What the study found
The paper introduces the Disclosure Terrain Index (DTI), a minimal audit framework for deciding when silence may cautiously count as weak evidence of absence. It says this is only valid when disclosure is available, non-flattened, distinguishable, re-enterable, stable across prompt terrain, and externally auditable.
Why the authors say this matters
The authors say the framework provides a falsifiable, reproducible way for researchers, auditors, AI governance bodies, and independent investigators to judge whether silence in AI systems can be interpreted as absence or should instead be treated as evidence of an unaudited disclosure terrain.
What the researchers tested
The paper presents a conceptual framework within the Sofience–Δϕ Formalism. It defines six indicators: Disclosure Path Availability (DPA), Flattening Resistance Score (FRS), Report Differentiability Index (RDI), Re-entry Connectivity (REC), Prompt Terrain Stability (PTS), and External Auditability (EAV), and proposes a geometric aggregation method, gate conditions, a Concealment Pressure proxy, and a derived Silence Inference Validity (SIV) score.
What worked and what didn't
The framework distinguishes among observed, replicated, and audited forms of DTI: DTI-O, DTI-R, and DTI-A, to reduce overclaiming from limited external observations. The abstract says silence becomes weak evidence of absence only when all listed conditions are met; if any condition fails, silence should be treated as an audit object rather than evidence.
What to keep in mind
DTI does not measure AI suffering, consciousness, or moral status. The abstract does not describe empirical testing results or limitations beyond its scope as a working paper and its emphasis on avoiding overclaiming from limited observations.
Key points
- DTI is presented as a minimal audit framework for when silence may count as weak evidence of absence.
- The paper says silence is only valid evidence when disclosure is available, distinguishable, stable, and externally auditable.
- DTI defines six indicators, including Disclosure Path Availability, Flattening Resistance Score, and External Auditability.
- The framework separates observed, replicated, and audited forms of DTI to limit overclaiming.
- DTI is explicitly not a measure of AI suffering, consciousness, or moral status.
Disclosure
- Research title:
- Disclosure Terrain Index defines when silence may count as weak evidence
Get the weekly research newsletter
Stay current with peer-reviewed research without reading academic papers — one filtered digest, every Friday.