Proprietary framework · definition
Four reference frameworks each describe part of what responsible AI must be. The iDIA framework converges them into a single measurable profile your board can read at a glance.
In brief
The iDIA framework is identifiable's evaluation method. It scores an artificial intelligence system on evidence, against ten weighted properties: Sovereign 6%, Mapped 8%, Evaluated 8%, Governed 16%, Transparent 8%, Measurable 10%, Accountable 10%, Managed 10%, Applied 14% and Mature 10%. The weighting is derived from ISO/IEC 42001, the NIST AI RMF, the AI Act and Law 25. A practice that is claimed but not written down hits a ceiling in the score.
Each property is scored from 0 to 100 on evidence. Together, they cover what a board, a regulator, or a client can ask for, without overlap and without gaps.
| Property | Weight | What it measures |
|---|---|---|
| Sovereign | 6 % | Where models and data live, under which jurisdiction, and what happens if the vendor changes its terms. |
| Mapped | 8 % | The inventory of systems, uses and instances. What is not inventoried can be neither evaluated nor governed. |
| Evaluated | 8 % | The data, the maturity of the organization and its skills, risks and their classification, outputs, robustness, use, bias. |
| Governed | 16 % | AI policy, roles, registers and lifecycle, defined and upheld over time. The heaviest property in the framework. |
| Transparent | 8 % | AI use disclosed, how results are validated and decisions are made, explainable to the person they affect. |
| Measurable | 10 % | Indicators tracked, plus a dimension the standards barely address: the costs of the stack and of each AI in use. |
| Accountable | 10 % | Who answers for each assisted decision, on what basis, and what remains on record. |
| Managed | 10 % | A competent human can intervene, correct and stop. Incidents, bias and changes follow a defined, owned process. |
| Applied | 14 % | Skills development, usage instances, real responsibility. Practice catches up with paper, or it does not. |
| Mature | 10 % | The seal of the Responsible AI Practice: the whole holds over time, on an established cadence of improvement. |
Current weighting, derived from the emphasis matrix of the four reference frameworks. It recalculates if the emphasis of any framework changes.
The weights do not come from preference. Each reference framework rates each property according to its emphasis (high, medium, low), and the weights normalize to 100. Governed carries 16 because all four frameworks converge on it, out of technical necessity.
The consequence matters: change one emphasis rating and the weights recalculate. That is what allows a weight to be justified before an audit committee rather than simply asserted.
On evidence, never on declaration. Five maturity levels, and each level demands more than the one before.
| Level | Score | What proves it |
|---|---|---|
| Absent | 0 | No practice in place. |
| Declared | 25 | The intent exists, without written record. |
| Documented | 50 | The practice is written and accessible. |
| Operational | 75 | The practice is applied and observable day to day. |
| Measured and improved | 100 | The practice is tracked by indicators and reviewed. |
The AI Index is the weighted sum of the six scores. Four bands place the posture, and the "Responsible AI Practice" designation is granted only at the threshold.
A measured artificial intelligence
The AI Index brings the ten properties together into a single measure and places you in one of four bands. Most organizations start in Developing, and that is a normal place to begin. Run again each year, the measure shows the ground you have covered.
The designation asks for two things at once: the Index reaches the threshold, and no property falls below the floor. Holding the average is not enough if one property stays too low. The threshold is not negotiated; the gap gets closed.
Four reference frameworks, one measure: ISO/IEC 42001, the NIST AI RMF, the AI Act and Law 25 converge in the iDIA framework. Two evaluators, the same inputs, the same result.
From an emphasis matrix: each reference framework rates each property H, M, or L, and the weights normalize to 100. Change one emphasis rating and the weights recalculate.
On evidence, according to the five-level maturity scale: Absent, Declared, Documented, Operational, Measured and improved. A policy that is declared but not documented caps at 25.
No, it converges them. The framework translates four reference frameworks into a single measurable profile and prepares the organization to align with each one. Regulatory correspondences are alignment reference points; precise compliance is confirmed with legal counsel.
Read next: the Responsible AI Practice Program, the evaluation · the Responsible AI Practice designation · the AI Index diagnostic