Status
Claims, labelled by evidence.
INSPEKTORIUM applies claim discipline to its own work. Each capability below is marked by how far it has actually been demonstrated — implemented, tested, passed internally, in progress, planned, or explicitly not claimed.
Implemented & tested
Evidence span validation
Answers are checked against the exact source spans they cite.
Implemented + tested
Citation & source provenance
Every supporting citation is bound to a retrievable source.
Implemented + tested
Deterministic fingerprints
Runs produce stable, verifiable fingerprints for re-checking.
Implemented + tested
Internal research
Artifact-integrity fault containment
Corrupted artifacts blocked in a controlled internal experiment.
Internal experiment
Execution governance
Authorization and fail-closed execution gates, under study.
Internal research
In progress & roadmap
Semantic scientific inference checking
Does the evidence warrant the inference — not just the citation?
Passed internally — not yet in public repo · Phase 8
Independent disagreement review
Structured second-opinion and disagreement between reviewers.
In progress — Phase 9
Canonical audit & release workflow
A reproducible audit trail and release-integrity workflow.
Roadmap — Phase 10
Explicitly not claimed
General AI safety
INSPEKTORIUM studies specific properties under defined conditions.
Not claimed
Regulatory compliance
No compliance, approval or legal guarantee is provided.
Not claimed
Certification
Findings are documented research interpretations, not certifications.
Not claimed