Version and provenance
Associate an output with the model version and the relevant recorded events. The goal is to explain which system produced an advisory and what changed around it.
RAIN / Resources
A model decisionshould leave a trail.
Understand RAIN’s recorded model history, the evidence an evaluation needs and the limits of the current prototype.
Audit framework · no independent report published
INTELLIGENCE BELONGS WHERE THE DATA LIVES.
01 / THE IDEA
Follow the evidence behind a change.
RAIN’s simulator includes a signed, linked history of inferences and model changes, including rejected updates. This page explains that evidence model. It is not a published third-party LLM audit, and it does not establish that a particular language model is suitable for an operational task.
Evaluation and promotion
Retain the candidate comparison, evaluation conditions and promotion outcome. Failed candidates matter because they explain why the incumbent remained active.
Limits of the record
A signature does not prove that an input was accurate or an advisory was useful. Hardware-backed signing, field reliability and any independent assurance require further work.
02 / IN PRACTICE
Put the thinking into practice.
Define the scope, preserve the boundary, and make the outcome something a person can inspect.
- 01
Define the audit question
Decide whether you are tracing an advisory, reviewing a model change or assessing the reliability of the wider system.
- 02
Inspect the supporting evidence
Compare model identity, event ordering and evaluation results. Check what information is available and make gaps explicit.
- 03
Separate findings from assumptions
Record what the evidence supports. Treat operational usefulness, source accuracy and suitability for a new task as separate questions.
03 / A FEW DETAILS
Worth understanding.
Where can I download the LLM audit report?
No independent LLM audit report has been published on this website. This page describes the current audit-history approach and its limitations.
Does an audit trail prove a model is correct?
No. It helps trace outputs and changes. Model accuracy, data quality and operational usefulness need their own evaluation and field evidence.
BUILD WITH RAIN
One site. One question.
A useful next step.
RAIN is preparing for its first field pilot. Start by defining the environment, the data boundary, and what success should look like.