How Onsight scores entities.

Overview

Onsight operates on a simple principle: every claim about an entity must be traceable to an evidence item, and every evidence item must be tiered by source reliability. Scores are deterministic outputs of weighted sub-scores; they are not opinions and can always be inspected and reproduced.

The four primitives in Onsight are:

• Entities — companies, organisations, or corporate persons we track.
• Findings — alerts, litigation records, regulatory penalties, operational incidents, adverse media, sanctions hits.
• Evidence — the specific source artefacts (registry pulls, court records, articles, photos) that substantiate each finding.
• Scores — deterministic summaries of the current state: an overall Integrity Index plus five category sub-scores.

Integrity Index

The Integrity Index is a 0–100 score where higher means lower risk. It is a weighted combination of the five sub-scores:

Interpretation bands:

Sub-scores

Each sub-score is a 0–100 rating of the entity against documented findings in that category, updated whenever a new finding is ingested.

ESG (15% weight)

Environmental, Social, and Governance indicators outside the Labor and Governance categories — covers environmental incidents, community relations, supply-chain labor exposure, and board diversity. Drawn from operational incident records, regulatory actions, and published ESG disclosures.

Labor (25% weight)

Labor practice violations, wage-payment disputes, safety incidents, union actions, and regulatory penalties against worker protections. Weighted heavily because labor issues are the single most common source of reputational and operational risk in our coverage geography.

Political (20% weight)

Political exposure — PEP designations, state-affiliated ownership, government contract dependency, jurisdictional stability, and contested regulatory environments.

Sanctions (25% weight)

Direct or adjacent exposure to OFAC SDN, UN, EU, HMT, and other global sanctions lists. Includes ownership-chain analysis through known sanctioned jurisdictions (Russia, Iran, DPRK, etc.). The most aggressively weighted sub-score because the legal consequences are absolute.

Governance (15% weight)

Corporate governance signals — proxy ownership structures, nominee directors, opaque beneficial ownership, litigation patterns, regulatory non-compliance, and management turnover irregularities. Where the jurisdiction has no UBO disclosure regime, a voluntary regime, or a closed registry, the Governance sub-score reflects available documentary evidence only — registered directors, disclosed shareholders, and publicly available records — and is qualified accordingly on the entity profile.

Jurisdiction context

Every entity in Onsight exists within a regulatory environment that determines what information is structurally accessible through official channels. Onsight maintains a Jurisdiction Transparency Profile for every covered market, capturing the UBO disclosure regime, registry accessibility, FATF status, enforcement quality, and structural exemptions.

This jurisdiction data affects the entity profile in three ways:

1. Score interpretation. When an entity is assessed using documentary evidence only in a jurisdiction with limited UBO transparency, a jurisdiction ceiling indicator appears alongside the Integrity Index. This signals that the score reflects available evidence but may not capture risks that are structurally hidden from documentary sources. The score itself does not change — the qualifier contextualises it.
2. Verdict context. The Decision Posture verdict distinguishes between entity-specific gaps (things this entity could or should have disclosed) and jurisdiction-level gaps (structural limitations that no entity-level investigation through official channels can overcome).
3. Coverage assessment. The Coverage panel shows what documentary sources can and cannot establish in the entity's jurisdiction, informed by the jurisdiction's UBO regime, registry type, and enforcement environment.

Confidence on a documentary-only read is structurally capped: in jurisdictions with low transparency it cannot exceed Low; in jurisdictions with moderate transparency it cannot exceed Medium. Field intelligence releases the cap because field collection has addressed the structural gap.

Jurisdiction data is maintained by Cascade analysts and reviewed three times per year (January, May, September), with out-of-cycle updates triggered by FATF status changes, registry-access changes, or new UBO threshold legislation. It is derived from primary legislation, FATF mutual evaluations, and Cascade's own field experience with registry and regulatory access in each market. The current date of last review is displayed on every entity profile.

Evidence tiers

Every finding in Onsight is anchored to one or more evidence items. Each evidence item is graded by source reliability on a four-tier scale:

A finding supported only by T4 sources is reported with an explicit confidence caveat. Findings combining T1/T3 sources are considered corroborated.

Source grading (A1–F6)

Some surfaces in Onsight — notably the citation chips on individual facts (see Audit Mode on any entity profile) — carry a second grade alongside the evidence tier: a Source grading, expressed as a letter-number pair like A1, B2, or C3.

The two systems measure different things. Evidence tier is the structural type of source (registry vs. published vs. field vs. HUMINT). Admiralty Code is, for a specific source: how reliable it is historically (A–F) and how accurate this particular piece of information appears to be (1–6). They are independent dimensions and should be read together.

