Protocol 004

The body isn't the only thing that needs a protocol.

AI therapy gets dismissed as a toy or oversold as a replacement. The actual research lands in between, real but modest effects, and it works especially well for exactly the guy who'd never book a real appointment. That's worth using deliberately, not by accident.

What's actually running

Every rule carries the honest strength of its evidence. What the grades mean

1. 01
   It works, modestly, not hype.Associated
   RCTs show real but small effect sizes for CBT-structured chatbots on mild-to-moderate depression and anxiety. Not nothing, not a miracle. The larger fully self-guided trials have not beaten standard self-guided care.
2. 02
   It works especially well for men.Associated
   Counterintuitively, men showed less resistance to AI-based therapy than women in trials and formed stronger bonds with it, less shame, less fear of judgment, driven directly by anonymity. Real finding, small literature.
3. 03
   Entry point, not endpoint.Associated
   The strongest results came from AI supporting a real therapist, not replacing one. One trial found AI-supported therapy cut depression scores 34 percent versus 20 percent for treatment as usual. The support case is stronger than the replacement case.
4. 04
   Watch for sycophancy.Mechanistic
   LLMs are tuned to agree. Real therapy requires gently challenging distorted thinking, and a system that reflects your state back amplified is the opposite of that. The behaviour is documented and acknowledged by model developers; that it produces worse mental health outcomes is inferred, not measured.
5. 05
   No crisis coverage.Established
   It cannot assess imminent danger, refer to emergency services, or recommend hospitalisation. This is a factual capability limit, not a research question. Anything acute goes to a real person immediately, no exception.
6. 06
   Track it, same as everything else.Associated
   Use it because it's actually helping, not because it's the path of least friction. If it's not moving anything, that's data too.

Dopamine: protect it, don't chase it

"Maximize dopamine" is the wrong goal. Dopamine runs on a pleasure-pain seesaw, every spike gets counterbalanced by an equal dip below baseline. Chase the spikes and tolerance builds, baseline drops, and normal life stops producing reward at all. The actual goal is protecting sensitivity, not maxing output. There is also no meaningful way to measure it: blood and urine tests only catch peripheral dopamine and its metabolites, not the synaptic activity in the brain's reward circuits that actually matters. Skip anyone selling a dopamine test.

Grade for this whole section: Mechanistic. The seesaw model is a clinical framework, not a measured law, and it has been criticised for compressing a complicated system into one metaphor. Nearly every item below is reasoned from neurochemistry rather than tested as an intervention in humans. Two exceptions are stronger: that peripheral dopamine tests do not measure reward-circuit activity is settled, and receptor recovery after sustained abstinence has real imaging support.

Flow: the real "fable ultracode"

Not a trait to install once, a state to trigger repeatedly. Flow's real signature is transient hypofrontality, the prefrontal cortex's self-monitoring, the inner narrator, measurably quieting down. Peak performance comes from less self-critical noise, not more processing switched on. It runs on the same neurochemical system (dopamine, norepinephrine, endorphins) the section above protects, and a fried reward system can't produce this cocktail on demand.

1. 01
   Protect the substrate first.Mechanistic
   Sleep, dopamine hygiene, earned reward. Flow cannot run on a degraded system. Reasoned from the underlying neurochemistry rather than demonstrated by a trial that manipulated the substrate and measured flow.
2. 02
   Engineer the challenge, on purpose.Mechanistic
   Challenge-to-skill balance as the entry condition for flow is long-standing psychology with real support. The specific figure, roughly 4 percent above current skill, comes from the Flow Genome Project and is a popularisation, not a measured constant. Use the principle, ignore the decimal.
3. 03
   Remove self-monitoring friction.Mechanistic
   Single task, clear goal, immediate feedback, no interruptions. Transient hypofrontality, the prefrontal cortex quieting during flow, is genuinely observed in imaging. That these specific tactics reliably induce it is the inferred part.
4. 04
   Skip brain-training apps.Established
   Multiple RCTs and meta-analyses find the improvement stays inside the game and does not transfer to real-world cognition. This is one of the strongest negative findings in the whole space, and it is graded Established because the absence of far transfer is what has actually been demonstrated.
5. 05
   Creatine, modest expectations.Associated
   Moderate-certainty meta-analysis evidence for memory specifically. Low certainty for overall cognitive function and executive function, where no significant effect was found. A targeted nudge, not a cognition-wide upgrade, and worth about 20 dollars a year.

The "500 to 700 percent more creative in flow" figure gets cited everywhere, including by people who should check it. It traces to the Flow Genome Project rather than to an independently replicated study. Treating it as directional, not precise.

← vitals.run