Tickmark Bench
Menu
Get the sample

An XBRL question answering dataset from 2026 SEC filings.

Tickmark Bench Core is 149 verified tasks from 67 companies. Each question is answered from tagged XBRL facts in a 10-Q or 10-K, and each task lists the facts behind its reference answer.

What XBRL gives a benchmark

Public companies tag the numbers in their financial statements with XBRL. Each tagged fact has a concept, such as us-gaap:AssetsCurrent, a period, a unit, and any dimensions, such as a segment. The SEC publishes these facts in bulk as the Financial Statement Data Sets. Tickmark Bench builds its tasks from those files.

Because the reference comes from tagged facts, the answer key does not depend on reading a table by eye. A Core task is kept only when every reference fact matches the filing's XBRL instance exactly: exact period dates, dimensions, and unit.

What a fact looks like

This is one fact from the Momentus Inc. task in the free sample, written out as a list.

  • concept: us-gaap:AssetsCurrent
  • value: 113112000
  • unit: USD
  • period_end: 2026-06-30, duration_quarters: 0, a balance at that date
  • segments: none
  • accession: 0001628280-26-055624, a 10-Q filed 2026-08-11
  • entity: Momentus Inc., CIK 0001781162

The task divides this fact by current liabilities at the same date, 8254000 USD, and asks for the current ratio to 2 decimal places. The reference is 13.70.

What the questions ask

Four task families are generated today. single_number asks for one figure, often a ratio or margin. multi_period asks for a change across periods, or a figure from a segment. claim_check asks for a yes or no verdict on a stated claim. comparison asks for the same metric for more than one issuer.

How answers are graded

Numeric answers are checked in code against a reference value taken from, or computed from, the filed XBRL facts, within a tolerance. The unit and the scale have to match the question. Core claim checks are scored on the verdict alone. Core is scored entirely by grade.py.

What you get

Core is one zip: tasks.jsonl, grade.py, README, datasheet, license and checksums. Core is $299 one-time, on the pricing page. The free sample has 20 tasks, five of each type, in an earlier sample format. The full field list is on the Rights page, and the filters are on the datasheet.