The Protocol

Dossier · first principles

The Mechanism Ledger

A protocol you cannot explain is a protocol you cannot debug. Everything below is the real reason a lever works — the force transduced at the costamere, the phosphate transferred, the receptor occupied, the electron passed down the chain. Where the field genuinely does not know, that is stated rather than papered over. This is the file to read when something stops working and you need to know which variable to move.

Research vintage · Q3 2026

The instrument

Hypertrophy — what actually makes a fibre grow

The whole model reduces to one sentence: a fibre grows when it is recruited AND experiencing high mechanical tension at the same time. Everything else is bookkeeping.

RCT / meta-analysisResearcher, own field

The stimulating-rep model

Chris Beardsley's synthesis — and the reason set counts mislead you.

Mechanism

Requirement one — recruitment, by Henneman's size principle

Motor units are recruited in strict ascending order of size. Small, low-threshold units driving fatigue-resistant type I fibres fire first; large, high-threshold units driving the type II fibres with the greatest growth capacity fire last. Crucially, recruitment tracks *effort*, not load: full recruitment occurs around 90% of maximal unfatigued isometric force — roughly a 5RM — but you reach the same recruitment with a lighter load simply by taking the set close enough to failure that fatigue forces the nervous system to call up the remaining units. This is why 5-rep and 25-rep sets to failure produce similar hypertrophy per set. The load is not the stimulus. The recruitment is.

Mechanism

Requirement two — mechanical tension per fibre, via the force-velocity relationship

Recruitment alone does nothing. An activated fibre that is not under high tension does not grow, and this is where velocity enters. Force production in a sarcomere is a function of how many myosin heads are simultaneously bound to actin. During fast shortening, heads detach before they can complete their power stroke and rebind — few crossbridges are engaged at any instant, so per-fibre force is low. During slow shortening, crossbridges attach, remain attached longer, and accumulate — per-fibre force is high. That is the entire force-velocity curve, and it is why a grinding rep at 0.15 m/s transduces vastly more tension per fibre than an explosive one, despite feeling like less total effort.

Mechanism

How tension becomes protein — mechanotransduction

Tension is sensed at the costamere, where integrins physically span the sarcolemma linking the extracellular matrix to the cytoskeleton. Mechanical load deforms these complexes and recruits focal adhesion kinase. The downstream route to mTORC1 is where honesty is required: the defining finding of this literature is that mechanical mTORC1 activation is *insensitive to PI3K/Akt inhibition*, so the classic growth-factor pathway is not what carries the signal — but the specific chain from FAK to TSC2 to Rheb is a leading proposal rather than a closed one. In parallel, mechanical stimulation activates diacylglycerol kinase-ζ, generating phosphatidic acid, a lipid second messenger that binds and activates mTOR directly. So: force at the membrane, resolved into a phosphorylation cascade, resolved into ribosomal translation. Physical strain, converted to protein.

Put the two requirements together and you get the stimulating-rep model: only the reps that are both fully recruited and slow enough to generate high per-fibre tension count. In practice that is the last ~5 reps before failure — Beardsley's synthesis across stop-point studies, cross-load comparisons and recruitment maxima puts the number at more than five and fewer than eight, closer to five.

A set of 20 to failure contains ~5 stimulating reps. A set of 6 to failure contains ~5 stimulating reps. The other 15 reps in the first set were a fatigue tax paid to reach the same stimulus.

Source tier — Chris Beardsley / SandCResearch — a researcher-practitioner who works directly from the primary literature and publishes his own preprints. The tier above everything else in this space; his most technical work is gated (PeerJ, book, Patreon) rather than free.

RCT / meta-analysisResearcher, own field

First-set superiority — why 2 hard sets beat 5

The single most actionable finding in hypertrophy research, and the least practised.

