# Accident Investigation Prover MCP for AI Agents AI Agent Connect > Accident Investigation Prover enforces ICAO Annex 13 and NTSB standards on your AI client. It stops lazy 'pilot error' conclusions by forcing evidence correlation across FDR data, CVR transcripts, and maintenance logs. It ensures every finding maps to a multi-causal chain, including organizational factors and specific, measurable, addressed recommendations to prevent recurrence. ## Overview - **Category:** aviation - **Price:** Free - **Endpoint:** https://edge.vinkius.com/vk_preview_lW5eq02uyQWDJiNGsxX901icN7Fl8AQ8SwMOpToG/ai-agent-connect - **Tags:** aviation, accident-investigation, icao, ntsb, hfacs, reason-model, safety, cvr, fdr, root-cause, prover ## Description Accident Investigation Prover enforces NTSB and ICAO Annex 13 investigation rigor to ensure your aviation safety audits are technically sound. When you're analyzing incidents, the biggest risk is a shallow investigation that blames a human without looking at the system. This Connector changes that by acting as a rigorous gatekeeper. Instead of letting your agent give you a surface-level summary, it forces the tool to verify every claim against hard data. It demands FDR parameters, CVR timestamps, and maintenance records before it'll let you move to a conclusion. It ensures you're looking at the Swiss Cheese model of failure, tracing issues from active errors back to latent organizational pressures like scheduling, budget, and regulatory gaps. By using this through the Vinkius catalog, you turn your AI into a technical expert that follows the HFACS taxonomy and produces actionable, evidence-linked recommendations. It moves you away from 'improve training' as a generic fix toward specific, trackable actions for named authorities. This isn't about guessing what happened; it's about proving why it happened by connecting the dots between the cockpit, the hangar, and the boardroom. ## Tools ### validate_accident_investigation Correlates evidence chains and classifies factors into the HFACS taxonomy. It forces the agent to identify organizational root causes and write measurable recommendations. ## Prompt Examples **Prompt:** ``` Investigate a B737 crash where the autothrottle was deferred for 5 days and the pilot had been on duty for 11.5 hours. ``` **Response:** ``` ### Multi-causal Analysis * **Probable Cause:** Loss of airspeed awareness * **Contributing Factors:** * Autothrottle deferral (Maintenance/Level 2) * Fatigue at 11.5h FDP (Precondition/Level 2) * Inadequate CRM monitoring (Supervision/Level 3) * 24-month training gap (Organizational/Level 4) * 28-month audit gap (Regulatory/Level 4) **Conclusion:** 'Pilot error' rejected as root cause. Investigation reaches Level 4 organizational factors. ``` **Prompt:** ``` Analyze an engine failure caused by a third-party part that was 23% cheaper than the OEM. ``` **Response:** ``` ### Organizational Factor Analysis * **Probable Cause:** Uncontained engine failure * **Root Causes:** * **Cost-driven supplier selection** (Organizational/Level 4) * **Traceability gap** in part documentation (Supervision/Level 3) * **Inadequate authority oversight** of third-party suppliers (Regulatory/Level 4) **Recommendations:** 1. Establish a supplier audit frequency. 2. Implement a traceability chain verification. 3. Create an OEM vs. third-party risk assessment framework. ``` **Prompt:** ``` Review a terrain impact where the FO was on their first flight to a mountain airport. ``` **Response:** ``` ### CFIT Investigation * **Probable Cause:** Controlled Flight Into Terrain (CFIT) * **Contributing Factors:** * Visual illusion below MDA (Level 1) * Inadequate crew pairing for high-risk airport (Supervision/Level 3) * Absent special airport qualification program (Organizational/Level 4) **Note:** GPWS response time was insufficient (4s). Investigation must address organizational factors regarding crew qualification. ``` ## Capabilities ### Correlate FDR and CVR data Matches flight recorder parameters with cockpit voice transcripts and ATC tracks. ### Construct multi-causal chains Maps the probable cause and contributing factors using NTSB-style formatting. ### Classify factors via HFACS Categorizes every finding into one of four levels, from unsafe acts to organizational climate. ### Trace organizational pressures Identifies systemic issues like maintenance economics, training budgets, and regulatory gaps. ### Generate actionable recommendations Creates specific, measurable, and evidence-linked actions for safety authorities. ## Use Cases ### Engine Failure Analysis Investigating a turbine disk separation to find out if it was a maintenance gap or a supplier cost-cutting issue using validate_accident_investigation. ### Terrain Impact Review Analyzing a CFIT incident to see if visual illusions were compounded by inadequate crew pairing