replit.md

EDCM Analyzer Platform

Overview

Privacy-first Energy-Dissonance Circuit Model (EDCM) Analyzer with multiple modes:

• Dating Mode: Safety & clarity checks for personal interactions
• Political Mode: Rhetoric and discourse pattern analysis
• Lab Mode: Scientific visualization of interaction dynamics
• Analysis Mode: AI-assisted conversation analysis with artifact management

The platform analyzes text and audio inputs for interaction dynamics (pace, pressure, clarity, balance) while strictly avoiding emotion detection, intent inference, or diagnostic claims.

Architecture

• Local-first: All data stored in IndexedDB by default
• Cloud sync: Optional, explicit opt-in only
• Privacy constraints: No emotion detection, no intent inference
• Database: PostgreSQL with Drizzle ORM for server-side persistence

Recent Changes (January 2026)

• Added EDCM core types (shared/edcm-types.ts)
• Implemented focus-locked EDCM engine with heuristic metrics
• Created /api/edcm/analyze endpoint for EDCM analysis
• Added /api/analytics/collect with privacy guards
• Built local analytics storage using IndexedDB
• Added Settings page for cloud sync preferences
• Added Trends page with SVG sparkline charts
• Updated AI Assistant description with accurate capabilities
• Added tutorial button for onboarding
• News Ingest Module: Primary/secondary source classification with distortion scoring
  • Quote extraction with speaker attribution
  • Editorial marker detection
  • Headline spin analysis
  • Distortion score (higher = more distortion)

Key Features

EDCM Metrics

• C (Constraint Strain): Composite pressure indicator
• R (Refusal Density): Resistance patterns
• D (Deflection): Topic avoidance
• N (Noise): Uncertainty signals
• E (Escalation): Intensity patterns

Quality Flags

• LOW_CONFIDENCE_PARSE, INSUFFICIENT_CONTEXT
• MISSING_SPEAKER_TAGS, OVER_SHORT_SAMPLE
• POSSIBLE_INTENT_INFERENCE_RISK, MODEL_DRIFT_RISK

HmmItem Tracking

• Structured uncertainty with severity levels
• Evidence and suggested fixes
• Tags for categorization

Analytics

• Local-first with IndexedDB
• Optional cloud sync (off/metrics_only/include_text)
• Export to JSON
• Trends visualization

Project Structure

client/src/
  pages/
    Analysis.tsx        - AI-assisted EDCM analysis
    Settings.tsx        - Cloud sync preferences
    Trends.tsx          - Analytics visualization
    Home.tsx, Dashboard.tsx
  components/
    analysis/HmmmPanel.tsx
    shared/ModeHeader.tsx, etc.
  edcm/
    engine.ts           - EDCM analysis (client-side)
    audio-adaptor.ts    - Audio→EDCM converter
    types.ts            - Legacy types
  lib/
    analytics.ts        - IndexedDB storage
    edcm-client.ts      - API client + logging

server/
  routes.ts             - All API endpoints
  edcm-assistant.ts     - AI processing + EDCM analysis
  storage.ts            - PostgreSQL operations
  audio-discernment.ts  - Audio processing

shared/
  edcm-types.ts         - Core EDCM types
  edcm-assistant-types.ts
  schema.ts             - Database schema (Drizzle)
  audio-types.ts

API Endpoints

• POST /api/edcm/analyze - Run EDCM analysis
• POST /api/analytics/collect - Privacy-guarded analytics
• POST /api/edcm-assistant/parse - Parse text to turns
• POST /api/edcm-assistant/process - AI assistant streaming
• GET/POST /api/edcm-assistant/artifacts - Artifact CRUD
• POST /api/political/analyze-news - News distortion analysis

Ethical Constraints

• No emotion detection from audio or text
• No intent inference or prediction
• Safety Notes describe "patterns that can precede harm" not "dangerous person detection"
• Consent banner required for audio recording
• Raw metrics never shown by default (projection governance)
• Text upload requires explicit opt-in

User Preferences

• GitHub: erinepshovel-code, repo: EDCM
• Privacy-first approach mandatory
• "hmm" must be first-class UI element, not a comment
• Focus-locked analysis: instrument-only, no advice