Accident Investigation Prover MCP. Forces NTSB-grade rigor into every incident report.

You're running into a problem with most standard models; they treat accident conclusions like some kind of narrative fiction instead of what they actually are: an evidence-based reconstruction. The validate_accident_investigation tool forces your agent to perform the rigorous, multi-axis analysis required by ICAO Annex 13 and NTSB standards. It doesn't just spit out a probable cause; it builds the entire investigative backbone you need.

It starts with correlating technical evidence across every single available data point. You feed in FDR parameters, CVR transcripts, ATC radar tracks, and maintenance logs, and the tool cross-references all that against physical wreckage analysis. It doesn't accept a standalone finding; it demands to know how those pieces of tech correlate with what was physically found.

This forces an airtight link between recorded data and reality.

When mapping out the probable causes, you aren't getting vague suggestions. The tool constructs multi-causal chains following the established NTSB format: 'The probable cause was [X], contributing to which were [Y, Z, W].' It doesn't stop there; it applies Reason's Model to trace systemic root failures through inadequate defenses and latent conditions.

You’ll see how the agent uses a five-whys approach to drill down past immediate operational errors until it hits the core system failure.

The investigation mandates classifying every single human error factor using the HFACS taxonomy across four distinct levels. It separates Level 1 (Unsafe Acts) from Level 2 (Preconditions), Level 3 (Supervision), and Level 4 (Organizational factors). If everything you give it clusters at just one level, the tool flags that investigation as too shallow; it forces a deep dive into human performance failure points.

This moves beyond simply blaming the pilot.

The analysis then targets systemic weaknesses by tracing failures back to organizational pressures. It won't let you ignore scheduling utilization rates, training budget shortfalls, maintenance deferral metrics, or regulatory oversight gaps. The tool proves that operational issues are never solely about the crew in the cockpit; they're rooted in system pressure points.

Finally, when it comes time for recommendations, it refuses to accept vague directives like 'Improve communication.' Instead, it forces you into drafting truly actionable items. Every recommendation has to be Specific (what action must happen), Measurable (how do we prove success?), Assigned (to a specific named authority), and Tracked with a clear follow-up timeline linked directly back to the original evidence found during the investigation.

You'll walk away with findings that are not just smart, but actionable for actual regulatory change.