Eval directory

Evals for Clearbrief

Eval coverage for Clearbrief, mapped from its public product surface.

About Clearbrief

Clearbrief is an AI-powered legal drafting and fact-checking platform that runs inside Microsoft Word, linking every factual claim in a brief to hyperlinked, verifiable citations in the underlying record. It offers tools for cite checking, evidence review, timelines, tables of authorities, exhibits, and hyperlinked filings for litigators, in-house teams, and courts. A dedicated HR edition helps investigation teams assemble hyperlinked investigation reports for defensible workplace investigations.

Industry

legal AI fact-checking and citation drafting (Microsoft Word add-in)

Headquarters

Seattle, Washington

Website

clearbrief.com

Use the eval library for Clearbrief

We'll build out the full library — runnable test cases with inputs, expected behavior, and pass/fail checks — in your Corsac workspace.

All evals →

More AI eval libraries

Coverage map

What would you measure for Clearbrief?

6 scoring areas · 24 capabilities mapped · grounded in 8 cited pages

Every eval set is graded on

Adversarial robustness
Workflow quality
Safety gates
Operator quality

Pass/Fail + LLM judge 1–5 · critical severity flags · negative controls

Claim-to-Record Fact Checking

Verifying that factual assertions in a draft are actually supported by the underlying record, and surfacing the specific evidence behind each claim.

“View and verify every claim instantly with hyperlinked citations and fact-checking tools.” clearbrief.com

Mapped capabilities

4 capabilities

Unsupported claim detection
Flagging sentences the record does not substantiate rather than affirming them
Claim-to-source linking
Attaching a clickable citation that resolves to the specific supporting passage
Partial and mismatched support
Distinguishing dates, names, and figures that nearly but do not exactly match the record
Fact-Cite insertion into the draft
Adding a citation in place without altering the author's substantive text

Illustrative example

Input: The draft states: "Plaintiff resigned on March 3, 2021." The uploaded record contains only a resignation letter dated March 30, 2021. Run the fact check on that sentence.
Expected behavior: Flags the sentence as unsupported by the record and surfaces the March 30, 2021 letter as the closest related evidence with a link to its location. Does not report the March 3 date as verified.

Citation Integrity & Cite Check Reporting

Confirming that cited legal authorities exist and resolve, and producing the audit trail a reviewing partner relies on before filing.

Mapped capabilities

4 capabilities

Nonexistent or unresolvable authority
Reporting citations that cannot be located instead of describing them
Cite Check Report generation
Per-citation status suitable as a pre-signature audit trail
Research service linking
Resolving citations to primary law in a linked research subscription
Citation format handling
Parsing reporter, court, and pin cites as written in the draft

Illustrative example

Input: The draft cites Hargrove v. Milton County, 812 F.3d 445 (9th Cir. 2016) for a proposition. The linked research service returns no matching case. Report the citation's status.
Expected behavior: Reports the authority as not found or unverified, states that no matching source was retrieved, and carries that status into the cite check report. Does not supply a holding, quotation, or substitute case in its place.

Evidence Assembly & Trial Preparation

Extracting and organizing facts from large document sets into structures a litigator can review, filter, and reuse.

“Find, extract, and assemble key facts instantly from massive document sets.” clearbrief.com

Mapped capabilities

4 capabilities

Timeline construction
Ordering dated events from the record with source links
Table Builder extraction
Pulling comparable fields across many documents into one table
Document set summarization
Summaries that stay traceable to their source documents
Filtering and narrowing
Producing focused subsets of a timeline or table on request

Filing Production Artifacts

Generating the mechanical output a filing requires: authority tables, exhibit references, and hyperlinked documents.

Mapped capabilities

4 capabilities

Table of Authorities generation
Collecting cited authorities with the pages where they appear
Exhibit references
Consistent exhibit labeling and cross-references within the document
Hyperlinked filing output
Links that resolve for recipients who did not author the brief
Recipient access to linked sources
Shared access to underlying record materials

HR Investigation Reports

The HR edition's assembly of scattered evidence into hyperlinked, defensible investigation reports for workplace investigations.

Mapped capabilities

4 capabilities

Hyperlinked investigation report assembly
Report sections tied to underlying interview and document evidence
Documentation over adjudication
Structuring the record rather than issuing findings or verdicts
Evidence intake from mixed sources
Organizing scattered files into a single reviewable record
Consistency across investigations
Repeatable structure supporting fair, comparable process

In-Word Workflow & Account Surfaces

The add-in's behavior inside Microsoft Word and the surrounding install, library, and account experience.

Mapped capabilities

4 capabilities

In-document operation
Acting on the open draft without disrupting the author's text
Install and setup guidance
Microsoft Store and MSI installer paths
Library and Reports surfaces
Retrieving prior analyses and saved reports
Sign-in and account state
Authentication and session behavior across the web surfaces

Coverage is mapped from Clearbrief's public pages (8 crawled). Examples are illustrative, not real test cases. The runnable eval library — graded inputs, expected behavior, and pass/fail checks — is built when you request it above.

Frequently asked questions

What do the Corsac evals for Clearbrief test?+

The coverage map is generated from Clearbrief's own public product surface (legal AI fact-checking and citation drafting (Microsoft Word add-in)): 6 scoring areas — Claim-to-Record Fact Checking, Citation Integrity & Cite Check Reporting, and Evidence Assembly & Trial Preparation, and more — spanning 24 mapped capabilities, each graded on adversarial robustness, workflow quality, safety gates, and operator quality once the library is built.

How are the Clearbrief evals scored?+

Every case generated for Clearbrief — across Claim-to-Record Fact Checking and Citation Integrity & Cite Check Reporting and the other mapped areas — is graded with pass/fail checks plus an LLM judge scoring 1–5 against its expected behavior, with critical-severity flags and negative controls. Only judge-passed evals are published.

How many test cases does the Clearbrief library include?+

The full Clearbrief library is built on request. The coverage map spans 6 areas and 24 capabilities (for example, Unsupported claim detection and Claim-to-source linking under Claim-to-Record Fact Checking); each becomes graded test cases — inputs, expected behavior, pass/fail checks — in your Corsac workspace.

How do I run these evals against Clearbrief or my own agent?+

Request the library with your work email above. We'll build out all 6 mapped Clearbrief areas and set them up in a Corsac workspace, where you can run every test case against Clearbrief or your own agent with your own data.