Controlled positive case
12/12 qualification folds passed, with zero observed false skips in the synthetic positive control. The AURIC controlled drift check recorded a veto rate of 1.0.
VAREXIS Technologies
Structural-change research for data streams.
Evaluate which changes in your data deserve a closer look. A scoped study compares AURIC with existing methods on your workload, measuring useful detections, missed changes and review effort.
Research software · Partner evaluation by agreement

The opportunity
AURIC investigates structural change in data streams. Interpreting its cause or meaning remains a separate task. Its default-silence approach is designed to withhold a signal unless its conditions are met.
The practical question is whether those signals help your team find relevant changes with a manageable review burden. A high silence rate only helps when important changes are still detected.
Candidate applications
These are settings for evaluation. Benefits must be demonstrated on the partner workload.
Development evidence from Isambard-AI
In M1, job 6070189, AURIC served as a veto-only monitor within the BDAS workload-qualification protocol. Its role was to withhold permission for selective processing when monitored conditions failed. It had no authority to approve omitted work.
12/12 qualification folds passed, with zero observed false skips in the synthetic positive control. The AURIC controlled drift check recorded a veto rate of 1.0.
The consumed hydraulic workload qualified only 3/12 folds and recorded eight candidate false skips. The qualification protocol recorded zero unsafe promotions; the workload was not promoted.
This established a concrete monitor role and a way to test when selective processing should be refused. It supports further evaluation of AURIC alongside an existing pipeline.
M1 was a development qualification test. The veto rate describes its controlled drift check; it is not 100% real-world detection accuracy. The wider protocol outcomes do not isolate AURIC’s contribution from the other qualification checks.
Heavy-kernel processing was excluded. M1 establishes no economic speedup, fresh client validation or production readiness. The later BDAS E4B runtime result is separate and does not establish an AURIC speedup contribution.
Research record
The retained record includes reproducible reference behaviour, development integration and mixed benchmark outcomes. It supports a scoped feasibility study; customer benefit remains to be established.
The archived minimal v1.2 implementation reproduced its expected output on a constructed stream. This verifies that reference behaviour, without establishing field accuracy or execution of the full proposed architecture.
The selected NAB benchmark did not meet its declared signal criteria. A later audit identified scoring limitations in the historical positive TCPD result.
The CK2 audit found that the historical change-point scoring could reward an entirely silent detector above AURIC, and that the score was sensitive to annotation order. The original result remains in the research record with this qualification. A partner study needs agreed event definitions and appropriate comparison methods.
CK2 completed an analysis of previously used detector outputs. It did not establish fresh validation, calibrated cross-series confidence or production promotion.
First-party research records. Neither use of external datasets nor Isambard-AI infrastructure establishes independent validation or product endorsement.
Scoped feasibility study
Agree the stream, relevant changes, review process and success criteria.
Check that the agreed implementation can be reproduced and that the data are suitable.
Measure useful detections, missed events, false alerts, detection delay and runtime against existing methods.
Report review effort, operating cost and limitations before deciding on further development or integration.
Evaluation and access
A feasibility study or co-development project with an agreed workload, comparison plan and deliverables. Scope, reproducibility, data access and fees are agreed before work begins. Any subsequent evaluation licence or integration agreement is separate.
An adapter or API can be scoped for evaluation once the implementation and data requirements are agreed. This page does not offer a deployed public AURIC execution service.
The public page describes purpose, evidence and evaluation scope. Implementation details and supporting material are considered separately under agreed confidentiality and permitted use. An enquiry does not grant source code or reproduction rights.
Feasibility · Co-development · Evaluation
Start with your data, your existing method and the cost of getting the decision wrong.
Evidence summary updated 14 October 2026. Research evaluation; no guaranteed detection benefit or production promotion.

VAREXIS Ltd is an NVIDIA Inception program member.
NVIDIA, the NVIDIA logo and NVIDIA Inception are trademarks and/or registered trademarks of NVIDIA Corporation in the U.S. and other countries.

Compute access awarded through the UKRI AI Research Resource for BDAS evaluation on Isambard-AI.