Probily Public beta
Methodology/ how the report is built

The method is the product.

Probily's value isn't a verdict — it's a transparent, reproducible path from a URL to the public record. Here is exactly how a report is built, scored, and cited, end to end.

We don't tell you what's true. We show you where to check.

A value in a report is a pointer to a primary source, not an assertion. Where a record exists, we link to it. Where none exists, we say so. The reader stays the judge — the method just removes the busywork of assembling the evidence.

The pipeline

Five stages, fully documented.

01

Capture the source

The article URL is fetched and archived to the Wayback Machine, then hashed with SHA-256.

Wayback + SHA-256
02

Read into structured records

People, organisations and other named entities are extracted from the text and organised into a structured report.

deterministic · reproducible
03

Enrich across registries

Each entity is queried across the connected open, public registries in a single pass.

open sources only
04

Score & rank

Sources are weighted by reliability; each value gets a confidence and an Admiralty code.

LiftWing · Admiralty
05

Assemble a cited report

A deterministic report is rendered where values link to their primary sources.

same input → same report
Sources

Open sources, one catalogue.

Reports query open, public registries only — identity, sanctions and watchlists, courts, corporate and integrity registers, leaks and offshore records, academic and web archives. No proprietary or sealed sources. Every hit links to its originating record, and a hit means a query returned a record in that source at that time — nothing more.

Scoring

How a value earns its confidence.

Source reliability

Wikipedia sources are weighted by their WP:RSP standing and Wikimedia's LiftWing quality models — reliable outlets count for more than fringe ones.

Generally reliableNo consensus
Admiralty code

Each corroborated value carries a NATO Admiralty rating — source reliability (A–F) and information credibility (1–6) — the standard analysts already use.

A1B2C3
Confidence

Extraction and match confidence combine into a single gauge per value — so a name-only sanctions match reads very differently from a QID-confirmed one.

92%
54%
What a hit means
A hit is a query that returned a record in database Y at time T — not a conviction, a sanction, or a determination about a person.
Built from records, not written as prose
The report is assembled deterministically — the same article yields the same structured report, and its values trace back to source records rather than to generated prose.
Provenance

Evidence that can't quietly change.

At report time the source page is captured to the Wayback Machine and hashed with SHA-256, computed from the captured HTML. If the page is later edited or deleted, your report still points at the version you actually read.

Every report also has a public authenticity check: paste its ID at /verify to confirm it is a genuine Probily report and when it was generated.

Wayback
web.archive.org/web/20240805182004
source captured at report time
2024-08-05 18:20 UTC
SHA-256
af3c1029b7e5…7d201bbf
hash of the captured HTML
immutable
Standards

Aligned with the field's standards.

Aligned · not a signatory

Journalism Trust Initiative (JTI)

Our transparency and sourcing practices are built to align with JTI's indicators. We are not a certified signatory, and we don't claim to be.

Aligned · not a signatory

IFCN principles

We follow the spirit of the IFCN code — transparency of sources and method, and an open corrections path — without claiming membership.

Not a consumer-reporting agency. Probily reports are not "consumer reports," and Probily is not a consumer-reporting agency. Do not use Probily for decisions covered by the FCRA — employment, credit, housing, or insurance eligibility.

Limits

What the method can't do.

Extraction

The text is read automatically; misattribution and missed names are possible. Cross-check before relying on a name.

Coverage

Open registries only. Proprietary and sealed sources are out of scope — absence of a record is not proof of absence.

Freshness

Registry data reflects the last sync; upstream records change. Re-run the report for the latest state.

Language

Non-English articles and transliterated names can reduce extraction accuracy and match quality.