Risk inference helps Minerva turn relevant, sourced text discovered during a search into explainable screening and Client Risk Rating signals. It complements direct watchlist and configured-list matches; it does not replace them.
Access: Requires the Admin role or above. In the sidebar, go to Administration > Configuration, then open Risk Inference under Screening behaviours.
Use this guide when you need to:
- understand the difference between list-backed and text-derived risk signals
- see when PEP, Criminal, and High Risk Industry inference run
- tune a workflow that is producing too many false-positive signals
- add local terminology or suppress a recurring phrase without writing regular expressions
- configure PEP terms separately for tiers 1 through 4
- audit or roll back a Risk Inference change
Keywords is the default strategy for PEP, Criminal, and High
Risk Industry inference in all four screening channels. Risk Inference is
workspace-scoped, so a Calibration workspace can be tuned without changing
Live.
Turning an inference off can reduce analyst-visible signals or Client Risk
Rating evidence. Change one channel at a time, record why the change was made,
and validate representative true positives as well as false positives.
What Risk Inference Does
Minerva combines structured records with public information discovered while a search is running. Open Source Intelligence (OSINT) is sourced public-web information. When Open Source is requested, Minerva retrieves that information in real time and carries sourced occupation, organization, notes, document titles, industry, employer, and other business context into the resolved profile.
Risk inference evaluates resolved profile text from all applicable sources. Open Source supplies most newly discovered live-web context, but it is not the only possible source of text. Minerva does not treat every web mention as fact. A keyword finding is an explainable risk signal that analysts should review with the source, identity evidence, and surrounding context.
The three inference types have different outcomes:
| Inference type | What it looks for | Effect when matched |
|---|
| PEP | Language indicating that an individual holds or held a qualifying public role. | Flags the result as a potential PEP match, can assign the applicable PEP tier, and can contribute to the Inferred PEP Client Risk Rating criterion. |
| Criminal | Language indicating criminal events, allegations, charges, or convictions. | Flags the result as a potential Criminal match and can contribute to the Inferred Criminal Client Risk Rating criterion. |
| High Risk Industry | Language indicating that a person or organization is connected to a higher-risk business category. | Adds evidence to the High Risk Industry criterion in Client Risk Rating. It does not create a PEP- or Criminal-style potential match. |
A Criminal keyword finding is not a determination of guilt or verified
wrongdoing. Analysts should review the identity match, source reliability,
date, legal context, and disposition of the matter.
When Each Inference Runs
There are two conditions to keep distinct:
- the feed gate that allows an inference engine to run
- the feed combination that supplies the most useful real-time text
| Inference type | Engine condition | Feed combination with the clearest effect | Subject types |
|---|
| PEP | The PEP feed must be requested and the selected channel must use Keywords. | PEP + Open Source. PEP alone is primarily a watchlist/list search; adding Open Source lets Minerva evaluate sourced public-role text found while exploring the web. | Individuals |
| Criminal | The Criminal feed must be requested and the selected channel must use Keywords. | Criminal + Open Source. Open Source can contribute sourced notes and document titles that describe criminal events beyond direct Criminal list evidence. | Individuals and organizations |
| High Risk Industry | The selected channel must use Keywords and the returned profile must contain applicable business, industry, employer, occupation, or related context. | Ownership + Open Source for organizations is the main intended combination because it supplies business classification and ownership context. This is not a hard feed gate. | Primarily organizations; also employer or occupation context on individuals |
If the objective is a list-only PEP check, PEP without Open Source keeps the
search focused on watchlist evidence. If the objective includes discovering
public-role context from the live web, request PEP and Open Source together.
| Channel | What it covers |
|---|
| Onboarding | First-pass screening while a new customer or profile is being onboarded. |
| Ongoing monitoring | Recurring searches for existing profiles. Changes can affect alert volume across the monitored population. |
| Direct API calls | API-driven screening submissions associated with the workspace. |
| Risk assessments | Searches run inside risk assessment and due-diligence workflows. |
Main Configuration Page
Each channel has its own PEP, Criminal, and High Risk Industry strategy. You can copy one channel into another or apply the active channel to all four after validating it.
Strategy Settings
Each inference currently supports two strategies:
| Strategy | PEP and Criminal effect | High Risk Industry effect |
|---|
| Keywords | Evaluates sourced text using Minerva’s built-in term library plus the workspace’s additions and suppressions. A finding can flag the potential match and contribute the corresponding inferred CRR factor. | Evaluates applicable business text and can add the High Risk Industry factor and annotation to Client Risk Rating. |
| None | Stops text-derived potential-match flagging and the corresponding Inferred PEP or Inferred Criminal CRR contribution for that channel. Direct watchlist and configured-list results remain active. | Stops the High Risk Industry factor and annotation from being added to Client Risk Rating. All other rating criteria remain active. |
None is not a global screening off switch. It changes only
the selected inference type in the selected workspace and channel.
Default Values
New and previously unconfigured workspaces use Keywords for all three inference types in all four channels. The built-in library is versioned and displayed in the dashboard with a Minerva defaults badge.
