01
Evaluation
Show what was tested, against which criteria, and with which limitations at release time.
Ormedian Labs · Evidence-led AI assurance
AI systems should come with their evidence.
Our approach connects evaluation, provenance, risk decisions, and monitoring. People can understand the evidence behind an AI system, not just its claims.
Talk to us about this workAssurance framework / 01
An Assurance Pack for each release
The question
What supports this release decision?
A coherent evidence record makes the answer easier to find, inspect, and revisit.
The connected record includes
- 01
Intended use
Purpose, boundaries, and ownership
- 02
Evaluation
Runs, results, thresholds, and limitations
- 03
Risk decisions
Known risks, controls, and reviewers
- 04
Monitoring plan
Signals, actions, and review cadence
Why we’re here
The evidence is often there. The context is missing.
AI work produces test results, documents, reviews, and operational signals. When they live apart from the system version and decision they support, it becomes harder to explain what was known at release time and what changed afterwards.
Ormedian’s approach brings those pieces together as a usable evidence record across an AI system’s lifecycle.
02
Provenance
Link data, models, policies, and review decisions to the release they support.
03
Monitoring
Define the signals, actions, and owners that keep assurance useful after deployment.
Our approach
Assurance should follow the release.
An Assurance Pack is a release-level evidence record. It brings intended use, evaluation, risk decisions, provenance, and a monitoring plan into one place. The record stays meaningful as the system evolves.
Read about Assurance PacksHow the pieces connect
From a release question to an evidence record.
The Ormedian workflow connects definition, evaluation, and ongoing monitoring. Each stage leaves a clear record for the next, instead of another disconnected document.
- 01 / DEFINE
Set the boundaries
State the intended use, owner, known risks, and what good performance would mean.
- 02 / EVALUATE
Test the claim
Record the test conditions, results, limits, and the decision they support.
- 03 / MONITOR & BUNDLE
Keep evidence useful
Connect the release record to signals, owners, and changes after deployment.
Research and writing
The thinking behind the work.
assurance
What is an Assurance Pack?
A practical definition of an Assurance Pack, what goes inside, and how it keeps evidence current as systems change.
Read articlemonitoring
Monitoring by default: the minimum viable signals
The smallest set of production signals that catch real failures early without building a monitoring cathedral.
Read articleprovenance
Provenance: what to capture from day one
A checklist of provenance artifacts that make results reproducible and evidence defensible before you need them.
Read articleStay connected
Help shape what comes next.
Interested in building this with us?
Explore open roles