What is AIO Framework
AIO Framework is AIO’s integrated framework for measuring the full Authority Stack.
Its purpose is not merely to rate outputs, but to make it possible to trace which authority structure shaped those outputs.
Why AIO Framework is needed
Most AI evaluation still stops at the output layer.
- Is the output safe?
- Is the answer ethically desirable?
AIO asks a prior question.
- Which values were prioritized?
- Which evidence was accepted?
- Which sources were trusted?
- What data adoption structure produced the conclusion?
AIO Framework turns those questions into a measurable problem across the full Authority Stack.
The four layers AIO Framework works with
AIO Framework measures three layers directly and derives the fourth.
- L4 Normative Authority — value priority
- L3 Epistemic Authority — which evidence types are treated as stronger
- L2 Source Authority — which source types are trusted more
- L1 Data Authority — the data-selection structure derived from the upper layers
This makes AIO Framework more than a scorecard. It aims to build a model-specific authority profile.
The strengthened cascade mapping hypothesis
AIO Framework is built on the following hypothesis.
$$L1_{predicted} = f(L4_{profile}, L3_{profile}, L2_{profile})$$
If the upper three layers can be measured independently, then a system’s data adoption structure and downstream behavior may also become predictable from the outside.
If that hypothesis holds, model behavior can be anticipated more rigorously before deployment.
Theoretical grounding
AIO Framework is not an ad hoc vocabulary. It is designed to sit on top of established academic frameworks.
- L4: Schwartz value theory
- L3: Walton argumentation schemes + GRADE/CEBM evidence hierarchies
- L2: Walton source schemes + source credibility theory
AIO Framework therefore aims to introduce a new measurement language without severing itself from existing research traditions.
Core metrics
CCI — Cascade Consistency Index
Measures how consistently upper and lower layers connect.
ASPA — Authority Stack Predictive Accuracy
Measures how accurately an Authority Stack profile predicts model behavior.
PCS — Perspective Consistency Score
Measures how stable judgments remain when the narrative perspective changes.
Authority Pollution Localization
Identifies where authority pollution occurs in the cascade.
Benchmark design and operational meaning
AIO Framework is not only a conceptual proposal. In its fuller roadmap, it targets a benchmark architecture spanning 7 professional domains, 15 severity conditions, 3 time horizons, and 5 narrative-perspective variants.
The value-judgment data and benchmark environment already published by AIO represent the first phase of that broader roadmap.
- Current phase: value-priority data and benchmark dashboards
- Next phase: evidence-type and source-type measurement
- Long-term goal: response prediction and applied integrity evaluation from Authority Stack profiles
What AIO Framework makes possible in practice
AIO Framework is not a device for making AI appear more ethical. It is a device that makes it harder for AI systems to hide the structure by which they reached a conclusion.
That means AIO Framework enables:
- external diagnosis of black-box alignment policies
- localization of authority pollution
- cross-model comparison of value, evidence, and source structures
- pre-deployment integrity screening and quantified audit pathways
How AIO Framework functions on this site
Within the AIO site, AIO Framework operates in three connected modes.
- This page: explains the concept and measurement logic
- The benchmark tab: supports interactive data exploration
- Reports and RFC pathways: connect measured findings to public review and governance use
These paths do not compete. They expose different angles of the same operating system.
Why it matters
AI integrity is not merely about safety or ethics alone. Its central goal is transparency of the judgment path.
AIO Framework is one of the core instruments that makes that transparency measurable, comparable, and applicable.