The tables below summarize the current default categories and representative built-in terms. The terms shown on the configuration page are the authoritative library for the selected workspace.
PEP Defaults by Tier
PEP has one fixed category for each tier. Tier 1 represents the most senior public exposure; tiers 2 through 4 cover progressively more regional, state-linked, and local roles. If text matches more than one tier, Minerva retains the highest applicable exposure level.
| PEP tier | Default category | Representative built-in terms |
|---|
| 1 | National and international leadership | prime minister, supreme court justice, federal judge, member of parliament, foreign minister, secretary of defense, dictator |
| 2 | Regional and senior public office | ambassador of, chief justice, governor of, senator, attorney general, premier of |
| 3 | State-owned organizations and agencies | director of a crown corporation, state-owned corporation, federal agency, government corporation |
| 4 | Local public office | mayor of, municipal judge, city council, inspector general, special prosecutor |
Criminal Defaults
The default Criminal events and allegations category includes terms such as arrested, incarcerated, convicted, charged with, wanted for, human trafficking, bribery, corruption, organized crime, tax evasion, and terrorism.
Admins can add more Criminal categories when a local policy needs separate terminology or review ownership.
High Risk Industry Defaults
High Risk Industry starts with these categories and representative terms:
| Default category | Representative built-in terms |
|---|
| Gambling and gaming | casino, lottery, sports betting, gambling |
| Money services, payments, and exchanges | money transfer, remittance agency, currency exchange, cryptocurrency |
| Cannabis and recreational drugs | cannabis, marijuana, dispensary |
| High-risk lending and offshore finance | payday lending, offshore lending, unsecured loan, unregistered fund |
| Precious goods, art, and luxury assets | gold bullion, precious metal, fine art, auction house, yacht |
| Vehicle, vessel, and travel dealers | used car, car dealer, travel agency, boat sales, vehicle dealer |
| Weapons and defence | arms dealer, firearms dealer, defense contractor |
| Charities and non-profits | non-profit, not for profit, fake charity, unregistered charity |
| Adult entertainment | adult entertainment, strip club, pornographic website, unlicensed massage |
| Property, holding companies, and private capital | real estate, private equity, venture capital, holding company |
Editing Keyword Logic Without Regex
The expression builder is designed for compliance users. It stores structured, plain-language rules rather than asking an admin to enter regular expressions.
For each category, you can:
- review the active Minerva built-in terms
- add a local term or phrase
- suppress a built-in term or add a false-positive suppression
- choose whether any or all added terms must match
- choose how an added term is recognized
- restore the Minerva defaults
| Match option | Meaning | Example |
|---|
| Contains phrase | Finds the words together in the entered order. | charged with matches “was charged with bribery.” |
| Whole word | Finds the complete word, not the same letters embedded in another word. | mayor matches “mayor” but not a longer word that only contains those letters. |
| Word starts with | Finds words beginning with the entered text. | fraud can match “fraud” and “fraudulent.” |
student senator in PEP tier 2 or charged with overseeing in Criminal inference after confirming that those phrases are producing irrelevant hits.
When to Tune Risk Inference
Tune from evidence rather than from one unusual result.
| Pattern observed | Recommended first action |
|---|
| PEP potential matches repeatedly come from a harmless phrase or role that does not meet policy. | Add a narrow suppression in the affected PEP tier and channel. Keep the other tiers unchanged. |
| Criminal potential matches repeatedly describe a non-criminal use of a phrase. | Add a phrase-level suppression to the relevant Criminal category. |
| A local public role or criminal expression is consistently missed. | Add the narrowest accurate phrase to the applicable tier or category. |
| High Risk Industry is raising CRR scores for a business classification your policy does not treat as high risk. | Suppress the specific phrase or tune the affected category. Confirm that other legitimate HRI classifications still contribute to the rating. |
| False positives are widespread and cannot be isolated to a small set of terms. | Use None temporarily for the affected inference and channel while the team calibrates. Re-enable Keywords after representative testing. |
- collect a representative set of false positives and known true positives
- identify the exact category, term, source field, and channel responsible
- make the smallest term or suppression change that addresses the pattern
- test the changed channel in a Calibration workspace
- compare result volume, true-positive retention, and CRR changes against the baseline
- review the change summary and save a clear reason
- switch to the Live workspace, deliberately reapply and save the reviewed settings, then monitor the affected workflow
Avoid broad suppressions such as a country, common job word, or generic legal
term. They can hide unrelated true-positive signals. Prefer the longest phrase
that describes the known false-positive context.
Review, History, and Rollback
Risk Inference changes use the same workspace deployment controls as other tenant configuration sections.
Before saving, Review changes shows:
- each affected channel and inference strategy
- added, changed, or removed terms and suppressions
- category and match-rule changes
- an optional change description
The history page records who changed the configuration, when it changed, the affected sections, and the saved description. Rollback restores a prior snapshot by creating a new history entry; it does not erase the audit trail.
