---
title: Verify the Numbers in AI Monitoring Reports
description: Check the subject, period and source behind a plausible explanation.
url: "https://bestnetllc.co.jp/en/docs/the-story-of-local-llm-fabricating-monitoring-report-numbers-and-countermeasures/"
updated: "2026-11-13"
category: Tech Blog / AI Security Ops
---

# Verify the Numbers in AI Monitoring Reports

2026.04 / Tech Blog / BASTION

Updated 2026-11-13

# Verify the Numbers in AI Monitoring Reports

Check the subject, period and source behind a plausible explanation.

The original case found inconsistencies between an AI monitoring report, its input and observation records. This article focuses on how to verify the output.

## Check the scope of the input

Numbers with the same label can mean different things when the subject or reporting period differs. Verify the input against the intended conditions before using it.

## Trace the output back to its source

Fluent prose does not ensure that numbers or explanations match the evidence. Make it possible to return to the input and observation records when a discrepancy appears.

## State what was actually checked

Passing a limited comparison does not prove that every explanation is correct. Check numerical consistency separately from the validity of a causal explanation.

The lessons from the original case have been reorganized for this public article. See the BASTION introduction for current features and deployment conditions.

[About BASTION](https://bestnetllc.co.jp/en/bastion-6/)

[Contact us](https://bestnetllc.co.jp/en/contact-6/)