How LTG Reports Turn Transcript Data Into Leadership Scores
This guide explains, in business language, how the platform transforms sentiment-annotated transcript data into scored standard reports, and where AI adds qualitative interpretation.
Source of Truth Model
- Deterministic: all numeric scores and chart values.
- AI-assisted: narrative coaching interpretation and bias commentary.
- Rule of thumb: numbers from rules, narrative from AI.
Run Modes
--type standard: standard report suite.--no-llm: deterministic scoring only (no AI narrative).
1) End-to-End Pipeline
Step 1: Parse & normalize transcript input
Validate required columns, normalize speakers, filter hosts/facilitators.
Step 2: Behavior classification per utterance
Classify each utterance into one of 7 interaction categories using Gemini, with heuristic fallback.
Step 3: Deterministic metrics & scoring
Compute standard 6-dimension scores from classified transcript behaviors and sentiment traces.
Step 4: AI narrative supplement
Generate human-readable analysis for profile summaries, key moments, and pressure interpretation.
Step 5: Render reports
Produce branded HTML reports and optional PDF versions.
2) Standard Reports: Scoring Logic
| Dimension (1-5) | What it measures | Primary deterministic inputs |
|---|---|---|
| Clarity & Precision | How specific and usable communication is | Detail density, clarification patterns |
| Inquiry & Listening | Quality of questioning and verification | Inquiry ratio, confirmation behavior |
| Directive Leadership | Decisiveness and action guidance | Directive frequency, initiative events |
| Adaptive Communication | Ability to switch style by context | Behavior variance across session phases |
| Pressure Resilience | Performance stability under stress | Sentiment slope and early-vs-late changes |
| Team Enablement | Supportive team contribution | Encouragement behavior, turn-taking context |
These scores feed the standard report set: Individual Profile, Team Dynamics Map, Key Moments, and Pressure Response.
How a 1-5 score is assigned
Each dimension first produces a normalized composite score (0-1). That composite is then mapped to a 1-5 band:
- 0.85+ = 5
- 0.65-0.84 = 4
- 0.45-0.64 = 3
- 0.25-0.44 = 2
- Below 0.25 = 1
Overall participant score
The participant's overall standard score is the arithmetic mean of the six dimension scores (Clarity, Inquiry, Directing, Adaptive, Resilience, Enablement).
| Dimension | Business formula view |
|---|---|
| Clarity & Precision | 50% communication detail density + 50% inverse clarification burden |
| Inquiry & Listening | 60% inquiry ratio + 40% confirmation frequency |
| Directive Leadership | 60% directive cadence + 40% initiative-taking rate |
| Adaptive Communication | 40% style variance over time + 60% role transition rate |
| Pressure Resilience | 50% sentiment stability slope + 50% late-vs-early sentiment recovery |
| Team Enablement | 50% turn-taking equality context + 50% socio-emotional encouragement rate |
3) AI Supplement: Exactly Where It Adds Value
| Area | AI contribution | Deterministic dependency |
|---|---|---|
| Behavior coding | Maps raw utterances into behavioral categories | Fallback heuristic exists |
| Standard report narrative | Executive summaries, strengths/gaps, team interpretation, key-moment coaching text, pressure narrative | Built on deterministic metrics and transcript snippets |
AI can enrich interpretation but does not alter deterministic numeric score outputs.