> ## Documentation Index > Fetch the complete documentation index at: https://docs.dataspike.io/llms.txt > Use this file to discover all available pages before exploring further. # How Risk Scoring Works > How Dataspike turns a screening hit into a percentage and a band — walked through three worked examples. Every screening hit comes back with a **percentage** and a **band**. This page explains how we get from “what we found” to that number, then walks through three cases. ## How the score works The **band** — Low, Medium or High — is the decision signal. It drives alerts, and a **High** fails an automated identity check outright. The **percentage** is the detail behind it: how much corroborating evidence sits inside that band. | Band | Range | | - | - | | 🟢 **Low** | 0 – 29% | | 🟡 **Medium** | 30 – 69% | | 🔴 **High** | 70 – 100% | **Rule one — the band is the single worst finding.** One sanctions listing makes a case High, and nothing in the record talks it back down. Equally, twenty minor findings stay Low — volume never escalates a band. **Rule two — the percentage moves only inside that band.** Extra findings of the same severity raise the number with sharply diminishing returns, and it is capped so it can never spill into the band above. ### How factors combine Each risk factor starts from a base weight set by its severity: | Severity | Base weight | | - | - | | 🔴 High | 0.70 | | 🟡 Medium | 0.30 | | 🟢 Low | 0.05 | | Unknown | 0 | Factors are combined probabilistically: two factors with weights *a* and *b* give `1 − (1 − a) × (1 − b)`. Repeated factors of the same severity are scaled down logarithmically, so each one adds less than the one before — that is why two Low factors add 5.5%, not 10%. The result is then capped at the top of the band set by the most severe factor. > [!WARNING] > > A clean search reports **Low**, not blank. Zero matches and a minor finding look alike at a glance, so read the hit count alongside the band. ## Worked examples Cases 1 and 2 are constructed illustrations. Case 3 follows production scoring of a real, heavily sanctioned record, anonymised here. Percentages are rounded. ### Case 1 · Country risk alone *A clean individual whose only exposure is where he is from.* **John Doe** `Person` `System default profile` One hit, on a national company register — he is listed as a director. No sanctions, no adverse media, no enforcement history. His citizenship is a jurisdiction on the **FATF grey list**. | Finding | Factors | Severity | | - | :-: | - | | Company register entry *(the register itself is a low-risk source)* | 1 | 🟢 Low | | “Company registry” label *(labels are scored separately from the source)* | 1 | 🟢 Low | | Citizenship — FATF grey list *(default rulebook: grey list = Medium)* | 1 | 🟡 Medium | | **Total** | **3** | | `Medium 30% · Low 5.5% (2 factors) → 1 − (1 − 0.30) × (1 − 0.055) = 0.34` > [!NOTE] > > 🟡 **34% · Medium** — Nothing in his own history is serious. The grey-listed jurisdiction alone lifts him out of Low, which is exactly what a grey listing is meant to signal. > [!TIP] > > We read the FATF lists live, so a country moving on or off the grey list changes this score on the next screening — no configuration change needed. ### Case 2 · The client’s own country rule *The same clean record, but the client has told us which jurisdictions they will not accept.* **John Doe Inc.** `Company` `Custom profile` A Venezuelan-registered company with one hit, on a national company register. Nothing adverse at all. This client’s rulebook sets Venezuela to **High for companies** and Medium for individuals. | Finding | Factors | Severity | | - | :-: | - | | Company register entry | 1 | 🟢 Low | | “Company registry” label | 1 | 🟢 Low | | Country — Venezuela *(client rulebook, company rate — a rated country replaces the FATF reading entirely)* | 1 | 🔴 High | | **Total** | **3** | | `High 70% · Low 5.5% (2 factors) → 1 − (1 − 0.70) × (1 − 0.055) = 0.72` > [!NOTE] > > 🔴 **72% · High** — A spotless company scores High purely on the client’s own country policy. Run the identical record as an individual and the same rulebook returns Medium instead. > [!TIP] > > This is the strongest lever a compliance team has. Country policy is set per client, separately for individuals and companies, and it overrides the FATF default for that country. ### Case 3 · A saturated record *A heavily sanctioned individual, screened on the default profile — what the ceiling looks like.