Skip to content

Assurance OS for model releases

Ship AI systems you can defend.

Ormedian turns high-stakes AI releases into living evidence trails: release contracts, evaluation gates, monitoring signals, provenance, and Assurance Packs.

View the open-source scaffold on GitHub →
  1. 01

    Release

    Version, ownerintended use

  2. 02

    Gates

    Evaluation, riskpolicy

  3. 03

    Operate

    Monitoringincidents, controls

  4. 04

    Assure

    Living packapproval trail

A model release moves through evidence gates and monitoring into a living Assurance Pack.

Release evidence trail

01

Release contract

intent locked

02

Evaluation gate

run_7f91 passed

03

Risk control

0 critical open

04

Monitoring

signals live

Assurance Pack

claims-routing-v4.1

Current

Intended use Complete

Evaluation Passed

Risk register Controlled

Monitoring Live

The evidence gap

Defensibility fails when evidence loses its release context.

Evidence usually exists

Metrics Risk notes Approvals

Metrics, risk notes, logs, approvals, and incident context usually exist. They just do not travel together as one release record. Ormedian makes the release itself the organizing unit.

01

Customer diligence

Which release did this evidence support?

Show the exact model, data, eval run, approval, and monitoring plan behind a customer-facing claim.

02

Incident response

What changed, and who accepted the risk?

Trace the release diff, risk owner, and control that was in force when live behavior shifted.

03

Audit readiness

Can you reproduce the decision trail?

Turn scattered notebooks, dashboards, documents, and chat threads into one defensible artifact.

Assurance Packs

One living dossier per AI release.

An Assurance Pack is the destination of the release evidence trail: compact enough to inspect, complete enough to defend.

Assurance Pack

claims-routing-v4.1

Review-ready

decision: approved-with-control - evidence updated 12m ago

Intended use

Route enterprise support claims to the right specialist queue with human review on uncertain or policy-sensitive cases.

v4.1
Release
claims-routing-v4.1
Decision
Approved with controls
Owner
Applied AI Platform
Freshness
Monitoring active

Release gates

Routing accuracy delta -0.7%
Policy override rate 0.04%
Open critical risks 0

Control note

Manual review required for claims above the financial-risk threshold.

Intended use, scope, assumptions, exclusions, and policy constraints

Evaluation results with run IDs, baselines, thresholds, and failure notes

Risk register entries with owners, controls, residual risk, and review cadence

Monitoring thresholds, alert rules, incidents, and escalation paths

Provenance linking data, model, policy, reviewer, and release decision

Approval history that stays coupled to the deployed version

Workflow

Governance becomes part of the release path.

Ormedian turns review gates into concrete outputs, so the pack is assembled while work happens instead of after everyone has moved on.

  1. 01

    Define release contract

    Capture intended use, exclusions, success metrics, evaluation criteria, and policy constraints before review.

  2. 02

    Evaluate against gates

    Attach run IDs, baselines, thresholds, failures, and deltas from the evaluation suite.

  3. 03

    Monitor deployed behavior

    Connect production signals, drift thresholds, incident notes, policy violations, and alert ownership.

  4. 04

    Bundle current evidence

    Publish the Assurance Pack as the current record for the deployed version and release decision.

Release review

claims-routing-v4.1

Evidence gates
  1. 01

    Define intended use

    intended-use.yaml

    Locked
  2. 02

    Run evaluation suite

    eval-run-7f91.json

    Passed
  3. 03

    Review residual risk

    risk-register.md

    Control added
  4. 04

    Attach monitoring plan

    signals.yaml

    Active
  5. 05

    Bundle release evidence

    assurance-pack-v4.1

    Current

After deployment

Evidence keeps moving after launch.

Monitoring signals, provenance edges, incidents, and approvals remain coupled to the deployed version instead of drifting into separate operational tools.

Live monitoring

Release health cockpit

Streaming

Drift

0.18

stable

Regression

-2.1%

within gate

Policy events

4

reviewing

Release health

97%

healthy

Signal trail

Alert threshold

Evidence updates

Policy spike

Financial advice classifier exceeded review threshold.

Dataset shift

New market segment observed in production traffic.

Pack refreshed

Monitoring evidence linked to the current release record.

Provenance graph connecting Eval Run, Dataset, Model, Policy, Reviewer, and Decision with a fixed release evidence topology.

Provenance

Release evidence graph

Version-coupled
EVAL RUNrun79f1 Dataset claims_eval06_5 Model model:v4.1.18 Policy risk-policy:12 👤 Reviewer risk + platform approved Decision

Review moments

Built for the moments where claims need evidence.

Regulated launch

Package intended use, evaluation coverage, residual risks, monitoring plans, and approvals before production.

Customer evidence request

Respond with a release-specific assurance record instead of scattered screenshots and ad hoc documents.

Drift or incident

Trace what changed, which thresholds fired, who reviewed it, and which control applied.

Model release review

Give engineering, product, risk, and compliance the same evidence record before sign-off.

Procurement review

Show how the AI system is evaluated, monitored, governed, and kept under control.

Policy or ownership change

Keep controls, owners, assumptions, exclusions, and review cadence explicit.

Evidence control

Evidence tied to deployed versions

Evidence control

Reproducible evaluation references

Evidence control

Risk controls with named owners

Evidence control

Monitoring plans that stay current

Early access

Start with one release you need to defend.

Join the waitlist for a sample Assurance Pack, beta access, and practical templates for evaluation, monitoring, provenance, and release governance.