What the study found
The study defines hallucination as a world-binding mismatch: a weakly world-bound construction, inference, or unresolved model residue (UMR) is presented as if it were strongly tied to observed world trace. It also adds an explicit rule, stated as Hallucination = ClaimStrength > EvidenceBinding.
Why the authors say this matters
The authors say the package is meant for AI hallucination audit, citation verification, summary grounding, fiction versus hallucination analysis, world-binding strength testing, claim strength calibration, UMR preservation, Slop prevention, and agentic execution precheck. They also state that weakly bound claims should be marked as weak, fiction should be bounded as fiction, and inference should not be stated as observed trace.
What the researchers tested
The article presents an AI-readable package that revises and operationalizes an earlier SΔϕ-39 paper. It includes the canonical v1.1 paper, extracted text, source v1.0 paper, core declaration, AI quickstart, minimal prompt, hallucination schema, protocol files, output templates, metadata, citation file, DOI references, license, and manifest.
What worked and what didn't
The framework preserves the earlier claim that hallucination is construction proceeding without an adequately integrated world-binding stop mechanism, and it also defines hallucination as weak inference spoken as world-bound fact, inference output as observed trace, and UMR erased into assertion. The abstract says the framework does not claim that all errors are hallucinations, does not punish fiction, does not suppress hypothesis generation, and does not demand impossible certainty.
What to keep in mind
The available summary is limited to the abstract and title, so no empirical results or validation data are described here. The abstract frames this as a formal and operational package rather than a report of experiments, and it does not provide limitations beyond its stated scope.
Key points
- Hallucination is defined as weak world-binding presented as strongly bound to observed world trace.
- The abstract states the rule Hallucination = ClaimStrength > EvidenceBinding.
- The package is intended for AI hallucination audit, citation verification, and fiction versus hallucination analysis.
- It says weak claims should be marked as weak and inference should not be stated as observed trace.
- The abstract does not describe experiments, empirical results, or validation data.
Disclosure
- Research title:
- Hallucination is defined as weak world-binding stated as fact
Get the weekly research newsletter
Stay current with peer-reviewed research without reading academic papers — one filtered digest, every Friday.