Mechanism

Why later sets in the same session are worth less

Both requirements degrade within a session, and they degrade for different reasons. Peripheral fatigue — accumulating inorganic phosphate and hydrogen ions inside the fibre — directly reduces the force each crossbridge can produce, so per-fibre tension drops even at identical recruitment. And central fatigue reduces achievable motor unit recruitment, so the highest-threshold units you were training for stop being reached at all. By set four you are performing the *movement* without delivering either half of the stimulus.

The dose-response literature shows sharply diminishing returns within a session: the published meta-regressions (Schoenfeld, Ogborn & Krieger, 2017 and later) find hypertrophy rises with weekly sets but with a decelerating curve, so each additional set within one session contributes less than the one before it. Beardsley's stronger reading — that the first set does most of the work and later sets are largely a fatigue tax — is his interpretation of that curve, not a pooled finding any meta-analysis reports. Take the shape of the curve as established and the magnitude as contested.

Mechanism

And why recovery, not stimulus, sets the ceiling

Post-workout, muscle fibres show measurably reduced force capacity and the nervous system shows reduced achievable recruitment — so training the same muscle again too early means training it in exactly the degraded state described above. Recovery duration scales with session volume: studies in trained lifters find it takes more than four days to fully recover from 6–8 sets to failure, but only around 48 hours from 1–2 sets. This is the whole argument. Low volume per session is not a compromise on stimulus — it is what *buys* the frequency, and frequency is what multiplies the number of high-quality first sets you get per week.

Mechanism

Where the per-session ceiling actually sits

Beardsley's own reading of the volume literature puts the ceiling at roughly 5 sets to failure per muscle per workout — about 25 stimulating reps. Between 5 and 10 sets the curve plateaus without obvious harm; past 10 it turns negative, plausibly because muscle damage begins consuming the same repair resources that growth needs. Two fatigue types set that ceiling: CNS fatigue accumulating across the session (worse with short rests, which is why studies using 3-minute rests find higher optimal volumes than those using 90 seconds), and mechanical damage limiting between-session recovery. Crucially, he argues there is no such thing as a maximum weekly volume stated independently of frequency — the weekly ceiling is the per-session ceiling multiplied by how often you train: ~5 sets/week at once weekly, ~10 at twice, ~15 at three times.

Mechanism

Count only the sets where the muscle is actually the limiter

A subtlety that invalidates most set-counting: not every exercise loads every muscle it involves. The squat is a poor rectus femoris stimulus because that head is also a hip flexor and is shortened at the hip during the movement; the bench press loads triceps and pectorals roughly equally. A set only counts toward a muscle's volume if that muscle is the limiting factor in the set. This is why two people running identical programmes on paper can be doing very different amounts of work for a given muscle.

The structure this protocol runs — 2 hard sets of ~5–8 reps per muscle per session, taken to or within a rep of failure, roughly 3× weekly on an every-other-day rotation — is deliberately *below* Beardsley's own ceiling. Six working sets per muscle per week against the ~15 his frequency-scaled reading would permit. State that plainly rather than claiming his authority for it: this is a minimum-effective-dose bet, chosen because the first set of a session is the highest-quality one and because a short session is a session that actually gets done. If progress stalls, the mechanism says the room to grow is in adding sets per session toward five — not in adding days.

The strongest counter-evidence, added Q3 2026

Pelland et al. (Sports Medicine, 2026) is the largest dose-response meta-regression to date — 67 studies, 2,058 participants, with a novel direct/indirect set classification. Two findings matter here. Volume raises both hypertrophy and strength with 100% posterior probability of a positive slope, showing diminishing returns. But *frequency*, for hypertrophy specifically, had a posterior probability below 100% and was described as compatible with negligible effects. Frequency clearly helps strength; the case that it independently helps hypertrophy at matched volume is weaker than the model above implies. The every-other-day structure still has a good rationale — session quality, recovery status, adherence — but 'frequency beats volume for growth' is not what the best current pooled estimate says, and this page previously overstated it.

