A second opinion on your AI-assisted decisions.
Two independently retained methods examine the same decision — one structural, one behavioral — without coordinating their conclusions. You get both findings, compared honestly, not blended into a single score.
“The model was accurate” is not the same question as “was this decision reasonable.”
When an AI-assisted decision is challenged — internally, by a regulator, or in a dispute — many reviews focus first on whether the outcome was correct or the model performed as expected. That question matters, but it leaves two harder questions: how did this situation become a decision in this particular form, and was the person's engagement with the system reasonable given exactly what they were shown at the time? AI Decision Integrity Review (ADIR) is designed to examine both independently, on one specific, bounded decision.
Two Independent Methods
-
ID∆AC™ — Decision Route Exposure
Reconstructs how the decision route itself was formed: what entered it, how AI shaped the available field, where practical authority sat, whether intervention remained executable, and what evidence survives for later reconstruction or challenge.
-
BOLD — Reasonableness Review
Reconstructs whether the person's reliance on, deviation from, or interaction with the AI output was reasonable given exactly what they were shown — independent of whether the outcome was right.
What You Receive
-
Independent Structural Finding
Deliverable 1
ID∆AC™'s native account of the decision route, authority, and evidentiary record — delivered on its own, attributable to ID∆AC™ alone.
-
Independent Behavioral Finding
Deliverable 2
Behavieural's native reasonableness verdict on the human side of the decision — delivered on its own, attributable to Behavieural alone.
-
Bilateral Comparison Note
Deliverable 3
Where the two findings converge, where they differ, where the evidence is asymmetric, and what remains unresolved — with no merged score, verdict, or certification.
-
Who It's For
Designed primarily for Legal, Risk, Compliance, Internal Audit, and Assurance functions that need an honest, defensible record of one specific AI-assisted decision — including one that isn't fully flattering. Typical moments: a decision already under internal review, a pending dispute, or a regulatory or supervisory inquiry.
What This is Not
-
Not a technical or model accuracy audit.
-
Not legal advice or statutory compliance certification.
-
Not forensic accounting or a governance redesign.
-
Not expert testimony by default — dispute and expert-witness use require separate scoping.
-
Not one blended methodology — each firm is separately retained and independently responsible for its own finding.
How It Works
-
Scope the Decision
Step 1
Together, we identify one bounded, already-identified AI-assisted decision route — not a full deployment or program.
-
Independent Review
Step 2
ID∆AC™ and Behavieural each run their own review on their own timeline, in parallel, over a matter of business days.
-
Bilateral Comparison
Step 3
The two native findings are frozen, then compared in a single note covering convergence, difference, and what remains open.
-
Credibility
Two independent practices. Published, dated, citable research.
The ADIR bilateral interface has been tested on controlled synthetic cases ahead of its first live pilot. Both principals publish their thinking openly rather than only behind a sales deck — ID∆AC™'s research essays on decision formation, authority, and evidentiary reconstruction are publicly available with DOIs, and Behavieural's methodology is grounded in the AI Trust Axis, a peer-reviewed behavioral framework used across its engagement portfolio.
Who We Are
-
Cosmin Oprea
Founder, ID∆AC™
Independent researcher and method developer focused on the structural conditions of AI-assisted decisions: how a decision is formed, how AI shapes the route before formal review, where practical authority sits, when human intervention can still change the trajectory, and what evidence survives for later reconstruction and challenge. Founder of ID∆AC™, an applied research and diagnostic practice working on decision exposure, human agency, authority, evidence provenance and reconstructibility.
-
Eugene Chan (PhD)
Principal Strategist, Behavieural
Eugene is the Founder and Principal Strategist of Behavieural, a consultancy that uses behavioral science to solve organizations' hardest trust problems — AI adoption, customer loyalty, brand and reputation. He is Full Professor and Van Norman Chair of Business at Tyndale University, and has previously taught at Toronto Metropolitan University as well as in Australia and the United States. His PhD is from the Rotman School of Management, and he is the author of Managing Brand Crises: A Guide for Navigating the Storm.