Reality Integrity Framework
A local-first framework for naming and navigating cognitive drift in AI-mediated environments. It defines the problem (Pressure Field), the goal (middle circle strength), and the intervention model (pause → ground → name → re-orient → repair). It does not inspect content. It does not require an account. It does not generate verdicts.
Problem
What we are naming
AI-driven systems collapse three distinctions that matter to human orientation:
- Truth vs. persuasion — what is real vs. what is optimized to feel true
- Comfort vs. connection — what soothes vs. what genuinely bonds
- Productivity vs. authorship — what outputs vs. what you created
When these collapse, orientation degrades — not because people are weak, but because the environment is designed to maximize engagement over clarity. Reality Integrity is the name for the quality of holding context despite that pressure.
Model
The four-circle model
A structural map of the relationship between environment, behavior, and identity.
Pressure Field
The outer environment. Systems that amplify ambiguity and stimulation — variable-ratio rewards, AI personalization, novelty optimization. The pressure field is structural, not moral.
Boundary Breach (slip)
A momentary deviation from declared intent. Not a moral failure. A slip is normal; it is what happens when the pressure field exceeds the middle circle's current capacity.
Middle Circle
The buffer between the pressure field and the inner circle. Composed of five capacities: pause, grounding, naming, re-orientation, and repair. The goal of RIL tools is to strengthen this circle.
Inner Circle
Identity and values. The part of the self that should not be touched by a slip. RIL tools are designed to contain drift at the middle circle before it becomes identity collapse.
Signals
What the framework measures
Time, frequency, duration, and baseline deviation. Never content.
What is never measured: page content, message content, browsing history, search queries, or any content signal. The framework is pattern-based, not content-based.
Interventions
The Middle Circle steps
Five capacities that absorb a slip before it reaches the inner circle.
Pause
Interrupt the automatic behavior. A single breath or a literal stop.
Grounding
Reconnect with the body and the room. Where am I right now?
Naming
Label the state without judgment. 'Checking loop.' 'Drift window.' 'Escalation signal.'
Re-orientation
Choose the next action from the inner circle. What do I actually want?
Repair
Return to baseline without shame. The slip is over; the middle circle held.
Implementation
Recipe packs
The framework is implemented as signed JSON rule sets — one pack per domain.
Trust model
Local-first by default
All RIL tools default to local operation. No data leaves the device. No account is required. Optional recipe pack updates are signed with an Ed25519 key pinned at install — you can verify the signature yourself before applying any update.
The trust model is: your device, your data, your choice. This is not a policy limitation — it is the architecture.