Tasks from
the filing.

Core references trace to cells in the filing.

grade.py scores Core. Verifiers format.

Get the sample

Sample task, 10-Q

Blackline, Inc.

Filed 5 Aug 2026. CIK 0001666134. Accession 0001628280-26-053353.

Was gross margin at least 80 percent for the quarter ended 30 June 2026?

Gross profit
$142,678,000
Revenue
$187,822,000
Gross margin
75.96%

Verdict no. The claim misses the filing.

Core references trace to cells in the filing.

Tickmark Bench checks answers against SEC EDGAR bulk XBRL. In the Core set, every reference traces to cells in the filing. Core is scored entirely by grade.py. Numeric answers are scored within the stated tolerance, and claim checks on the verdict alone. The official score is score = passed/149. grade.py's total is the official score.

Tasks ship in the Prime Intellect verifiers format. Built for AI labs and RL environment vendors. An agent can read this page. Core is $299 through Stripe Checkout on Pricing. The grading account is on How it's graded.

Core is 149 verified tasks, the real_v1 batch. The record of the pack is the datasheet.

1885 stock certificate of the Chicago, Burlington and Quincy Railroad, with an engraved border and share text. Jean-Francois Millet's print The Seaweed Gatherers, laid on hand-ruled ledger paper.

Four task types

What the pack checks today. A count of each type is not published.

Single-number. One figure, often a computed ratio or margin, from one filing period. Code checks it against a reference value taken from, or computed from, the filed XBRL facts, within a tolerance.

Multi-period and segment. A change across periods, or a figure from a segment. Code checks it against the figure computed from the filed XBRL, within a tolerance.

Claim checks. A yes or no on a statement about the filing. grade.py scores the verdict alone.

Comparisons across companies. The same metric for more than one issuer. Code checks it against reference values taken from, or computed from, the filed XBRL facts, within a tolerance.

Front of an 1880s NAHV stock certificate, with coupons and engraved lettering.

Solve rates

Solve rates: not yet published. We publish only measured numbers, with model, attempts and date.

First, the filing.

Every question is generated from bulk XBRL and kept only when the filing is from Q2 2026 or later.

Then the question.

Single-number, multi-period, segment, claim check, or a comparison across companies.

Numbers, in code.

Single-number, multi-period, segment, and cross-company answers are checked against a reference value taken from, or computed from, the filed XBRL facts. A model is not in that loop.

Then the verdict.

A claim check is a yes or no. grade.py scores the verdict alone. The explanation is not scored.

Worked checks, with the references from the sample, are on How it's graded.

Sources, the filing window, and the grading method are on the datasheet.

Who it is for

AI labs. Eval and RL on questions whose references come from filings dated Q2 2026 or later.

RL environment vendors. Tasks in the Prime Intellect verifiers format. grade.py scores the number within the stated tolerance and the claim check on the verdict alone.

Agents. The catalog, the free sample, and the prices are in llms.txt.

Get the pack

Download the 20-task sample

Five tasks of each type, in an earlier sample format. They are not part of Core.

See pricing

Core is 149 verified tasks in the real_v1 batch. Prices are on the pricing page.

Facts an agent can read

The machine-readable catalog is llms.txt. A JSON copy of these facts is in the script with id sec-task-pack.

Product
Tickmark Bench
Page status
Home page. The Blackline filing in the hero is a real task from the committed sample. Solve rates: not yet published.
Source
SEC EDGAR bulk XBRL. Not a product of the SEC.
Filing rule
Filings from Q2 2026 or later.
Solve rates
Solve rates: not yet published. We publish only measured numbers, with model, attempts and date.
Datasheet
datasheet.html
Task types
Single-number calculations. Multi-period or segment. Claim checks, with a yes or no verdict. Comparisons across companies.
Numeric grading
Code against a reference value taken from, or computed from, the filed XBRL facts, within a tolerance.
Claim checks
grade.py scores the verdict alone. The explanation is not scored. The official score is score = passed/149.
Format
Prime Intellect verifiers. Not a Prime Intellect product.
Free sample
20 tasks, five of each type, in an earlier sample format. They are not part of Core. The sample page is sample.html.
Pricing
pricing.html
Catalog
llms.txt

Pricing

Core, and what it costs.