The protocol

Everything below follows from three findings, each established by more than one source:

1. Load beats modality. Exercise training measurably rebuilds mitochondria and aerobic capacity in humans, and the size of the change tracks training load (volume x intensity), not the type of training (Molmen 2025, T1). Interval and continuous training are statistically indistinguishable on the molecular surrogate (g = 1.29 vs 1.01, p > 0.05; Abrego-Guandique 2025, T2). Since load is what accumulates and adherence is what accumulates load, the right modality is the one you will still be doing in a year.

1. Timing beats amount. Bright light in the morning shifts sleep earlier and improves it; the same light in the evening shifts sleep later and worsens it (Dautovich 2019, T1). The shape of every light decision is more light early, less late, not less light.

1. Energy and glucose beat diet branding. Across five independent literatures, diet effects keep tracing back to total energy intake, weight change and glucose, not to any named diet's special mechanism (synthesis). The levers on that pathway with the best evidence and the fewest costs are training, light-timed sleep and total intake, which is what this page is built from.

On waking: go toward bright light

Get bright light early, ideally outdoors. Above roughly 1000 lux in the morning is associated with an earlier sleep phase and better self-reported sleep (Dautovich 2019, T1, 45 studies). Morning is the one point in the day where more light helps on both axes: it improves the coming night and costs nothing now.

Training: pick what you will keep doing, then progress the load

• Choose modality on preference and adherence. Interval vs continuous is statistically a tie on the measured surrogate (Abrego-Guandique 2025, T2); load predicts the outcome (Molmen 2025, T1).
• Progress volume x intensity gradually. Load is the dose variable the evidence can currently specify; an optimal prescription beyond that is an open question the field itself acknowledges (Bishop 2025, T4, cited as the open question, not as evidence).
• If you are unfit, expect the largest gains. The magnitude of change in mitochondrial content, capillarization and VO2max is largely determined by initial fitness, with greater changes in those starting lower (Molmen 2025, T1).
• Age, sex and existing disease are not reasons to skip it. The ability to adapt is maintained throughout life, irrespective of sex and presence of disease (Molmen 2025, T1).

Eating: control energy and glucose, not ketones

• A Mediterranean pattern is the best-ranked default for glycaemic control (SUCRA 88.15% vs 55.7% for low-carbohydrate; Yuan 2024, T2). Low-carbohydrate ranks first for body composition, with differences the authors call "mostly small and often trivial".
• If reducing carbohydrate, know the shape of the benefit: real, modest and front-loaded. HbA1c falls about 0.29 percentage points, most at 3 months, attenuating after (Mongkolsucharitkul 2025, T1, 27 RCTs, n = 2,870). Beyond 12 months the glycaemic and weight advantage is gone while the lipid benefit persists, HDL up and triglycerides down (Ichikawa 2024, T2).
• Do not chase ketosis for metabolic health. Full ketogenic restriction adds nothing measurable for glycaemia or weight over control diets (Choy 2023, T2), and across 27 randomised trials not one arm sustained carbohydrate below 10% of energy (Mongkolsucharitkul 2025, T1).

"No arm sustained carbohydrate intake < 10% of total energy; therefore, findings generalize to low-to-moderate carbohydrate rather than very-low-carbohydrate."

The last 3 to 4 hours before bed: take the room down

• Ordinary room light is already the problem. Under 200 lux in the hours before bed delayed melatonin onset in 99% of 116 people and cut melatonin duration by about 90 minutes (Gooley 2011, T1). Dim the room, not just the phone.
• If you act on one display number, it is melanopic EDI, not lux, not colour temperature, not a "night mode" label. Melanopic irradiance, isolated from brightness and colour, is what governs melatonin and sleep latency, and displays can be tuned without visibly changing (Schollhorn 2023, T1).
• Do not trust how it feels. Screen conditions that suppressed melatonin and delayed the clock were rated identical in comfort by the people affected (Cajochen 2011, T2; Chang 2015, T1). Handle evening light by fixed rules, not perception.
• Know the tradeoff you are making. Evening screen light acutely improves alertness and memory while damaging circadian timing and next-morning alertness (Cajochen 2011, T2; Chang 2015, T1). Late-evening screen work trades tonight's sharpness against tomorrow's.
• Blue-blocking glasses are a narrow tool. The supported outcome is falling asleep faster, not sleep quality, duration or mood (Hester 2021, T2, 16 RCTs, 453 patients). Reasonable to trial; set the expectation there.