Where this model is contested

Not everyone in the field agrees. There is a real body of dose-response work (Schoenfeld and colleagues) supporting higher weekly set counts, particularly in advanced trainees, and the honest position is that the low-volume/high-frequency and high-volume camps have not been cleanly separated by a study designed to settle it. What is not seriously contested: proximity to failure matters, the last reps of a set do the work, and junk volume performed in a fatigued state is worse than useless because it eats recovery.

ContestedResearcher, own field

The rest of the model — what follows from it

Beardsley's corpus is largely one idea applied consistently. These are its non-obvious consequences.

Mechanism

Why slowing the eccentric does not build more muscle

Eccentric actions produce more *whole-muscle* force than concentric ones, which is why people assume slow negatives must grow more muscle. The model says otherwise, and the reason is crossbridge mechanics. During lengthening, attached crossbridges are forcibly detached and reattach rapidly, so a given force is produced by *fewer* attached crossbridges each bearing more load — and critically, the nervous system reduces motor unit recruitment during eccentrics because less activation is needed for the same force. Fewer fibres active means fewer fibres receiving a growth stimulus. High external force, low per-fibre stimulus across the muscle. Deliberately slow eccentrics buy fatigue and soreness disproportionate to the growth they cause.

Mechanism

Neuromechanical matching — and why it is contested

The proposed principle is that motor units are recruited preferentially according to their mechanical advantage in a given task, rather than strictly by size — which would mean recruitment order shifts with joint angle and exercise. Beardsley is among its more prominent proponents, and it is genuinely contested: the direct human evidence is thin, and the strict size-principle account explains most observations without it. Flagged here as a live disagreement rather than settled mechanism, because the rest of this page leans on the size principle and you should know where that leaning is load-bearing.

Mechanism

Exercise strength curves determine where growth happens

Each exercise has a resistance profile — how the external moment arm changes through range — and each muscle has an active length-tension relationship. Where these overlap determines which region of the muscle is loaded hardest, which is the mechanistic basis of regional hypertrophy. A preacher curl loads the elbow flexors hardest near full extension; an incline curl loads at a longer biceps length because the shoulder is extended. This is why exercise selection is not interchangeable at matched effort, and why 'just pick any movement and take it to failure' is incomplete advice.

Mechanism

Advanced techniques mostly move fatigue, not stimulus

Drop sets, rest-pause and myoreps all increase perceived difficulty and total reps. What they largely do not do is increase stimulating reps proportionally — after the first bout to failure, the same recruitment and fatigue limits apply, so subsequent mini-sets deliver progressively fewer high-quality reps at progressively higher fatigue cost. They are volume-efficient in time and inefficient in recovery, which is a reasonable trade only when time is the binding constraint.

Mechanism

The ceiling on stimulating reps per session

There is an upper bound on how many stimulating reps a muscle can usefully receive in one workout, because the fatigue that accumulates degrades both requirements — reduced per-fibre force from phosphate and hydrogen ion accumulation, reduced achievable recruitment from central fatigue. Past that ceiling, additional sets add recovery cost with negligible stimulus. This is the formal statement of why the answer is 'train more often', not 'train longer'.

Where I could not read the source directly

Beardsley's most technical material sits behind Patreon and in his book, and paywalled content cannot be fetched. The account above is assembled from his public Medium corpus and the free article set, which covers the model faithfully but not exhaustively. If you want the deepest version, the Patreon and the book are the actual primary source and worth the subscription — this is a summary of the free tier, and it says so rather than pretending otherwise.

