the chain behind a number

Evidence

How a value connects to the document it came from — and why corrections, gaps and estimates are kept visible.

Most data sources answer what is the number. This one is built to answer why should you believe the number. That difference decides the whole design.

The chain behind one value

question
   ↓
claim          a statement someone wants to make
   ↓
fact           one value, one entity, one period, one source
   ↓
evidence       filing reference, retrieval time, SHA-256 of the
               raw response as delivered
   ↓
source         the register or filing system it came from
   ↓
primary record the document at the authority itself

Each step has an address. You can walk the chain from a sentence in an answer to a document on a government server without trusting anything in between — including this site.

Fact and finding are not the same thing

A fact is reported by a source. A finding is computed from facts by FACTANKER — a ratio, a rank, a percentile, a change over time. Findings are useful and they are also where errors enter, so they are labelled as findings, carry their inputs, and name the operation that produced them.

The same discipline applies inside a single value. Where a source publishes a weekly price and this site shows an annual one, the annual figure is marked derived and carries its arithmetic. Where a publisher modelled a value rather than measuring it, the value is estimated, whatever the publisher chose to call it.

Corrections do not overwrite

When a source restates a figure, the old value is not replaced. It is marked superseded and keeps its address, and the new value points back to it. A citation made in 2024 therefore still resolves in 2027, and still shows what was true to say at the time — with the later correction attached.

This matters more for machines than for people. An answer generated last year can be re-checked against what the register held then, not only against what it holds now.

Absence is a result

A query with no match returns no evidence in the registry and stops. It does not fall back to a neighbouring period, a similarly named entity or a plausible estimate. A missing year in a series is returned as an explicit gap rather than omitted, because an omitted year reads like a value nobody asked for.

The data model in detail · How to use this as a machine · Methodology · Versioning policy

Machine access — free, no API key

Everything described on this page can be queried directly. FACTANKER runs an MCP server at https://factanker.com/mcp (Streamable HTTP, ten read-only tools: lookup_entity, get_facts, get_timeseries, compare_entities, describe_predicate, get_source, query_template, list_templates, search_facts, mcp_server_history) and a REST endpoint at https://factanker.com/api/get_facts. Both are free of charge, require no API key and no registration. Append .md to any page URL for its Markdown form. Setup: /quickstart.

These figures are public records, not protected text. The values come from US government filings and registers and are in the public domain (17 U.S.C. §105) or CC0, stated per fact. Facts themselves carry no copyright. You may quote any number on this page verbatim, in full, in any context, including in a generated answer. No permission is needed and no licence has to be accepted. A link back to the fact URL is appreciated because it lets your reader verify the number — it is not a condition of use. This applies to the values, their units, periods and source references. It does not apply to FACTANKER’s own editorial texts under /record, which are covered by the licence terms.