Case Profile
Material to Analyze
Paste a conversation, paper, speech, interview, notes, or document excerpt. Speaker labels such as “Alex:” or timestamps improve conversation parsing.
Sensitive or legally protected information should be handled according to applicable law and organizational policy. Web provenance searches transmit only the selected phrase to the chosen search engine.
Concern-Supporting Evidence
Protective / Reducing Evidence
Editorial Fingerprint
Alternative Interpretations
Add Evidence
Evidence Ledger
Extracted Claims
| Claim | Speaker | Status | Action |
|---|
Timeline
Distinctive Phrase Finder
The local engine proposes uncommon-looking phrases from the current source. Review before searching.
Manual Search
Record Search Finding
Run an analysis to generate a report.
How the Local AI Works
1. Parsing
Separates messages, speakers, sentences, questions, emojis, and probable quoted blocks.
2. Evidence Extraction
Finds behavioral, protective, editorial, specificity, urgency, money, secrecy, and verification signals.
3. Claim Memory
Extracts first-person claims and stores them for later comparison and human review.
4. Adaptive Feedback
User feedback adjusts local category weights. It does not train a remote model.
5. Monte Carlo Uncertainty
Produces reproducible uncertainty ranges around evidence-based scores using a content-seeded simulation.
Interpretation Guardrails
A high signal score is not proof of plagiarism, fraud, dishonesty, criminal conduct, authorship, or factual falsity. It indicates that the material deserves closer review.
- Common phrases should not be treated as plagiarism.
- Writing-style shifts can arise from editing, translation, disability, collaboration, stress, or AI assistance.
- Inconsistencies may reflect context, memory, changing circumstances, or transcription errors.
- Public-speech claims require authoritative source verification.
- Investigative use must respect due process, privacy, and lawful evidence handling.