10-Q
Blackline, Inc.
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.
Tasks from
the filing.
Graded on the filing, from Q2 2026 on.
Checked against the filed XBRL. Graded in code.
Get the sampleGraded questions from filings a model has not memorized.
Tickmark Bench is graded questions generated from SEC EDGAR bulk XBRL. Filings are from Q2 2026 or later, so the figures are not ones an earlier model would have memorized. A number answer is checked in code against the filed value, within a tolerance. A claim-check explanation is scored by a small decision-model judge, calibrated against known-good and known-bad answers, and a frontier model spot-checks that judge.
Tasks ship in the Prime Intellect verifiers format. Buyers are AI labs and RL environment vendors. An agent can read this page and buy. The longer grading account is on How it's graded.
Example data. 48,200 tasks, filed through .
48,200
Example tasks in the mockup pack. The example window runs from through . Example data. The rule that stays is Q2 2026 or later.
Four task types
Counts are example data. The definitions are the product.
Single-number calculations. Margins, ratios, and other figures from one period of XBRL. The answer is one number, such as a gross margin. Code checks it against the filed XBRL value, within a tolerance.
19,280
Multi-period and segment. A change across periods, or a number taken from a segment. Code checks it against the figure computed from the filed XBRL, within a tolerance.
12,050
Claim checks. Whether a management claim matches the XBRL. The answer is a yes or no, plus an explanation. The explanation is scored by a small decision-model judge, calibrated against known-good and known-bad answers, and a frontier model spot-checks that judge.
9,640
Comparisons across companies. The same metric for more than one issuer. Code checks it against the filed XBRL values, within a tolerance.
7,230
Pass rates
by task type
Illustrative scores for layout review. In this example, the pass rate falls as the question gets broader. Nothing here is a measured eval. Model names are placeholders. Example data.
| Task type | Tasks | Share |
|---|---|---|
| Single-number | 19,280 | 40% |
| Multi-period and segment | 12,050 | 25% |
| Claim check | 9,640 | 20% |
| Cross-company | 7,230 | 15% |
| Model | Single-number | Multi-period | Claim check | Cross-company | Overall |
|---|---|---|---|---|---|
| Frontier model A | 82% | 56% | 48% | 36% | 61.8% |
| Frontier model B | 74% | 48% | 40% | 28% | 53.8% |
| Open model C | 66% | 39% | 32% | 21% | 45.7% |
| Open model D | 48% | 22% | 18% | 10% | 29.8% |
Overall is the example mix applied to the four task types: single-number, multi-period, claim check, then cross-company. Example data.
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 the filed XBRL value, or the figure computed from it. A model is not in that loop.
Then the judge.
A claim check is a yes or no, plus an explanation. The explanation is scored by a small decision-model judge, calibrated against known-good and known-bad answers. A frontier model spot-checks the judge.
Who it is for
AI labs. Eval and RL on financial reasoning, using filings from Q2 2026 or later rather than numbers a model may already have memorized.
RL environment vendors. Tasks already in the Prime Intellect verifiers format, with a code grader for numbers and a calibrated judge for claim-check explanations.
Agents. The catalog, the free sample, and how to purchase are written for automated buyers. Start with For agents, or read the facts below.
Get the pack
Five tasks of each type, from the committed sample, in the same format as the pack.
The full pack. The task count on this page is example data, not a quote.
Book a call for custom or exclusive work
A custom cut, or a window that is not sold to anyone else.
Facts an agent can read
Purchasing steps will be on For agents. Figures marked Example data are placeholders. The same placeholders carry data-example="true". A JSON copy, with figures nested under example, is in the script with id sec-task-pack.
- Product
- Tickmark Bench
- Page status
- Home page. Task counts, pass rates, and model names on this page are example data. The Blackline filing in the hero is a real task from the committed sample.
- Source
- SEC EDGAR bulk XBRL. Not a product of the SEC.
- Filing rule
- Filings from Q2 2026 or later.
- Example task count
- 48,200. Example data.
- Example filing dates
- through . Example data.
- Task types
- Single-number calculations. Multi-period or segment. Claim checks, with a yes or no verdict plus an explanation. Comparisons across companies.
- Numeric grading
- Code against the filed XBRL value, within a tolerance.
- Claim explanations
- A small decision-model judge, calibrated against known-good and known-bad answers. A frontier model spot-checks the judge.
- Format
- Prime Intellect verifiers. Not a Prime Intellect product.
- Free sample
- 20 tasks in the committed sample. The sample page is sample.html.
- Pricing
- pricing.html
- Custom or exclusive
- Book a call
- For agents
- agents.html
Pricing
The full pack, and what it costs.