AuraOne
Start a project
CUSTOMERS · CASE STUDIES

Stories you can run, not logos you can’t verify.

Real programs, anonymized by field. Each one shows the work that came in, the record it left behind, and the gate it set for the next release. Bring yours and run the same loop.

Start a projectBack to customers
BY FIELD
Six programs

Post-training, model risk, manufacturing, clinical, robotics, compliance. Jump to yours.

EACH STORY
Work in. Record out.

The work that came in, the record it left, and the gate it set for the next release.

ANONYMIZED
By field

Attributed by company type, not by name. A private reference path opens at procurement.

ON THE RECORD · ANONYMIZED BY FIELD

The pain that started it, the record it left behind.

Six programs, attributed by company type rather than by name. Each one is a pattern you can run today. When evaluation moves to procurement, a private reference path opens.

POST-TRAINING · FRONTIER LAB
↳ EVIDENCE PACKET
A frontier AI lab
An escaped regression had nowhere to go.

RLHF that compounds, instead of resetting.

Specialist review, regression capture, and rubric versioning stayed attached to the same packet across releases. When a model slipped, the failure became a gate the next checkpoint had to pass. Override chains turned inspectable, and the lab kept the weights.

One packet
review, rubric, and override in one record
Signed
who reviewed each case, under what rights
Held
regression returns as a release gate
Record trail
POST-TRAINING · EVIDENCE PACKET
A frontier AI lab. Who made it, who reviewed it, under what rights.
MODEL RISK · UNDER EXAMINATION
↳ GOVERNANCE PACKET
A regulated decisioning program
A score does not survive examination.

A governance packet does.

Borderline credit, fraud, and AML decisions moved from a screenshot debate to one inspectable record per run. Risk, engineering, and change management worked the same approval surface, with escalation and regression memory attached. Examiner questions earned one answer per audit cycle.

One record
per release, examiner-ready
Escalated
borderline calls routed to a named reviewer
Remembered
the same miss is caught next cycle
Record trail
MODEL RISK · GOVERNANCE PACKET
A regulated decisioning program. Who made it, who reviewed it, under what rights.
MANUFACTURING · QUALITY RELEASE
↳ RELEASE PACKET
A manufacturing quality program
Detection is not release.

Before the part ships.

Visual defect detection flagged the suspect part. The release happened when a qualified reviewer signed it, with the image, the rubric, and the decision in one record. Every escaped defect became a regression gate, so the same miss could not pass the line twice.

One record
image, reviewer, and decision together
Signed
the reviewer who cleared the part
Held
escaped defects return as gates
Record trail
MANUFACTURING · RELEASE PACKET
A manufacturing quality program. Who made it, who reviewed it, under what rights.
CLINICAL · SECOND READ
↳ REVIEW RECORD
An academic medical center
A triage model is not a reviewed record.

Second reads went to the right reader.

Uncertain inferences routed to the radiologist qualified for the body region, with the rubric and override chain attached to the case. The record proved who reviewed the hard read and why. Calibration drift surfaced from real escalation data, not from anecdote.

Routed
to the reader qualified for the region
Signed
who read the hard case, and why
Measured
drift from real escalations, not anecdote
Record trail
CLINICAL · REVIEW RECORD
An academic medical center. Who made it, who reviewed it, under what rights.
ROBOTICS · MANIPULATION DATA
↳ REPLAY SUITE
A humanoid program
Failure data lived in the gig-platform mess.

Capture, review, export — one chain.

Demonstration capture, teleop review, and export lineage stayed in the same workflow that shipped the policy. Task-diverse failures came in rights-cleared and safety-reviewed, not from a teleop free-for-all. Replay suites turned escaped behaviors into gates the next checkpoint had to pass.

100–500
reviewed episodes in a pilot band
Rights-cleared
consent attached to each capture
Held
escaped behaviors return as gates
Record trail
ROBOTICS · REPLAY SUITE
A humanoid program. Who made it, who reviewed it, under what rights.
COMPLIANCE · TRAINING-DATA PROVENANCE
↳ PROVENANCE TRAIL
An enterprise procurement program
Most teams cannot trace where training data came from.

Provenance the auditor can read.

Every datapoint carried who created it, who reviewed it, and under what rights. When the EU AI Act provenance clock arrived in August 2026, the record was already there: a neutral second source the buyer could defend under audit, with the controls and logs attached to the workflow operators used day-to-day.

78% / 77%
of orgs cannot validate or trace training data
Aug 2026
EU AI Act provenance clock
Second source
neutral, independent, audit-defensible
Record trail
COMPLIANCE · PROVENANCE TRAIL
An enterprise procurement program. Who made it, who reviewed it, under what rights.
Reference-style example · attributed by field, not by name

“We stopped asking who approved this and reading silence. Every decision carries who made it, who reviewed it, and under what rights — so the record survives the audit.”

A model-risk officer · a regulated decisioning program
BRING THE WORK

Bring the work. Keep the proof. Own the model.

Bring the release, the review queue, or the escaped issue that matters most. We will show how it runs, what it leaves behind, and how the weights stay yours.

Start a projectBack to customers
Case Studies | Program examples for AuraOne | AuraOne