Protocol

At a glance

Philosophy

I treat health optimization as a portfolio problem under uncertainty. Each intervention has an expected benefit (measured in QALYs), a cost, and uncertainty about whether the observed effect is causal or confounded.

The framework: estimate each intervention's hazard ratio from literature, shrink it 30% toward null to correct for publication bias, apply a Bayesian confounding prior to discount observational evidence, simulate 50,000 Monte Carlo samples through CDC life tables, then greedily select the highest expected-value stack under a budget.

Key parameters: willingness-to-pay of $200K per QALY, 40-year horizon, diminishing returns (each additional item gets 95% of the prior item's benefit, floor at 80%), and a complexity penalty beyond 3 items (0.0005 QALYs/yr per extra item). This penalizes sprawling regimens while still allowing justified expansion.

I am not a doctor. I have no medical training. This is expected value reasoning applied to personal health decisions, not clinical guidance. Most of these effect sizes are small, uncertain, and could be entirely confounded.

Core stack

The top 15 highest-conviction items I take daily, ranked by expected value.

Daily schedule

How the stack breaks down by time of day. Generated from a private health database I maintain on GitHub. Bundles contain multiple catalog items.

Exercise

Daily morning workout, typically strength and cardio combined, following Bryan Johnson's workout protocol. Consistency matters more than intensity — the goal is a sustainable daily habit, not periodic heroic sessions.

Full catalog

All 59 items evaluated, organized by category. Items marked "bundled" are included in Blueprint products below.

Blueprint bundles

I use Blueprint products for items where bundling is cheaper than buying individually. Each bundle contains multiple catalog items.

Portfolio optimization

The optimizer greedily adds items by marginal net value (QALY gain × $200K − cost), applying diminishing returns and a complexity penalty. It stops when the next item would have negative marginal value.

Result: 30 items selected, +1587 expected days (4.3 years), $2,907/yr.

Sleep optimization

Sleep is the highest-leverage health behavior. I track it obsessively with two devices and have made several environmental interventions.

Devices

• Eight Sleep Pod 5 — bed temperature control, sleep scoring, HRV, respiratory rate, snoring detection, sleep stages
• Whoop MG — recovery score, strain, HRV, RHR, SPO2, sleep performance

Environmental interventions

• Hamvay-Lang PureDelight 800fp winter duvet ($885) — Bayesian analysis showed bed temperature has a +3.0 score/°C causal effect on Eight Sleep sleep score. The duvet raises bed temperature, improving sleep quality. Estimated 165-day payback period.
• Bed incline — Eight Sleep adjustable base, slight incline for reflux and breathing
• Nasal strips + Nasacort — first night with this combination: zero snoring (vs. 13.2 min/night average). The clearest win in the protocol.

Medications

• Trazodone 50mg — sleep maintenance. Whoop data confirms it wears off at 4-5 hours, leading to second-half fragmentation.
• Melatonin 300mcg — low-dose for sleep onset. Higher doses are not more effective.

Key findings

• Sleep duration is the #1 lever: +8.6 Whoop recovery points per hour
• Bedtime before 11pm yields a 13-point recovery advantage
• SPO2 drops below 95% on 36% of nights (flagged for apnea investigation)
• Disturbances trending up (+0.009/day, p<0.0001)

Tracking

I maintain a SQLite database of nightly health data with 12+ tables, synced from Eight Sleep and Whoop APIs. Metrics that actually influence decisions:

• Sleep score (Eight Sleep) — primary outcome for environmental experiments
• Recovery (Whoop) — composite of HRV, RHR, sleep quality; guides training intensity
• HRV — parasympathetic tone; trending metric for stress and recovery
• Snoring duration (Eight Sleep) — proxy for airway obstruction; guides nasal interventions
• SPO2 (Whoop) — blood oxygen; apnea screening metric

Most other metrics (strain, respiratory rate, sleep stages) I log but rarely act on directly.

What I'm investigating

• Sleep apnea — Bayesian estimate: 66% probability based on SPO2 patterns, snoring, and family history. Pending home sleep study. Treatment options: CPAP, Inspire nerve stimulator, Dayvigo (lemborexant).
• Rapamycin — strongest longevity candidate in the Rx list (ITP mice +26% lifespan). Requires physician willing to prescribe off-label. Immunosuppression concern.
• Semaglutide — SELECT trial HR 0.80 for MACE. High cost ($6K/yr) but enormous effect size. Watching for generic timeline.
• Lion's mane, cistanche — recently added to stack. Low evidence confidence but low cost.
• Trazodone alternatives — wears off at 4-5h. Evaluating lemborexant (Dayvigo) for better second-half sleep maintenance.

Methodology

This page is generated from the Optiqal Python package. The data pipeline:

1. Literature hazard ratios are entered into a structured catalog with uncertainty estimates and confounding priors
2. Each HR is shrunk 30% toward null (publication bias correction)
3. A Beta-distributed confounding prior discounts the causal fraction (e.g., Beta(2,4) means ~33% of the observed effect is believed causal)
4. 50,000 Monte Carlo simulations propagate uncertainty through CDC life tables for a 39-year-old male
5. Expected days gained = mean QALY gain × 365.25
6. Greedy portfolio optimizer selects items by marginal net value until returns go negative

Evidence confidence labels: high = strong causal prior + tight uncertainty; medium = moderate evidence or wider uncertainty; low = weak causal evidence, speculative.

Profile: 39yo male, BMI normal, never smoker. WTP $200,000/QALY, 40-year horizon, 50,000 simulations.