* **Jane Roe** `Person` `System default profile` Matched at full-name confidence against 13 lists, with three citizenships. We never score a record whole — it is broken into individually rateable **risk factors**, and this one yields 37. | Finding | Factors | Severity | | - | :-: | - | | Government sanctions listings *(EU, UK, Switzerland, Canada, Australia and other regimes)* | 12 | 🔴 High | | “Sanctions” label on each of those | 12 | 🔴 High | | Disqualified directors register | 1 | 🟡 Medium | | “Finance” and “Legal” labels | 9 | 🟢 Low | | Citizenships *(none rated above Low for an individual under the default profile)* | 3 | 🟢 Low | | **Total** | **37** | | `High × 24 factors · Medium × 1 · Low × 12 → capped at 1.00` > [!NOTE] > > 🔴 **100% · High** — Twelve independent government listings, each High on its own. The score saturates: as unambiguous as a screening result gets. Two things are worth noticing. The score reaches the 100% cap well before the last listing is counted — repeated findings have diminishing returns, so the tenth sanctions list cannot count as much as the second. And geography contributed almost nothing: the default profile sets no country policy of its own, so the citizenships were read straight from FATF and all came back Low. Case 2 is what happens when a client changes that. ## The rulebook behind the numbers None of these severities are hardcoded. Each client has a **risk profile** deciding what counts as serious. The default ships with these positions: | Rulebook entry | Default | | - | - | | Sanctions, Terrorism, Criminal labels | 🔴 High | | PEP, Leaks, Debarment, Regulatory warning labels | 🟡 Medium | | Finance, Legal, Media, Company registry labels | 🟢 Low | | FATF blacklist country | 🟡 Medium (person) · 🔴 High (company) | | FATF grey-list country | 🟡 Medium (both) | | Country policy | Not set — FATF reading is used | | Per-list severity | Not set — each list’s own rating is used | ### What overrides what Severity comes from the risk profile first and from the data itself second: | When… | …the severity comes from | | - | - | | A source (list) is configured in the risk profile | The profile setting, instead of the list’s own rating | | A source is not configured | The list’s own rating | | A country is configured in the risk profile | The profile setting, instead of its FATF status | | A country is not configured | Its FATF status: blacklist High, grey list Medium | | A country is neither configured nor on a FATF list | Low | | A source and its tag are rated differently | The source’s rating sets the bound | Country factors are read from citizenships, nationalities and addresses. Tags are read from source tags only. > [!TIP] > > Source settings are the most powerful part of a risk profile: they replace each list’s own rating, so this is where your organisation’s own view of a list belongs. > [!NOTE] > > Profiles are versioned, and every result records which profile and version produced it — so a score from six months ago can always be explained. ## Worth knowing * **Volume never escalates a band.** Fifty Medium findings still read Medium. Deliberate — it stops a noisy record manufacturing a High — but it means the band alone doesn’t convey how much evidence sits behind it. That is the percentage’s job. * **Only strong name matches raise the case.** Every matched record gets its own score, but it only lifts the overall search result if the name matched at high confidence. A High-risk record on a weak name stays visible to the analyst without escalating the case. * **Labels are read per listing.** A record carries a rolled-up label summary, but only the labels attached to each individual listing are scored. A label that appears in the summary and nowhere on a listing adds nothing. ## Next steps > **[AML Search](/aml-screening/search)** > > Run one-time AML searches and read the same score. > **[Daily Monitoring](/aml-screening/monitoring)** > > Get alerted when a score changes. > **[API Reference](/api)** > > All parameters and response formats. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.