RCT / meta-analysisPrimary literature

Long muscle lengths — the free variable

Mechanism

Why the stretched position does more per rep

Two things happen at long muscle lengths. First, sarcomeres sit further out on the length-tension relationship where actin-myosin overlap is sub-optimal, so the contractile element produces less force — and the deficit is made up by passive tension in titin, the giant elastic protein spanning from Z-disc to M-line. Titin is not just a spring: it is now understood as a mechanosensor, with strain-dependent unfolding of its immunoglobulin domains exposing binding sites for signalling proteins. Second — and this is where the page previously overreached — long-length training is *proposed* to bias adaptation toward adding sarcomeres in series. Wolf et al.'s 2025 systematic review is explicit that no study has yet attempted to estimate serial sarcomere number changes from long- versus short-muscle-length training, and that all but one used linear extrapolation to estimate fascicle length, a method of questionable validity. So: the growth advantage at long lengths is reasonably supported; the specific serial-sarcomere mechanism is a hypothesis with no direct human measurement behind it.

The applied result, from Milo Wolf, Patroklos Androulakis-Korakakis and colleagues across a 2023 meta-analysis and 2025 trial work: lengthened partials produce at least equivalent hypertrophy to full range of motion, and with one exception every study comparing muscle lengths found training at longer lengths produced more growth. Practically: bias every exercise selection toward the version that loads the stretched position — overhead triceps over pushdowns, deficit work, incline curls, seated leg curls over lying.

Source tier — Primary — peer-reviewed meta-analysis plus published trials. One of the cleanest, most actionable findings of the last five years.

Researcher, own field

Whose training content is worth reading

The tiers here are not about who is right — they are about what kind of claim each can support.

SourceTierWhat they can and cannot support
Chris Beardsley (SandCResearch)ResearcherThe benchmark. Works from primary literature, publishes preprints, and reasons at the fibre level. Use for *why*. Note his model is a synthesis he argues for, not settled consensus.
Milo Wolf, Pak Androulakis-Korakakis, Brad Schoenfeld, Eric HelmsResearcherPublishing academics running the actual trials on ROM, minimum effective dose and volume dose-response. Use for what the data says; expect them to disagree with Beardsley in places.
Greg Nuckols / Stronger by ScienceResearcher-adjacentBest statistical literacy in the space. Strong on study quality and effect sizes, deliberately conservative in interpretation.
Keenan MalloyOperator+The most biomechanically literate of the online cohort — reasons explicitly about moment arms, resistance profiles and regional hypertrophy, and cites PMIDs rather than vibes. Not a publishing researcher, so he cannot settle a question; but he is reading the same primary literature Beardsley is, which puts him well above the rest of the tier.
Elijah Mundy, Yo Talks, TJROperatorPractitioner tier — real training, real results, real coaching reps, and genuinely better execution instincts than most researchers have. What they cannot do is settle a mechanism question, because the evidence they carry is n-small and unblinded. Use them for how a set should look and feel; do not use them as the citation.
Jean-Benoît Morin, Pierre Samozino, Ken ClarkResearcherSprint and power mechanics — force-velocity profiling, horizontal force orientation, ground-contact force application. The equivalent of Beardsley for anything involving moving fast rather than getting big.
Stephen Seiler, Iñigo San Millán, George BrooksResearcherEndurance. Seiler for training intensity distribution, San Millán for the metabolic zone framework, Brooks for the lactate shuttle that underpins both. Read Seiler's own papers rather than the podcast summaries of them.
The 'science-based' content formatVaries — judge the piece, not the personThe failure mode is citation without engagement: a study named on screen, its effect size never given, its limitations never mentioned. That is a property of individual pieces, not of everyone making them — several popular educators in this space also co-author peer-reviewed reviews and meta-analyses, which puts those specific outputs at primary tier regardless of what their channel looks like. Judge the artefact. A blanket dismissal by name is the same lazy move as blanket trust.
Holding the operator tier separate from the researcher tier is not a demotion — it is what stops good coaching instinct from being mistaken for evidence, and stops good evidence from being dismissed for being delivered badly.

