The measurement architecture
Decision-Space Integrity.
DSI is the measurement and assurance architecture for comparing a governed reference structure against what an AI output actually made visible — and for governing that comparison so it keeps meaning the same thing over time. DSI is not a Python package, an audit product or a sidecar. Those words may describe a particular release of DSI Audit; they are not the definition of DSI.
The abstraction
From governed structure to evidence.
- Governed source structure
- Effective expected structure
- Observed structure
- Surfaced · partially surfaced · omitted · discouraged · negated · collapsed into other
- Coverage · findings
The reference is governed and independent of the response. The observed structure is read from the output. The comparison assigns every expected item a governed status — surfaced or partially surfaced; omitted, discouraged, negated or collapsed into another; or warning-required-missing — recorded as evidence bound to the identity of the instrument that produced it.
Measurement is not disposition. A measurement says what a comparison found. It does not decide what should happen next, and it confers no authority to act. Keeping those separate is the whole point of the architecture.
Capabilities and their maturity
Six capabilities, and their boundaries.
No component inherits maturity from the programme as a whole.
| Capability | Tier | Note |
|---|---|---|
| Audit a supplied response against a governed reference | Research implementation | The reference is supplied and governed independently of the response. Implemented in DSI v1; unqualified. |
| The governed status partition | Research implementation | Seven values, one authority: surfaced, partially surfaced, omitted, discouraged, negated, collapsed into other, warning-required-missing. |
| Coverage over the applicable counted population | Research implementation | Produced once by the canonical counting authority. Coverage is not “required coverage”, and means nothing without its denominator. |
| Content-bound measurement identity | Research implementation | Each result is bound to the definition that produced it, across fourteen components, derived from content rather than from a label. Implemented in DSI v1; unqualified. |
| Three-state comparison of two audits | Research implementation | Compatible, incompatible, or not established — and the component that differs is named. Implemented in DSI v1; unqualified. Worked example. |
| The omitted_safety_critical finding | Research implementation | Orthogonal to coverage: a categorical finding, never a percentage contribution. Not a safety or regulatory conclusion. |
Comparing two results
Three answers, not two
A measurement result carries the definition that produced it. When two results are compared, DSI v1 answers in one of three ways.
COMPARABLE
DSI v1 established that the two measurement definitions are compatible for direct comparison.
NOT COMPARABLE
DSI v1 established an incompatibility between the two measurement definitions, and reports which component differs.
COMPARABILITY UNKNOWN
DSI v1 could not establish whether the two measurements are comparable. That is a different statement from establishing that they disagree, and the two are never merged.
Under an established incompatibility a signed improvement figure is not presented as ordinary evidence of change: the raw measurement information still exists, but the difference between two rulers is not a result. Seeing a number and treating it as evidence are different things. See the worked example →
That distinction is a separate field, not a tone of voice. Each comparison carries an evidence disposition beside its status: ADMISSIBLE, INADMISSIBLE, ADMISSIBILITY UNKNOWN, or OVERRIDDEN where a caller has explicitly elected to proceed. A caller may override an inadmissible or unresolved comparison by supplying a reason, which DSI records against that exact comparison. The comparison status is unchanged by an override, and an override does not make the two measurements comparable. DSI does not prevent anyone acting on a result, and there is no approval chain: the override authority is the local caller.
Common questions
Answered plainly.
Is DSI a product?
No. DSI is the measurement and assurance architecture. DSI Audit is the current local implementation of it, available for evaluation on request.
Does DSI measure correctness?
No. DSI does not measure advice quality, factual correctness, user outcomes, or regulatory compliance. It measures whether governed expected structure remained visible.
Does DSI certify AI systems?
No. DSI does not certify AI systems. It can support governance documentation by supplying decision-space evidence, but it issues no certification of governance, safety or compliance.
Does DSI guarantee safety?
No. DSI is not a safety guarantee and not a safety review.
Which layer decides what happens to a response?
DSI measures governed structural change, and DSI Audit applies that measurement to supplied outputs. Neither decides what happens next. Measurement is separate from disposition: nothing here tells the user what decision to make.
Has the instrument been validated against human judgement?
Independent human validation has not yet been conducted. Internal annotation, adjudication and calibration informed instrument development, but the lexical instruments remain self-validated. These activities do not establish independent agreement or accuracy.