolution location block and Wide uses 0.50. A less aggressive variant keeps city and state at 0.10 each and gives country 0.55, still a 0.75 total. The same case scores 0.892: above the threshold, with more credit left for an accurate city. A country-only split is also valid: set country to 0.75 and address, city, and state to 0. Backward compatibility. A request or workspace that sets no component weight scores exactly as it did before the split: the single location weight is divided evenly across the four components, and a Balanced workspace keeps 0.1875 per component after resolution. An existing location weight needs no rewrite for the same reason. The block switches to component weighting only when at least one component weight is set in the request or the workspace. The full resolution and zeroing rules are under Location component weights.

Role-aware adverse media

Role-aware adverse media is a separate screening configuration that runs after adverse-media articles have been retrieved and analyzed. It helps identify whether a negative article appears to involve the screened subject or is likely only a name mention. Use Match Scoring when you need to tune candidate thresholds, candidate evidence weighting, matching behavior, merge behavior, or News retrieval and article-analysis behavior. Use Role-Aware Adverse Media when you need to show or filter confident article-subject relevance signals after adverse-media analysis has already run.

Get Started

User Guides

Training Videos

Key Concepts

How match scoring works

What a score means

Start from a preset

Change one workflow at a time

Find the setting you need

Test a change in the score simulator

Simulator steps

Reading simulation results

Calibrate in a controlled order

Review and save a change

Check history and roll back a change

Decision thresholds

Preset thresholds

Candidate score weights

Location component weights

Preset weight values

Candidate gates

Name matching behavior

Person names

Organization names

DOB and date matching behavior

Location matching behavior

Entity-resolution merge behavior

News controls

News geography filter

News name filter

Max requested articles

News location inference

Preset retrieval and merge values

Request-level overrides

Country-dominant matching (worked example)

Role-aware adverse media

​How match scoring works

​What a score means

​Start from a preset

​Change one workflow at a time

​Find the setting you need

​Test a change in the score simulator

​Simulator steps

​Reading simulation results

​Calibrate in a controlled order

​Review and save a change

​Check history and roll back a change

​Decision thresholds

​Preset thresholds

​Candidate score weights

​Location component weights

​Preset weight values

​Candidate gates

​Name matching behavior

​Person names

​Organization names

​DOB and date matching behavior

​Location matching behavior

​Entity-resolution merge behavior

​News controls

​News geography filter

​News name filter

​Max requested articles

​News location inference

​Preset retrieval and merge values

​Request-level overrides

​Country-dominant matching (worked example)

​Role-aware adverse media

​Related guides

How match scoring works

What a score means

Start from a preset

Change one workflow at a time

Find the setting you need

Test a change in the score simulator

Simulator steps

Reading simulation results

Calibrate in a controlled order

Review and save a change

Check history and roll back a change

Decision thresholds

Preset thresholds

Candidate score weights

Location component weights

Preset weight values

Candidate gates

Name matching behavior

Person names

Organization names

DOB and date matching behavior

Location matching behavior

Entity-resolution merge behavior

News controls

News geography filter

News name filter

Max requested articles

News location inference

Preset retrieval and merge values

Request-level overrides

Country-dominant matching (worked example)

Role-aware adverse media

Related guides