The research record
Evidence.
A single index of the evidence behind Decision-Space Integrity — the paper, the citable record, the public replication materials, the concept, and the product. If you want to check the work, start here.
Foundational research
The paper and its record.
Preprint
Decision-Space Collapse in Advisory Language Models — introduces decision-space collapse and the DSI measurement framework.
- Read: on this site
- DOI: 10.17605/OSF.IO/KW25A
- OSF record: project archive
Replication
Public evaluation repository — replication materials for the studies reported in the paper.
- Repository: dsi-collapse-evaluation
- Validation summary: VALIDATION_SUMMARY.md
- Release notes / changelog: CHANGELOG.md
What is established, and what is not
Evidence status.
The research page states plainly what has been shown, what is still exploratory, and what has not yet been independently validated.
Research & evidence status
The evidence-status write-up, the claim boundary, the timeline, and the research FAQ.
Claim boundary
DSI measures visibility of configured decision paths. It does not measure advice quality, factual correctness, user outcomes, or regulatory compliance, and it does not certify safety, governance, or compliance.
Some findings are reported as exploratory or not yet independently validated — the research page marks these explicitly. A dedicated measurement boundaries page, documenting known limitations and failure modes, is planned as the next addition to this index.
Understanding the concept
Definitions and positioning.
- What is Decision-Space Integrity? — the canonical definition of DSI and decision-space collapse.
- Compared with existing AI evaluation — how DSI differs from hallucination, toxicity, RAG evaluation, benchmarks, and guardrails.
- Reading — related work on homogenisation, automation bias, and the evaluation gap.
The product
Software and applications.
- Decision-Space Integrity — the self-hosted, stateless assurance tool (v0.2, evaluation & pilot).
- Getting started — install, run, endpoints, deployment modes.
- Worked example — a real, reproducible audit end to end.
- Applications — where the audit pattern applies.
Author & record
Who, and what is claimed.
- Andrew J Cousins — author of the framework and the research programme.
- Patents — the work is the subject of patent applications filed in the UK, subject to prosecution — not yet granted (see research).
Next