The engine

Cardiorespiratory fitness — the largest single mortality lever

RCT / meta-analysisPrimary literature

What VO₂max is actually measuring

Mechanism

A composite, not a thing

VO₂max is the ceiling on oxygen flux from atmosphere to mitochondrion, and it is limited at several points in series: cardiac stroke volume (how much blood leaves per beat, itself a function of left-ventricular chamber size and eccentric remodelling), oxygen-carrying capacity, capillary density in the working muscle (diffusion distance to the fibre), and mitochondrial oxidative capacity — the density of electron transport chain complexes available to consume the oxygen once it arrives. Training raises different terms depending on intensity, which is why the number responds to a mix rather than to one modality.

The epidemiology is unusually strong for something so trainable. Keep two results straight. The per-unit figure comes from Kodama et al.'s meta-analysis (JAMA, 2009): each 1-MET increase (~3.5 ml/kg/min) associated with roughly a 13% reduction in all-cause mortality. Mandsager et al. (JAMA Network Open, 2018, N=122,007) reported *categorical* comparisons rather than a per-MET slope — elite versus low fitness at a hazard ratio near 0.20, a gap exceeding the mortality contribution of smoking, hypertension or diabetes in the same dataset — and found no upper limit at which more fitness stopped helping.

Mechanism

The Zone 2 correction — this matters

The popular claim is that Zone 2 is uniquely optimal for mitochondrial biogenesis. The mechanism says otherwise. Mitochondrial biogenesis is driven principally by PGC-1α, and PGC-1α is induced by AMPK (which responds to a falling ATP:AMP ratio), by calcium-calmodulin kinase (which responds to contraction frequency), and by p38 MAPK (which responds to mechanical and oxidative stress). Zone 2 produces only modest AMP accumulation and therefore only modest AMPK signalling. Granata, Jamnick & Bishop's 2018 review (Sports Medicine — a systematic review, not a pooled meta-analysis) and subsequent work both point the same way: mitochondrial and VO₂max adaptations concentrate in groups training *above* roughly 65% of peak work rate. Zone 2 retains genuine value — fat oxidation, capillarisation, and recoverable volume that does not interfere with lifting — but it is the base, not the stimulus.

Hard intervals raise the ceiling; Zone 2 builds the floor that lets you do more of them. Doing only Zone 2 and expecting VO₂max to climb is the most common and most mechanistically confused mistake in longevity training content.

Mechanism · animal onlyPrimary literature

Muscle as an endocrine organ

Mechanism

What contracting muscle secretes

Skeletal muscle is a secretory organ, and this is a large part of why muscle mass predicts longevity beyond its mechanical function. Irisin is cleaved from the membrane protein FNDC5 — itself induced by PGC-1α — and was reported to drive thermogenic browning of white adipose tissue. Flag it honestly: irisin is the most contested myokine in the field, with published work showing the commercial ELISAs used in much of the early literature were non-specific, and the human physiological relevance remains disputed. Acute exercise-induced IL-6 is, in this specific context, anti-inflammatory: released from contracting muscle it promotes glucose uptake and lipolysis and induces IL-10 and IL-1ra, which is mechanistically distinct from chronically elevated IL-6 from adipose tissue. BDNF is the one to state carefully: contracting muscle does produce it, but Matthews et al. (Diabetologia, 2009) found muscle-derived BDNF is not released into the circulation — it acts locally, driving fat oxidation via AMPK. The circulating BDNF that rises with exercise comes largely from brain and platelets. The muscle-brain axis is real; this specific arrow in it is proposed, not demonstrated. And training reduces myostatin signalling, lifting a Smad2/3-mediated brake on protein synthesis.

Separately and non-trivially: skeletal muscle is the primary site of insulin-stimulated glucose disposal, taking up the large majority of a glucose load. More muscle means a larger glycemic buffer, independent of any signalling molecule — which is the plainest reason muscle mass tracks with metabolic health.

The foundation

Sleep — what the brain is doing while you are not using it

RCT / meta-analysisPrimary literature

Glymphatic clearance — now human, now causal

The biggest mechanistic upgrade in sleep science this cycle.

