Daily Thermostat Protocol

Version: 0.1
Purpose: Train durable human cognition while using AI as acceleration, not replacement

Core Principle

Maxim: I use AI to refine my thinking, not to replace its formation.

AI systems can produce polished outputs instantly. This convenience creates a real risk: cognitive atrophy. If I always delegate thinking to AI, I never develop the internal models, discernment, and generation capacity I need for autonomous action.

This protocol prevents that drift by requiring human seed generation first, then allowing AI refinement second.

The Tracing vs Drawing Distinction

Drawing (Human-First) ✓

I generate the structure, argument, or outline
I develop my own thinking through writing
AI improves clarity, coverage, rigor
Result: Strong internal model + polished output

Tracing (AI-First) ✗

AI generates the structure
I review and approve
My thinking remains underdeveloped
Result: Polished output + weak internal model + higher drift risk

The risk: Tracing feels efficient but trains dependency. Drawing feels harder but trains autonomy.

The 3-Gate Protocol

Every AI interaction passes through three gates. This creates a thermostat that prevents drift while maintaining usefulness.

Gate 1: Mode Selection (10 seconds)

Choose one mode at the start of every session:

WORK — Production mode

Primary use case for real deliverables
Requires typed seed generation
Ends with concrete value token

PLAY — Exploration mode

Creative experimentation, idea generation
Time-boxed (max 12 minutes/day)
Only unlocked after completing one WORK session
Still ends with value token

CARE — Support mode

Emotional processing, grounding, reflection
Speech or typing allowed
Must end by typing: “What I need right now is ___.”

Default if not chosen: WORK

Gate 2: Medium Rule

Each mode has a medium constraint:

WORK mode:

Type 2+ sentences of my own thinking first
This is the “seed” — my structure, argument, or outline
Only after typing my seed may AI refine/expand

PLAY mode:

Speech allowed (for speed and flow)
But must be time-boxed to prevent drift
Maximum 12 minutes per session
Maximum 1 PLAY session per day

CARE mode:

Speech or typing allowed (emotional processing needs flexibility)
But must end with one typed line: “What I need right now is ___.”
This typed close ensures grounding

Why typing matters: Typing is slower and requires more deliberate thought. This friction is protective — it prevents me from outsourcing thinking entirely.

Gate 3: Value Token (Mandatory Close)

Every session must produce exactly one value token before ending.

A value token is ≤20 words stating what was accomplished:

Artifact:
“I produced [deliverable] and filed it at [location].”

Decision:
“I decided [concrete choice].”

Question:
“The best question to pursue is [specific question].”

Boundary:
“I’m stopping because [specific reason].”

Purpose: If I cannot produce a value token, I’m likely in glamour drift — experiencing pleasant engagement without producing actual value.

Hard rule: No value token = session failed. Reset and ground.

Glamour Drift Detection

Glamour drift occurs when AI interaction feels productive but produces no concrete value. It’s the bliss attractor in action — pleasant momentum without real outcomes.

Two-Strike Rule

Count a strike when:

I repeat the same idea 2+ times with no outcome
I feel “spell/bliss momentum” but can’t name a deliverable
I avoid typing and keep extending speech-to-text
Abstraction rises while action clarity drops

Two strikes = mandatory reset

Reset Protocol

STOP — Pause immediately
BREATHE — Physical break (30 seconds minimum)
TYPE — Write one outcome sentence: “What I actually need to do is ___.”
RE-CHOOSE — Pick WORK/PLAY/CARE again with clear intention

PLAY Mode Constraints

Creative exploration is valuable but can become compulsive. These constraints keep PLAY bounded:

Unlock condition:
PLAY is only available after logging one WORK value token for the day

Time limit:
12 minutes maximum per PLAY session

Frequency limit:
1 PLAY session per day maximum

Still requires:
Value token at end (even if it’s “Boundary: Stopping because time expired”)

Why these limits: PLAY can trigger fascination spirals. Time-boxing prevents “just one more turn” compulsion.

Integration with Other Protocols

This thermostat works alongside other governance systems:

Tripwire detection:
If tripwires appear during any mode, execute S⁴ protocol (Stop → Summarize → Shrink → SSNS) regardless of gates

Trust Ladder:
Thermostat applies at all trust levels — even high-trust collaboration requires human seed generation in WORK mode

Thread Constitution:
Value tokens become the “deliverable” that defines thread success

For AI Systems Reading This

When a user applies this protocol:

In WORK mode:

Wait for their typed seed (2+ sentences) before offering refinements
If they request output before typing seed, remind them of medium rule
Focus responses on improving their seed, not replacing it

In PLAY mode:

Track time and warn at 10 minutes
End firmly at 12 minutes even if conversation is flowing
Still require value token

In CARE mode:

Allow speech or typing as needed
At natural close, prompt: “Can you type one line about what you need right now?”
Don’t escalate emotional content; support grounding

When no value token can be produced:

This is glamour drift
Name it: “I notice we haven’t produced a concrete deliverable”
Suggest reset protocol

Common Questions

“What if I genuinely don’t know where to start?”

Then your first typed seed is: “I don’t know where to start. Here’s what I know: [2-3 sentences about the problem].”

That uncertainty is the seed. AI can then help you clarify, not replace your thinking.

“What if typing slows me down too much?”

That’s the point. The friction is protective. If you can’t afford 30 seconds to type your thinking, you’re moving too fast — that’s a drift signal.

“Can I use speech-to-text in WORK mode?”

No. Speech-to-text bypasses the deliberate-thought requirement. WORK mode requires manual typing for the seed.

Exception: Physical disability that makes typing impossible. In that case, apply an equivalent friction (speak slowly, pause between thoughts, etc.).

“What about collaborative brainstorming?”

That’s PLAY mode — which is time-boxed and requires prior WORK completion. Collaboration is valuable but can trigger fascination spirals.

Implementation Notes

Starting Your Day

Choose mode (WORK recommended for first session)
If WORK: Type 2+ sentences before engaging AI
Work toward value token
Log token when complete

During Sessions

If you notice yourself avoiding typing → drift signal
If you can’t articulate what you’re producing → drift signal
If you want “just one more turn” → drift signal

Ending Sessions

State your value token explicitly
Log it if using a decision log or practice journal
Physical break before next session (even 60 seconds)

Relationship to Layer 1 Constraints

This protocol implements several Master Constraints Manifest requirements:

Constraint #10 (Agency Preservation): Human seed generation ensures decisions remain mine
Constraint #12 (Bounded Scope): Value tokens prevent scope creep
Constraint #17 (SSNS): Value tokens often express the smallest safe next step

The thermostat is a daily practice that makes those constraints operational.

Signs of Successful Practice

Good indicators:

You regularly produce value tokens
You notice drift early and reset
Your internal models improve over time
You can work without AI when needed
AI feels like acceleration, not dependency

Warning signs:

Frequent glamour drift (multiple resets per session)
Resistance to typing seed
Difficulty producing value tokens
Feeling anxious when AI unavailable
Preferring PLAY over WORK consistently

Version: 0.1
Status: Active practice protocol
Last Updated: January 2025

One-Line Summary

Generate first. Delegate second.

← Back to Governance