Absent
The option cannot answer the requirement or creates a stop-rule risk.
Each guide separates buyer fit from evidence strength. The commercial interest, decision rules, sources, review dates, and outside win conditions stay visible.
Inputs can include workflow, legal risk, workload shape, supervision capacity, ownership need, and time pressure. An input stays only when changing it can change the result.
The option cannot answer the requirement or creates a stop-rule risk.
The claim exists, but no current product, agreement, or source confirms it.
A current product screen, agreement, policy, or primary document supports the claim.
A saved first-hand test shows the requirement held under the stated conditions.
Legal AI guides use a stop rule: a 0 on omission behavior, reproducibility, data handling, or supervision fit stops the recommendation. Operability and exit cost remain cost decisions.
First-hand test or approved outcome record with date, version, and reviewer notes.
Current primary product, contract, security, policy, or vendor source.
Verified user report or credible third-party source with enough context to inspect.
Ambiguous, inferred, unsupported, or outside the review window.
Grade D evidence cannot support a main recommendation. A legal stop rule needs Grade A or B support.
| Fact type | Review window |
|---|---|
| Pricing, packaging, availability | 30 days |
| Feature, integration, workflow | 60 days |
| Data handling, security, legal term | 90 days |
| First-hand test | 90 days or after a major release |
| Verified user report | 180 days |
| Stable category definition | 365 days |
An expired or disputed fact cannot carry the main recommendation. It remains marked until reviewed, corrected, or removed.
A guide fails review if the publisher wins every buyer state. Close scores return a conditional answer. Weak evidence returns an insufficient-evidence result.
Send the exact claim, page URL, and supporting primary source to phil@counterbench.ai. A factual correction and an opinion disagreement are logged separately. Material recommendation changes receive a dated note.
Published August 20, 2026. Method version 1.0.