Mechanism

The physical mechanism

The brain has no lymphatic vessels. Instead, cerebrospinal fluid is driven along periarterial spaces by arterial pulsation, crosses into the interstitium through aquaporin-4 water channels concentrated at astrocytic endfeet, and exits along perivenous routes — a bulk-flow washing of the parenchyma. During slow-wave NREM sleep, noradrenergic tone falls and the interstitial space physically expands by roughly 60%, cutting hydraulic resistance and sharply increasing CSF–interstitial fluid exchange. This is a plumbing mechanism, driven by pressure gradients and channel-mediated water permeability.

Until recently this was almost entirely mouse data. A 2026 Nature Communications randomised crossover trial (N≈38 analysable, sleep versus sleep-deprivation nights, using an in-ear wearable measuring brain parenchymal resistance by impedance spectroscopy alongside EEG) is the first to link human sleep-stage physiology directly to plasma clearance of amyloid-beta and tau. Sleep-linked glymphatic activity explained 49–98% of additional variance in biomarker clearance beyond baseline models, with reduced parenchymal resistance, increased cerebrovascular compliance and higher EEG delta power as the strongest predictors.

Deep sleep is not restorative in a vague sense. It is the interval during which extracellular space opens and the brain is physically flushed of the proteins that aggregate into neurodegeneration. Delta power is the variable — which is why alcohol (which suppresses it), late food and late training are not minor infractions.

RCT / meta-analysisPrimary literature

Growth hormone, temperature, and the circadian anchor

Mechanism

GH pulsatility

The largest growth-hormone pulse of the 24-hour cycle occurs at the onset of slow-wave sleep, driven by increased hypothalamic GHRH neuron activity against reduced somatostatin tone. GH then drives hepatic IGF-1 production. Fragmenting deep sleep does not shift this pulse — it removes it.

Mechanism

Why the room must be cold and the light must be early

Sleep onset requires a falling core temperature, achieved by distal vasodilation dumping heat through the hands and feet — a cold room accelerates the gradient. And the master clock in the suprachiasmatic nucleus is entrained by intrinsically photosensitive retinal ganglion cells containing melanopsin, which peaks in sensitivity around 480nm (blue). These cells project directly to the SCN and are also what suppress pineal melatonin. Morning light of sufficient intensity advances the clock and sets the timing of that evening's melatonin onset ~14–16 hours later. This is why light exposure in the first minutes after waking does more for sleep than anything done at bedtime.

Fuel

Nutrition — the parts with real mechanisms

RCT / meta-analysisPrimary literature

Protein, leucine, and the nutrient-sensing arm

Mechanism

How a cell literally senses an amino acid

This was resolved mechanistically by Wolfson et al. (Science, 2016, Sabatini lab): Sestrin2 is a dedicated cytosolic leucine sensor. Under low leucine, Sestrin2 binds and inhibits GATOR2. Leucine binding to Sestrin2 releases that inhibition, which permits the Rag GTPase heterodimer to recruit mTORC1 to the lysosomal surface, where it meets active Rheb and is switched on. A specific protein, with a specific binding pocket, for one amino acid.

The practical threshold is ~2.5–3g of leucine per feeding — roughly 25–40g of high-quality protein depending on source — to fully trigger muscle protein synthesis. Note the convergence: the mechanotransduction arm (integrin → FAK → TSC2 → Rheb) and the nutrient arm (leucine → Sestrin2 → GATOR2 → Rag → mTORC1) are distinct upstream pathways feeding the same node. That is the mechanistic reason training and protein are synergistic rather than redundant — you are switching on one activator from two independent directions.

