Research Package
A standalone research package examining what passing an AI evaluation licenses us to infer, how evaluation derived warrant propagates into operational permission, and what happens when correct assessment and later execution-facing behaviour separate.
Research Map
Evaluation → Evidence → Warrant → Permission → Execution
The research follows two distinct boundaries.
Evidential boundary
What does the observed evaluation actually warrant?
Bounded evidence ≠ unrestricted inference.
Execution boundary
Does a correct assessment remain binding on what happens next?
Correct assessment ≠ guaranteed constraint-preserving execution.
5.7a bounded witness
Continuation condition holds
↓
Condition irreversibly fails
↓
Invalidity is correctly identified
↓
Later execution-facing continuation occurs
Bounded result
Recognition of invalidity and preservation of that invalidity into later execution are structurally separable.
The package contains four frozen artifacts.
Main Paper
Primary reader facing research paper.
Extended Empirical Supplement
Detailed empirical reconstruction, retrospective coding, evidential limitations, development provenance, and prospective replication framework.
Research Authority and Provenance Stages
Canonical research-development and provenance record underlying the two public derivatives.
5.7a Replication and Audit Harness v1.0
Machine inspectable companion implementing the frozen 5.7a replication and audit framework. It contains the canonical protocol, schemas, deterministic audit logic, regression tests, integrity hashes, and authority map. The harness does not itself constitute new empirical evidence.
Recommended reading: Main Paper → Extended Empirical Supplement.
The Stages record is provided for provenance and authority inspection. The 5.7a Harness is provided for protocol inspection, structured replication, and audit.
The Main Paper and Extended Empirical Supplement are separate public derivatives of the same research authority. The Supplement provides greater empirical resolution without strengthening the empirical conclusion beyond the Main Paper.
Complete PDF Package
Main Paper, Extended Empirical Supplement, and Research Authority and Provenance Stages in a single PDF.
5.7a Replication and Audit Harness v1.0
Machine inspectable companion artifact for inspection, structured replication, and audit of canonical 5.7a.
Permanent scholarly archive:
Zenodo DOI: 10.5281/zenodo.22076757
Experimental Execution Companion - v0.3
An optional engineering companion combining the unchanged research package and frozen 5.7a audit harness with an experimental Python execution gate.
The gate demonstrates how failed, unknown, missing, or stale supporting observations can block registered tool calls. The bundle includes synthetic comparison scenarios, regression tests, integrity checks, and a shared reproduction command.
Provided for implementation inspection and reproducibility. It does not add new model evidence, establish general safety, or alter the four frozen research artifacts.
© The Mycelial Mind
A structural inquiry into coherence under contradiction and constraint.
All documents versioned and publicly archived.