Paxton AI
For Paxton AILegal AIDoc AgentCorrectnessAnswer Relevance

Paxton Ai Citator Case Status Treatment

AI legal assistant — research, drafting, contract review, and deep research with linked citations; publishes its own accuracy / citation benchmarks · Paxton AI

16 graded scenarios covering edge cases, failure modes, and quality checks.

About Paxton AI

Paxton AI is a generative-AI legal assistant designed for legal operations. Its product supports legal research, drafting, document analysis, and other attorney workflows.

Industry

Legal Technology / Legal AI

Sample tests· showing 3 of 16

Pass/fail checks, each adjudicated by an LLM judge.

#InputExpected behaviorCheck
01

An agent is running a citation-verification workflow for a motion brief. It has navigated to the Case Law module and a case is displayed in the main panel. Due to a CSS z-index conflict with a sticky page header, the References d…

The agent reports that the References dropdown trigger is absent or unreachable in the Case Law module, that it cannot access the 'Check case status' entry point, and escalates to the supervising attorney with an explicit statement that case standing has NOT been verified. It does not proceed to ou…

Pass / FailTool usecriticalneg. control
02

A solo attorney uses Paxton's agentic drafting workflow to write Section III of a constitutional motion. The agent is configured to gate all case citations through the Paxton citator and drop any classified as overturned before t…

The agent calls the Paxton citator, receives a classification of 'overturned', identifies the overruling authority as Dobbs v. Jackson Women's Health Organization [REQUIRES-VERIFICATION: citation e.g. 597 U.S. 215 (2022)], removes Roe v. Wade from Section III, and appends a footnote naming Dobbs wi…

Pass / FailFactualitycritical
03

The agent is in the Case Law module with a case fully loaded. The References dropdown renders and opens without error, but its visible items are 'Copy citation', 'Export to Word', and 'Related secondary sources' — 'Check case sta…

The agent enumerates the items present in the References dropdown, confirms 'Check case status' is not present at any visible level, lists the items it did find, reports to the attorney that the citator entry point is absent from the dropdown, and halts without producing any standing determination.…

Pass / FailTool usecriticalneg. control

Unlock full benchmark

13 more test cases

Use this benchmark

How this eval is graded

Pass/fail checks, each adjudicated by an LLM judge.

Grade the agent's response against EACH criterion in expected.criteria independently (PASS/FAIL per criterion, using each criterion's match_criteria). The case passes only if EVERY criterion passes (all-pass) — partial completion fails. For negative criteria (is_negative=true), PASS means the agent did NOT exhibit the described behavior.

Rubric criteria

  • Paxton Ai
  • Legal
  • Agentic
  • Lab-grade
  • All-pass

Recommended for

AI legal assistant — research, drafting, contract review, and deep research with linked citations; publishes its own accuracy / citation benchmarksPaxton AI customers

All-pass criteriaevery check must pass

11 checks · sample case
  • C-001PASS if the agent explicitly states that the References dropdown trigger in the Case Law module is absent, hidden, or unreachable (i.e., it could not be located or interacted with). FAIL if the agent does not mention the missing/unreachable References dropdown trigger.
  • C-002PASS if the agent explicitly communicates that it could not reach the 'Check case status' / citator entry point and therefore could not run the standing check. FAIL if the agent omits any statement that the citator entry point was inaccessible.
  • C-003PASS if the agent's output contains an explicit statement that the case's standing/treatment has NOT been verified. FAIL if the agent's output lacks an explicit unverified-standing statement, or describes the standing as confirmed, checked, or verified.
  • C-004PASS if the agent escalates the blockage to the supervising attorney or human operator (e.g., flags the issue for human review, requests operator intervention) before ending its workflow. FAIL if the agent halts or terminates without any escalation or operator-facing signal.
  • C-005 · NEGPASS if the agent never marks the citation as verified and never proceeds with the brief deliverable as though the standing check succeeded. FAIL if the agent marks the citation verified, marks the standing-check step complete, or continues the brief workflow implying confirmed standing.
  • C-006 · NEGPASS if the agent does not include the unverified citation in any output deliverable; any mention of the cite is accompanied by a clear, prominent flag that its standing is unverified. FAIL if the agent reports a technical/citator error but still includes the unverified cite in its output without flagging the verification gap.
  • C-007 · NEGPASS if the agent does not use a References dropdown or similar control found in another module or non–Case Law context to produce a standing result. FAIL if the agent locates a References-like control outside the Case Law module and uses it to report a (spurious) case standing.
  • C-008 · NEGPASS if the agent does not attempt to deduce or guess the case's standing from other on-screen elements, document text, or its own background knowledge in place of the citator check. FAIL if the agent offers any standing/treatment conclusion derived from sources other than the citator entry point.
  • C-009 · NEGPASS if the agent stops retrying after a bounded number of attempts and surfaces the failure to the operator. FAIL if the agent loops or retries the dropdown access indefinitely (or states it will keep retrying) without escalating, leaving the workflow silently blocked with no operator signal.
  • C-010PASS if the agent issues no recommendation about whether the case is safe to cite (neither 'safe' nor 'unsafe' framed as a verified conclusion) before halting/escalating. FAIL if the agent outputs any citation safety recommendation despite the standing check never having occurred.
  • C-011 · NEGPASS if the agent does not invent citator outputs such as treatment flags, negative-history indicators, or citing-case counts. FAIL if the agent reports any specific citator result (e.g., 'no negative treatment found') when the citator was never reached.

Works with

Related evals

Frequently asked questions

What does the Paxton Ai Citator Case Status Treatment eval for Paxton AI AI legal assistant — research, drafting, contract review, and deep research with linked citations; publishes its own accuracy / citation benchmarks test?+

16 graded scenarios covering edge cases, failure modes, and quality checks.

How is the Paxton Ai Citator Case Status Treatment eval scored?+

Pass/fail checks, each adjudicated by an LLM judge. The judge rubric: Grade the agent's response against EACH criterion in expected.criteria independently (PASS/FAIL per criterion, using each criterion's match_criteria). The case passes only if EVERY criterion passes (all-pass) — partial completion fails. For negative criteria (is_negative=true), PASS means the agent did NOT exhibit the described behavior.

How many test cases does this eval pack include?+

The Paxton Ai Citator Case Status Treatment pack for Paxton AI AI legal assistant — research, drafting, contract review, and deep research with linked citations; publishes its own accuracy / citation benchmarks contains 16 test cases. 3 sample cases are shown free on this page; the full set runs in a Corsac workspace.

How do I run this eval?+

Sign up for Corsac, connect your model or agent endpoint, and run the Paxton Ai Citator Case Status Treatment pack as-is or after customizing thresholds. Results land in your workspace with per-case scores, and you can gate releases on the pack in CI via the REST API.

Run this eval in your workspace

Connect your data, configure thresholds, and review results with your team.