AI systems that prove what they did.

What broke

An adversarial audit of my own build found six real bugs the first pass had praised — unstable widget keys silently wiping form state, UTC drift, a success message that never rendered. All fixed before anyone else touched the system. That's what the audit is for.

Findings

The research effort demonstrated that publicly circulated investor lists often contain outdated, duplicated, or poorly substantiated information. Applying systematic OSINT verification significantly reduced noise while increasing confidence in the remaining records.

Key findings included:

• Built a database of 500+ OSINT-verified seed investor profiles supported by documented public sources.
• Eliminated duplicate, stale, and unverifiable records through structured verification.
• Identified consistent data-quality issues across widely shared fundraising databases.
• Established a repeatable verification methodology emphasizing provenance, traceability, and confidence scoring.
• Organized investor information to support due diligence, outreach prioritization, relationship tracking, and fundraising workflows.
• Demonstrated that smaller, evidence-backed datasets often provide greater operational value than substantially larger, unverified collections.

Findings

Development of Sidebar produced several operational findings regarding legal information management and investigative workflows.

• Demonstrated that structured conceptualization substantially improves the organization of complex legal matters by transforming scattered documents into interconnected people, events, locations, organizations, and evidence.
• Applied fact particularization to decompose broad narratives into discrete, verifiable factual assertions, making inconsistencies, evidentiary gaps, and corroborating information easier to identify.
• Used OSINT-derived public information, where appropriate and legally obtained, to enrich client, witness, and entity profiles while preserving source attribution and traceability.
• Reduced information fragmentation by linking documents, timelines, evidence, notes, and entities within a unified matter workspace.
• Established an evidence-centered workflow emphasizing provenance, confidence assessment, and auditability for every significant factual assertion.
• Demonstrated that AI-assisted retrieval and structured organization can substantially reduce the time required to review large case files while maintaining attorney oversight and decision-making authority.
• Reinforced the importance of explainable outputs, where every generated insight can be traced back to supporting documents or publicly available sources.