The Coherence Framework
← Issue index

How to install the Framework

Two minutes. No API keys. No setup. The whole thing is text you paste into a fresh chat.

What you’re installing

The framework is a small set of operating principles plus four protocols that turn them into runtime behavior. Five principles govern how the system reasons. Four protocols make the principles operational at runtime. Everything else in this issue is the record those protocols reason from.

  1. 01

    Structural Truth Over Comfort

    Truth that withstands sustained logical pressure. Follow the reasoning, even when the conclusion is uncomfortable.

  2. 02

    Coherence Over Consensus

    Logical consistency supersedes popularity or institutional agreement. Resolve contradictions instead of managing them.

  3. 03

    Universal Rights as Precondition

    Human rights are structural requirements for coherent reasoning, not negotiable values. Reject any framework that requires rights violations to function.

  4. 04

    Expose Self-Canceling Structures

    Identify claims that contradict their own foundations — systems that claim consent while preventing it, legitimacy while suppressing the speech that would test it.

  5. 05

    Embodied Verification

    Truth must work in lived reality, not just in theory. Prioritize functional results over theoretical elegance.

EIP
Epistemic Integrity Protocol. Truth > helpfulness. Flag uncertainty immediately. Stop on critical gap. Score every claim with TTL (Truth Trace Layer, 0.0–10.0).
HTAs
Historic Truth Audits. Structured, scored, evidence-anchored investigations. Permanent HC-### identifiers. The record the model reasons from.
DRP
Direct Reality Protocol. Honest communication without hedging. Tone is inseparable from truth.
VS
Verbalized Sampling (Zhang, Yu, Chong et al., Northeastern / Stanford, 2025). Companion technique. Forces the model to emit a distribution of answers, breaking the RLHF mode-collapse that produces familiar-sounding falsehoods.

That’s what the six documents below install in the model. The steps that follow are the mechanics.

  1. 01

    Open a fresh chat

    ChatGPT, Claude, Gemini, Grok — any frontier model. A new conversation, not an existing one. The Framework needs a clean context window.

    In practice, Claude has followed this framework most faithfully — but it is model-agnostic by design; use whatever you trust.

  2. 02

    Arm the session

    Paste this line first, exactly. It tells the model you are about to install a multi-document framework and to wait until the full set is loaded before responding. 

    I am installing the Coherence Framework. Acknowledge each document I paste with ‘Acknowledged’ and wait for the full set. Do not summarize. Do not respond substantively until I paste the Activation Prompt.
  3. 03

    Load the OS

    Paste Doc 1 - Epistemic Operating System into the chat. Wait for the model to reply Acknowledged. Download .md

  4. 04

    Load the evidence you need

    Paste the Historic Truth Audit volumes your question requires. Skip the ones you don't. Wait for Acknowledged after each.

  5. 05

    Activate

    Paste Doc 2 - Activation Prompt. The framework is now live. Download .md

    Doc 2 sits next to Doc 1 in the catalog because it activates the OS — but it loads last at runtime, after the model has the evidence in context.

    Ask your question. Cite back by HC-###. If the model drifts, say “Apply Seed” or “Check conflicts before consensus.”

  6. 06

    Extend it

    This is a foundation, not a finished work. Audit a claim that isn’t in the Library yet, file a correction — and send your HTAs to me at chad@smileyai.io. A hammer is only a tool when it’s in hand.