Dated results, with the tested scope visible

The September 29 and October 17, 2026 checkpoints tested specific citation-checking components and paths. The results below preserve their sample sizes and do not become a general accuracy promise.

Attorney input: records and instructions
Attorney input
Merit prepares material for review
Merit prepares
Attorney reviews facts, law, and dates
Attorney reviews
Attorney completes any manual submission
Attorney acts

Evidence at a glance

Standalone citation check

September 29, 2026. A fabricated citation was blocked, an unsupported real citation was blocked after its meaning was reviewed, and a supported real citation passed. Five related tests passed.

Additional test sets

September 29, 2026. A newly configured environment passed 79 citation-checking tests. The local test set passed 188 tests.

Targeted citation probes

October 17, 2026. The citation review blocked five of five planted citation errors in targeted probes.

What the results support

Three-way check behavior

The standalone check distinguished a fabricated citation, an unsupported real citation, and a supported real citation in the tested path.

Repeatability in the tested environment

The additional and local test sets support the citation-checking behavior observed at the September 29 checkpoint.

Detection in targeted probes

The five planted errors were blocked in the October 17 probes. The sample is reported as a targeted result, not a universal rate.

Scope limits

Component and path specific

These results apply to the specific citation checks, tested paths, and targeted probes described above. They do not establish broad legal accuracy, deadline accuracy, or client outcomes.

Separate from drafting

Citation checking is a separate step. Merit does not claim that every drafting or other workflow automatically runs the citation check.

Attorney verification remains required

The attorney opens and evaluates sources, checks proposition support and application, independently verifies every proposed deadline, and decides what to use.

What each test report includes

Each Merit test report identifies the details readers need to understand the result:

  • Test date
  • Software revision or release
  • Number of test items and runs
  • Inputs and expected outcomes
  • Observed result for the stated test
  • Whether attorney-review time is included
  • Whether the test covers a component or a complete workflow
  • Current disposition only when a dated retest supports it

A timing result appears with its corresponding correctness result. Merit is compared with manual work only when both conditions use a defined, comparable method.

What these tests do and do not show

The public results support only the components and paths named above. No claim is made here about monthly savings, every document type, complete source coverage, automatic citation checking in every workflow, broad legal or deadline accuracy, or client outcomes.

Book a personalized demonstration