Thinking Hash System
- Entity ID:
ent-20260419-b5fa0a2036af - Type:
concept - Scope:
shared - Status:
active - Aliases: thinking redaction, thinking trace hashes
Description
Server-side reasoning-preservation mechanism: Anthropic computes a hash of the true thinking trace, stores the trace in a database, and returns only the hash to the client. On subsequent turns the client echoes the hash; the server looks up and re-injects the real reasoning for LLM continuation. Consequence: local JSONL transcripts contain only hashes, so any analysis of local transcripts systematically undercounts thinking depth. Zero-hash turns indicate genuinely zero-reasoning regardless of effort setting.
Key claims
- Local JSONL transcripts undercount thinking depth by design
Relations
- Thinking Hash System --[informed_by]--> Andrey Kolkov
- /thinkback Command Family --[depends_on]--> Thinking Hash System