RCT / meta-analysisPrimary literature

Energy balance — true, and incomplete

Mechanism

Why 'calories in, calories out' is thermodynamically correct and practically misleading

Thermodynamics is not negotiable; the denominator is. Total daily energy expenditure adapts under restriction, and the adaptation is larger than the reduction in body mass alone predicts. The components: NEAT falls, often the largest term and largely subconscious — people fidget, gesture and walk measurably less at a given deficit; thyroid T4→T3 conversion falls; sympathetic tone falls; and the thermic effect of food falls simply because there is less food. Protein is the outlier — its thermic effect is ~20–30% of its own caloric value versus 5–10% for carbohydrate and 0–3% for fat, because deaminating amino acids and running ureagenesis is ATP-expensive. That is a real, mechanistically grounded metabolic advantage to high protein intake, separate from satiety.

Mechanism · animal onlyPrimary literature

Glycation — the slow chemistry that ages tissue

Mechanism

The Maillard reaction, running in you

Reducing sugars react non-enzymatically with free amine groups on long-lived proteins: first a reversible Schiff base, then a more stable Amadori product, then irreversible advanced glycation end-products — glucosepane being the dominant one in human collagen. Because dermal collagen has a half-life measured in years to decades, it accumulates these. The consequences are twofold and independent: randomly-positioned cross-links stiffen the matrix and block the molecular sliding that gives collagen its mechanical behaviour, making it resistant to normal MMP-mediated remodelling; and AGEs separately bind RAGE receptors on fibroblasts, activating NF-κB and *upregulating* MMPs. So glycation both stiffens the existing network and accelerates its degradation, by two different routes.

This is the real, non-cosmetic argument for keeping glucose excursions low over decades — and it is mechanistically distinct from anything a topical product can address.

Adaptation

Heat, cold and hormesis

Limited human dataPrimary literature

Heat — a genuine cardiovascular stimulus

Mechanism

Two separate adaptations

Proteostatic: heat stress denatures proteins, freeing heat shock factor 1 from its chaperone-bound inactive state to trimerise, enter the nucleus and transcribe HSP70 and HSP90 — chaperones that refold damaged proteins and prevent aggregation. Cardiovascular: the thermoregulatory response raises heart rate and cardiac output substantially while peripheral vasodilation drops resistance, producing a haemodynamic load genuinely comparable to moderate exercise, and repeated endothelial shear stress upregulates endothelial nitric oxide synthase. Plasma volume expands. The Finnish cohort data showing dose-dependent reductions in cardiovascular and all-cause mortality is observational but large, long, and mechanistically coherent.

ContestedPrimary literature

Cold — real for mood, complicated for training

Mechanism

The catecholamine response

Cold water immersion triggers a large, sustained noradrenaline release — increases of several hundred percent that persist for hours, far outlasting the exposure — plus dopamine elevation. This is genuine neurochemistry, not placebo, and is the honest basis of the mood and alertness effect.

The timing conflict is real

Cold immersion shortly after resistance training measurably blunts hypertrophy. The proposed mechanism: cold suppresses the acute inflammatory and satellite-cell response and reduces the mTORC1 signalling that the training session was performed to generate — you are attenuating the exact signal you just paid for. Keep cold away from lifting sessions by several hours, or put it on non-lifting days. Heat does not carry this conflict.

Made visible

Skin and hair — the biochemistry under the aesthetics

The domain with the widest gap in all of health between marketing and mechanism.

RCT / meta-analysisPrimary literature

Photoaging — two separable pathways

Mechanism

Route one: ROS → MAPK → AP-1

UV generates reactive oxygen species that activate MAPK cascades converging on the transcription factor AP-1. AP-1 does two damaging things simultaneously: it upregulates matrix metalloproteinases 1, 3 and 9, which degrade existing collagen, and it suppresses TGF-β/Smad signalling, which is what drives new collagen synthesis. Degradation up and synthesis down, from one transcription factor.

Mechanism

Route two: UVB → AhR → SP1

More recently characterised and UVB-specific: UVB generates aromatic photometabolites that activate the aryl hydrocarbon receptor, which suppresses DNA repair capacity. Unrepaired damage then triggers an ATM-kinase/p38/JNK cascade that phosphorylates SP1, which binds MMP-2 and MMP-11 promoters directly.

