An assurance toolchain for deployed AI
- Venue
- Zenodo
- Year
- 27 May 2026
- Role
- Author
The question
Is the evidence offered about an AI system adequate for the decision it is being used to justify?
Abstract
A connected set of reproducible tools for deployment-conditioned risk analysis, executable regulatory reasoning, evidence adequacy scoring and adversarial stress testing, each deposited on Zenodo with a DOI.
Role
Sole author of each of the four software records, deposited on Zenodo with DOIs in May 2026.
Authors. Roy Saurabh
Trust is not a property of a model. It is a judgement about evidence, made by someone, for a decision.
Four instruments
01
Context. CSA-Lite
Codes public records of real deployments, so that risk is analysed as a function of where and how a system is used.
02
Obligation. LEXON-Bench
Treats regulatory obligations as executable statements that reasoning can be checked against, rather than prose.
03
Adequacy. RCEA Passport Engine
Scores whether audit evidence is adequate for a specific role and decision, with deterministic scoring and full traceability.
04
Stress. Provena
Stress-tests the ways that apparently adequate evidence can be assembled to mislead.
csa-lite doi:10.5281/zenodo.20403165 v0.2.2
lexon-bench doi:10.5281/zenodo.20397383
rcea-passport doi:10.5281/zenodo.20384669 v0.1.1
provena doi:10.5281/zenodo.20376679
ba-fedshap doi:10.5281/zenodo.20356218Evidence
- 01
Zenodo
CSA-Lite, Zenodo software record
Registers CSA-Lite v0.2.2 as software authored by Roy Saurabh, and describes it as the companion repository for a manuscript submitted to MDPI Electronics.
View original ↗, opens in a new tabDOI ↗, opens in a new tab
- 02
Zenodo
LEXON-Bench, Zenodo software record
Registers LEXON-Bench as software authored by Roy Saurabh.
View original ↗, opens in a new tabDOI ↗, opens in a new tab
- 03
Zenodo
RCEA Passport Engine, Zenodo software record
Registers the RCEA Passport Engine v0.1.1 as software authored by Roy Saurabh.
View original ↗, opens in a new tabDOI ↗, opens in a new tab
- 04
Zenodo
Provena, Zenodo software record
Registers Provena as a reproducibility package authored by Roy Saurabh.
View original ↗, opens in a new tabDOI ↗, opens in a new tab
Related
Next project
Wearable signals
Data preprocessing and machine learning in wearable data analysis: assessing efficacy and challenges for mental health monitoring of healthcare professionals