Evidence Packs (L0 / L1 / L2)
- Executive Summary
- Working Knowledge
- Technical Spec
Evidence Packs are ReGenesis's key differentiator. While other AI coaching tools generate generic summaries, ReGenesis produces evidence-backed insights where every claim links to source material — with transcript quotes and video jump links.
The Three Levels
| Level | Name | What It Contains | Who Sees It |
|---|---|---|---|
| L0 | Insight | The core finding or pattern | Coachee + Coach |
| L1 | Reasoning | How the AI reached this conclusion | Coach (optionally Coachee) |
| L2 | Source Evidence | Direct quotes, timestamps, video jump links | Coach (configurable) |
Why This Matters
- Trust: Coaches can verify every AI insight before approving it — no black-box AI
- Hallucination prevention: L2 references are cross-validated against source transcripts. If the AI claims something, the coach can click through and hear the coachee say it.
- Competitive moat: No competitor offers this level of evidence chain. It transforms AI from a "nice suggestion" to a "defensible observation."
- Regulatory compliance: GDPR Art. 22 (right to explanation for automated decisions) is satisfied by L1 reasoning chains
- Coach empowerment: Coaches do not need to trust the AI blindly. They verify, edit, and approve — maintaining their professional authority.
Example Evidence Pack
L0 (Insight): "Coachee demonstrates a recurring pattern of avoiding delegation, particularly for high-stakes deliverables."
L1 (Reasoning): "This pattern was identified across 4 sessions (Jan 8, Jan 22, Feb 5, Feb 19). The coachee used phrases like 'I just can't let go of the details' and 'no one else will get it right.' This correlates with the control-oriented leadership style identified in their Hogan assessment."
L2 (Source): [Session Feb 5, 23:45] "I stayed up until 2 AM redoing the slides because I didn't trust that Marcus would..." [Click to jump to video]
How Evidence Packs Get Built
The Generation Flow
- Sasha processes a session transcript and identifies themes, patterns, action items, and growth areas
- For each insight (L0), Sasha also generates:
- The reasoning chain (L1): step-by-step logic explaining how it reached the conclusion
- Source references (L2): specific transcript excerpts with timestamps
- Cross-validation: The Evidence Pack Builder checks that L0 claims are actually supported by L2 references
- The coach reviews the complete pack before anything reaches the coachee
- The coachee sees L0 (always) and optionally L1 (if the coach/tenant allows)
Coach vs. Client Evidence Pack
The same insight has two views:
Coach View (Full)
- L0: The insight
- L1: Complete reasoning chain
- L2: All source references with video jump links
- AI provenance: model version, confidence score, generation timestamp
- Edit/approve/reject controls
Client View (Filtered)
- L0: The insight (as approved/edited by coach)
- L1: Reasoning summary (if tenant setting allows)
- L2: Usually hidden (coach-only by default)
- Feedback controls: helpful/not helpful, request more context
Video Jump Links
When a source reference includes a specific moment in a recorded session, the Evidence Pack includes a "jump link" — clicking it opens the session recording at that exact timestamp. This is the ultimate proof: the coachee can hear themselves say the thing the AI identified.
Implementation:
- Transcripts include word-level timestamps from the transcription service
- When Sasha references a quote, the Evidence Pack Builder looks up the timestamp range
- The jump link is constructed as:
https://app.regenesis.ai/sessions/{id}/recording?t={seconds} - The video player (on web and mobile) supports deep linking with timestamp parameters
Cross-Session Insights
The most powerful insights span multiple sessions. Sasha can identify patterns that build over weeks or months:
- "Over the past 6 sessions, coachee has progressively shifted from avoidance language to ownership language when discussing team conflicts"
- Each cross-session insight includes L2 references from multiple sessions, showing the progression
Confidence Scoring
Every insight has a confidence score (0.0 to 1.0):
| Score Range | Meaning | Coach Action |
|---|---|---|
| 0.9 - 1.0 | High confidence, strong evidence | Quick approve likely |
| 0.7 - 0.89 | Good confidence, solid evidence | Review recommended |
| 0.5 - 0.69 | Moderate confidence, partial evidence | Careful review needed |
| Below 0.5 | Low confidence | Auto-flagged for careful review |
Low-confidence insights are flagged with an admonition in the coach dashboard.
Evidence Pack Architecture Diagram
The cross-validation engine is the primary defense against AI hallucination. Every L0 insight must have at least one verified L2 source reference. If the validator finds zero valid L2 references, the insight is automatically flagged and sent for manual review. The coach sees a warning badge.
At current Anthropic pricing (Claude 3.5 Sonnet), generating a full evidence pack for a 60-minute session costs approximately $0.05-$0.15 in LLM API fees. This includes two LLM calls (insight generation + reasoning generation) plus embedding generation for cross-validation. See Tech Stack for detailed cost analysis.