Data quality

Every figure tells you how much to trust it

We check 28,319 statements against an independent second source. What disagrees gets flagged — not hidden.

curl "$API/v1/fundamentals/PSM/statements?stmt=BS"

{
  "stmt": "BS",
  "crossChecked": false,
  "crossCheckReason": "BS/2024: 5.61e+03 vs 5.78e+09 EUR (factor 9.7e-07)",
  "warnings": ["cross_source.disputed"],
  "structureConfidence": "low"
}

What you can do with it

Exclude disputed valuescrossChecked: false plus crossCheckReason names the discrepancy concretely — both numbers and the factor.
See how a figure came aboutstructureConfidence distinguishes XBRL tagging from a table parsed out of a PDF.
Spot derived statementsderived: true marks income statements that were split out of a combined statement of comprehensive income.

Why we built it this way

A wrong figure does more damage than a missing one, because nobody notices it. So we check against a separately sourced pool — total assets and revenue, at matching reporting currency.

The label deliberately says “disputed”, not “wrong”. Some hits are definitional differences: insurers carry separate-account assets in total assets under US-GAAP, data vendors strip them out. Both figures are correct. Labelling that an error makes users throw away good data.

For the same reason we do not check on revenue alone: Deutsche Börse reports net and gross revenue separately, both correct. A touchstone only works if both sides mean the same thing.

Improvements

Each document now carries its own quality mark instead of just an aggregate
The check runs on every import, no longer only once
Definitional differences are told apart from real errors
Assumed values are flagged when the report stated no unit

Endpoints

GET	`/v1/fundamentals/{identifier}/coverage`	Pools, range, metric coverage
GET	`/v1/fundamentals/mapping/gaps`	Most frequent unmapped lines

Give feedback or ask a question

Something missing, something off, or an answer you don't understand? Write it here — it reaches us directly.