or training gaps. ### Training Audit Evaluating why a pilot lacked specific skills by tracing back to budget cuts and syllabus coverage gaps. ### Maintenance Log Review Identifying if a recurring part failure is linked to MEL deferral rates or component life limit issues. ## Benefits - Stop 'pilot error' fallacies: The Connector rejects single-cause conclusions by forcing a multi-causal analysis of the entire accident chain. - Ensure evidence correlation: It checks that every conclusion has supporting data from FDR, CVR, and maintenance logs. - Map to HFACS taxonomy: Every factor gets classified into the correct level, ensuring the investigation goes deep enough into organizational issues. - Identify systemic root causes: It automatically looks for scheduling pressures, budget constraints, and regulatory gaps. - Create actionable fixes: You get specific, measurable, and tracked recommendations instead of vague 'improve training' statements. ## How It Works The bottom line is your agent stops guessing and starts performing a professional-grade aviation safety audit. 1. Provide your agent with the raw investigation data, including logs, transcripts, and wreckage analysis. 2. Call validate_accident_investigation to trigger a structured reflection on the evidence and causal chains. 3. Receive a rigorous report that identifies evidence gaps and maps every factor to the HFACS taxonomy. ## Frequently Asked Questions **How does Accident Investigation Prover help with aviation safety?** It forces your AI to follow ICAO Annex 13 standards, ensuring you don't miss systemic issues like maintenance logs or organizational pressures. **Can this tool help me find the root cause of a flight incident?** Yes, it uses Reason's Model and the HFACS taxonomy to trace active failures back to latent conditions like budget constraints or training gaps. **Does Accident Investigation Prover check FDR and CVR data?** It requires the AI to correlate FDR parameters and CVR transcripts with ATC tracks and maintenance records before it'll allow a conclusion. **Will this help me write better safety recommendations?** It rejects vague goals like 'improve training' and forces the agent to write specific, measurable, and evidence-linked actions for named authorities. **Is Accident Investigation Prover suitable for NTSB-style reports?** Yes, it's built specifically to enforce the rigor required by NTSB and ICAO for multi-causal accident analysis. **Can I use Accident Investigation Prover to audit airline maintenance?** It can help you analyze MEL deferral rates and component life limits to see if they contributed to a specific incident. **How does this prevent 'pilot error' as a root cause conclusion?** The engine maintains a semantic trap list of blame-language signals: 'pilot error,' 'crew error,' 'human error,' 'judgment error,' 'failed to,' 'negligence.' If the LLM uses any of these in the HFACS taxonomy field, the classification is rejected. Instead, the LLM must classify each factor at the correct HFACS level: Level 1 (what the pilot did — skill error, decision error, perceptual error, or violation), Level 2 (what conditions enabled it — fatigue, CRM failure, environment), Level 3 (what supervision allowed it — scheduling, training gaps), Level 4 (what organizational decisions created it — budget cuts, staffing, regulatory gaps). 'Pilot error' is Level 1 only — the investigation must reach Level 4. **What evidence sources must be cross-referenced?** Five mandatory sources: (1) Flight Data Recorder — minimum 88 parameters per ICAO Annex 6, with timestamps. (2) Cockpit Voice Recorder — last 2 hours of audio, transcribed with timestamps, correlated with FDR parameter changes. (3) ATC recordings — radar track, clearances, handoffs, weather advisories. (4) Maintenance logs — last A/B/C/D checks, MEL items, deferred defects, AD compliance, component life (TSN/TSO/CSN/CSO). (5) Wreckage analysis — impact signatures, fire patterns, fracture analysis (fatigue vs overload), metallurgical examination. The engine rejects speculative language like 'it appears that' or 'evidence suggests' when hard data from these sources exists. **Why does this require recommendations to be addressed to specific authorities?** Because 'improve training' with no addressee has zero accountability. The NTSB model requires each recommendation to name the authority responsible for implementation — the FAA, the operator, the manufacturer, or an international body. Each recommendation must include: what specific action to take, measurable success criteria, a response deadline (typically 90 days), a verification mechanism (audit, inspection, data review), and a direct link to a specific investigation finding. This is how aviation achieved its extraordinary safety record — not through vague wishes, but through tracked, accountable, evidence-linked changes.