Elenchus
Audit meeting transcripts for flawed reasoning and unfair conversation dynamics.
The problem
High-stakes decisions get made in meetings every day. Whether the reasoning was sound, and whether every voice in the room was actually heard, almost never gets audited. Compliance reviews, legal discovery, and HR investigations rely on transcripts that nobody has the time to read closely.
What it does, and who it is for
Elenchus reads meeting transcripts and flags 38 types of logical fallacies, plus interruptions, dominance, and silencing patterns. Each finding ships with a confidence score and an alternative reading, so reviewers see possibilities, not verdicts. Ships as a CLI, a Microsoft Teams bot, and a web app.
- Compliance officers and legal teams
- HR and investigations groups
- Regulators auditing fair-process requirements
- Executive teams reviewing high-stakes deliberations
Proof at a glance
38
Fallacy types detected
1,000+
Detection patterns
89.9%
Precision (CI-gated)
EU AI Act Art. 14
Human-override compliant
Capabilities
38 fallacy types across 7 categories
Ad hominem, straw man, false dilemma, slippery slope, appeal to authority, and 33 more, each with a confidence breakdown and an alternative interpretation.
Conversation dynamics
Interruption detection, speaker dominance, silenced participants, and overall meeting health on a 0 to 100 scale.
Defensible methodology
CI-gated precision and recall measured on a hash-pinned 10-transcript answer-key corpus. Documented error rates, testable methodology, chain-of-custody audit trail.
Fairness validation
Disparate-impact analysis using the four-fifths rule, tested across five demographic groups and four communication styles.
Multiple platforms
CLI for batch transcript analysis, Teams bot for live meetings, and a web app with audit logs and feedback loops.
Compliance and trust
Aligned with: EU AI Act Article 14 (human oversight), NIST AI Risk Management Framework, Daubert Standard for expert evidence
- Confidence-score breakdown on every detection, never absolute verdicts
- Alternative interpretations rendered alongside each finding
- Audit trail with session logging and SHA256 integrity hashing
- Explicit human-override capability built into the workflow
- Disparate-impact analysis available out of the box
How it deploys
Elenchus runs on-premises or in your cloud. The deterministic pattern engine has no external dependencies. An optional LLM-assisted layer is off by default and uses your own API key when enabled.
What is coming next
- Phase 8 cross-utterance fallacy detection
- Expanded answer-key corpus and additional language support
- Deeper Teams native integration