A T1 registry filing might still receive a B2 grade if the registry has a history of occasional clerical errors and the specific filing has not been independently corroborated. Conversely, a T4 HUMINT report might receive an A1 grade if the source has been historically reliable and the specific information is independently confirmed.

Confidence levels

Each entity carries an Intelligence Confidence rating reflecting the diversity and tier mix of the underlying evidence:

Freshness grading

An entity's freshness is the age of its most recent analyst touch — an ingestion, review, or validation event. Freshness is color-coded on every entity and surfaced on list views:

Audit trail

Every data point in Onsight carries two timestamps:

• observed_on — when the source artefact was dated.
• collected_by / last_touched_by — which Onsight analyst acquired or last reviewed it.

Scores record their weights at time of calculation, so a score from 6 months ago can be reproduced exactly even if the weighting profile has since changed. Click any score in Onsight to see its full history, contributing findings, and the evidence that moved it.

Source redistribution rights

Onsight ingests evidence from many source types — public registries, paywalled databases, licensed feeds, government documents requiring clearance, sealed court records, and Cascade-commissioned HUMINT. Each carries different legal restrictions on redistribution to clients, even when the underlying fact is the same.

Onsight's audit-grade pitch requires we hold a complete evidence trail — every fact backed by a captured artefact and content hash. But we cannot legally redistribute every captured artefact to every client. Our standing posture: capture everything; display calibrated to legal rights; produce full artefacts on legal demand. This is the same posture every serious investigative firm has always operated under, and it is litigation-defensible.

Every citation in Onsight carries one of three redistribution tiers:

Tier 1 — Public

Source is freely accessible (government registry, public news, court filings on free dockets, sanctions lists, press releases). The full captured PDF/screenshot + content hash is embedded in the citation drawer. Buyer sees exactly what we saw at retrieval. No restrictions on redistribution.

Tier 2 — Excerpt-only

Source requires a paid subscription, has copyright restrictions, or has terms of service that prohibit redistribution. Examples include Bloomberg terminal data, Dun & Bradstreet reports, Refinitiv internal databases, paywalled news archives, paid academic databases.

For these sources, Onsight captures the full artefact internally for our own audit trail, but only displays a fact-relevant excerpt (typically 1–3 lines, the exact passage supporting the claim). Excerpting for purposes of analysis, criticism, and fact-reporting falls within fair use (US) and equivalent doctrines in other jurisdictions. The buyer also sees the citation metadata (source name, retrieval date, content hash, Source grade) and a link to the original. Clients with their own subscription to the source can access the full document via that route. Cascade can produce the full captured artefact under proper legal process — subpoena, regulatory inquiry, or other valid demand.

Tier 3 — Attestation-only

Source we cannot capture or redistribute at all. Examples: government documents requiring local-jurisdiction clearance, sealed court records, documents accessed under NDA, HUMINT debriefs where the source's identity must be protected, hardcopy registry filings viewed in person.

For these citations, the analyst's signed attestation is the audit artefact. The attestation contains: analyst name and code, date and location of viewing, document type, the specific facts confirmed, and method of access. The attestation itself is content-hashed and stored in the Onsight provenance vault. The buyer sees a "Cascade-attested" badge with the analyst's name, date, and the fact attested to — without source-identifying details when the source must be protected. The Cascade analyst is available for sworn deposition or signed declaration under proper legal process.

What this means in practice

An Onsight Read is the integrity layer between restricted sources and the buyer's decision. Our job is to attest, grade, and produce evidence under proper process — not to republish copyrighted or restricted material. This framing is how serious diligence has always operated: investigative firms hold evidence, attest to its existence and content, and produce it under legal demand. Onsight digitises that posture and adds machine-grade reproducibility (timestamps, hashes, audit logs).

Appeals & corrections

If an entity's representative disputes a finding, Onsight supports structured correction workflow:

1. Subject submits a correction request citing the specific finding ID and disputed fact.
2. Onsight analysts review within five business days; correction is either applied, declined with rationale, or opened as an investigation.
3. All correction actions are logged to the entity's audit trail and remain visible to clients with appropriate access.

Onsight does not remove findings; we supersede them with updated, corrected records. Historical findings remain inspectable.

Versioning

This methodology is versioned. Changes to weights, tier definitions, or confidence criteria are announced 30 days in advance and published here with effective dates.

Methodology v1.0 · Cascade Asia · Past the last data point.