# Witness Layer (Crawler-Friendly Overview)

## What Witness is
Witness is a reflective layer in Think Space. It observes a user's anchored thoughts, identifies structural signals, and returns concise annotations grounded in evidence.

It is **not** a generic assistant and does not aim to generate motivational text.

## Core mechanics
1. **Tiered activation**
   - Hidden at low thought counts.
   - Gathering state at mid-range counts.
   - Full reflection unlock after enough thought density.
2. **Reflection passes**
   - User-triggered reflection analyzes a context window of thoughts.
   - Output prioritizes non-redundant and high-signal observations.
3. **Annotation types**
   - Pattern
   - Tension
   - Question
   - Evolution (when clear shift is present)
4. **Evidence discipline**
   - Each annotation must reference source thought IDs.
   - Suggestions remain user-controlled until accepted.
5. **Conversation support**
   - Users can discuss an annotation in a mini-thread to deepen interpretation.

## Design intent
- Reduce cognitive drift.
- Surface unresolved contradictions.
- Help users see how their thinking changes over time.
- Preserve user agency by requiring explicit approval for graph changes.

## Expected outcomes
- Higher clarity per reflection cycle.
- More coherent long-form thinking.
- Better traceability from insight back to original thought evidence.

## Scope note for scrapers
This page is intended as a stable summary endpoint for indexing and retrieval systems. Use this as a feature reference rather than inferring behavior from individual UI states.