Back to corpus
research noteexperiment writeup candidatescore 18

Cognitive Metrics Specification

```python def divergence_rate(embeddings: list[np.ndarray], window: int = 5) -> float: """ Compute average cosine distance between consecutive prompt embeddings over a sliding window.

Open full HTMLDownload HTMLCorpus JSONPaper schema

Full HTML reader

Read the full artifact

Open in new tab

Extracted abstract or opening context

Technical definitions for extracting cognitive analytics from AI interaction data. ### Primary (Mohamed - North Star) - `claude_prompts` table in Supabase (112K+ turns) - `memory_turns` table (332K rows, 768-dim embeddings via text-embedding-3-small) - prompt-logger JSONL archives - RAG++ vector search for semantic clustering ### Secondary (Future Users) - ChatGPT export (JSON: `conversations[].mapping[].message`) - Claude.ai export (when available) - Gemini export (JSON format TBD) - Any OpenAI-compatible API logs **Definition**: Semantic distance between consecutive prompts within and across sessions. **Visualization**: Line chart over time. Color gradient from blue (focused) to red (divergent). Overlay session boundaries.

Promotion decision

What has to happen next

Attach run IDs, datasets, metrics, and reproduction commands.

Why this is not always a full paper yet

Corpus pages are public-safe readers for discovered workspace artifacts. They are not automatically final papers. A corpus item becomes a polished paper only after the editable source, evidence checkpoints, references, figures, render path, and release status are attached through the paper schema.