Lights out: darkness is a real variable

As little as 5 to 10 lux with the eyes closed still produces a circadian response (Tahkamo 2019, T1), so bedroom darkness is worth engineering, not assuming. The same review notes the system is forgiving: melatonin recovers within about 15 minutes of the light stopping. A brief midnight exposure is not cumulative damage; a lit bedroom all night is a nightly intervention in the wrong direction.

Almost every endpoint above is a biomarker or an intermediate outcome, not a disease or mortality endpoint. The two exceptions that carry through to felt, next-day function are next-morning alertness (Chang 2015, T1) and sleep onset latency (Hester 2021, T2).

The library's evidence argues against spending effort or hope here. The full reasoning is in the synthesis and topic files.

• Ketogenic diet to treat cancer. No randomised trial shows a survival benefit; the trial built to test it found none (Voss 2020, ERGO2). The reviews claiming otherwise rest on historical controls.
• The Glucose Ketone Index as a target. Rarely reported, targets seldom reached, in the trials that would validate it.
• Antiparasitic drugs for cancer. Hundreds of preclinical papers, a handful of trials, no established efficacy.
• Lowering GDF15. It is a strong predictor of later disease, and Mendelian randomization in the same study did not support a causal effect (Guo 2026, T2). Lowering the marker treats the gauge, not the engine.
• Thyroid combination therapy as a transformation. Blinded trials and three independent meta-analyses show no group-level benefit of T4+T3 or desiccated thyroid over T4 alone (Millan-Alanis 2021, T1); no trial measured appearance, and the largest objective physical difference on record is ~1.4 kg (Hoang 2013, T2). Persistent symptoms on T4 are real and belong with a clinician, not with before/after photos.
• Anything sold on a framework paper. T4 papers contain no new data. They are reasons to run studies, not reasons to change behaviour.

Three uses of dietary therapy have real evidence but belong to clinical care, not to a daily protocol: ketogenic diets in drug-resistant childhood epilepsy (large randomised effect, standard care; Devi 2023, T1, with tolerability caveats in Martin-McGill 2020, T1), ketogenic diets alongside cancer treatment for symptoms and body composition, not survival (Zhang 2025, T2), and carbohydrate restriction in PCOS, where effects are inseparable from weight loss (Tosatti 2026, T2; Turetta 2025, T2). Each is a decision to make with a clinician.

Stated so this page reads as exactly as complete as it is.

• A sleep prescription. Sleep is named as a primary lever by the circadian literature itself, and not a single sleep-intervention paper is filed. The light rules above are the closest supported proxy.
• A stress or mind lever. The mitochondria psychobiology topic is entirely descriptive: psychological state tracks mitochondrial biology, but nothing filed tests an intervention. Exercise is currently the only lever with evidence in that direction.
• Resistance training. Everything filed is endurance or interval work; the strength-training literature is simply not in the library yet.
• A minimum effective dose, or detraining data. How little maintains the gains, and how fast they fade, are unaddressed.
• Long-term diet data. Every dietary synthesis filed runs weeks to months.
• A harms ledger. Adverse events sit inside individual reviews; nothing filed assembles them. The clearest signal: ketogenic diets caused discontinuation at several times the rate of usual care in childhood epilepsy trials (Devi 2023).
• You. Every number here is a population average from general-population studies. What it does for one person is measured, not assumed: the levers above pair naturally with a small set of home measurements (sleep timing, morning light exposure, training load, weight, and for the metabolic lever a periodic HbA1c and lipid panel through a clinician).