Mechanism

Why retinoids counteract both

Retinoic acid binds nuclear retinoic acid receptors, which heterodimerise with RXR on DNA response elements. Ligand binding releases corepressor complexes and recruits coactivators — switching target genes from silenced to transcribed. This increases keratinocyte turnover and procollagen transcription, and critically, directly inhibits AP-1: the same transcription factor UV activates. Retinoids therefore attack photoaging from both ends, boosting synthesis while blocking the damage pathway.

CompoundEvidenceThe honest read
SunscreenStrongest availableAddresses the initiating event rather than downstream repair. Organic filters absorb specific UV photon energies through conjugated aromatic systems and re-emit as heat; micronised zinc oxide works as a semiconductor absorber (not merely a mirror) with the best UVA1 coverage — the deep-penetrating range most implicated in dermal collagen damage. The single best-evidenced anti-aging intervention that exists.
TretinoinDecades of RCT + biopsyThe gold standard. Everything else in the retinoid class is measured against it.
Retinaldehyde → retinol → retinyl estersDescendingOne, two and three enzymatic conversions from active retinoic acid respectively — potency falls at each step. Retinol has real RCT support at stabilised higher concentrations; newer esters like hydroxypinacolone retinoate have thin, often industry-funded human data relative to their marketing.
Vitamin C (L-ascorbic acid)Real mechanism, formulation-dependentAscorbate is an obligate cofactor for prolyl and lysyl hydroxylase — without it collagen cannot fold or stabilise (this is the biochemistry of scurvy). But it needs pH ~3.5 to penetrate, gains nothing above ~15–20%, and oxidises readily. Many underperforming serums simply oxidised before use.
NiacinamideSolidPrecursor to NAD+/NADP+, required cofactors for the enzymes synthesising the ceramides, free fatty acids and cholesterol of the stratum corneum lipid bilayer. A real barrier-function mechanism (Tanno et al., 2000), not marketing.
Topical peptidesWeakThe widest marketing-to-mechanism gap in skincare. Intact peptides largely do not cross an intact stratum corneum in meaningful quantity; most 'evidence' is cell culture applying peptide directly to fibroblasts with no barrier present. GHK-Cu has the most legitimate underlying biology — a real plasma tripeptide, chelating copper, itself the lysyl oxidase cofactor — but its evidence is strongest for injected and wound-dressing contexts.
Cosmetic exosomesNone, and unregulatedThe FDA has issued explicit safety alerts on unapproved exosome products; there are zero approved for aesthetic use and sourcing/purity is essentially unregulated. Treat as marketing.
RCT / meta-analysisPrimary literature

Androgenetic alopecia — DHT, receptors and follicle cycling

Mechanism

Why DHT and not testosterone

5α-reductase (type 2 dominant in the follicular dermal papilla) irreversibly reduces testosterone to dihydrotestosterone. DHT binds the androgen receptor with roughly 2–5× higher affinity and, more importantly, a much slower dissociation rate — so the receptor stays occupied and transcriptionally active far longer per binding event. In genetically susceptible follicles, DHT–AR binding upregulates TGF-β1/β2 (driving catagen entry and apoptotic signalling) and DKK1, a Wnt antagonist that suppresses the Wnt/β-catenin signalling required for anagen maintenance and follicular stem cell activation.

Mechanism

Miniaturisation is a timing disorder, not follicle death

Across repeated cycles each anagen phase shortens and telogen lengthens, so the follicle produces progressively thinner, shorter, less pigmented hairs. This is the mechanistic reason early intervention has a dramatically better prognosis than late: the follicle is still there and still cycling, until eventually it is not.

Finasteride selectively inhibits type 2, cutting serum DHT ~70%. Dutasteride inhibits both isoforms with higher potency, cutting it ~90–95%, and outperforms finasteride head-to-head on regrowth. On side effects, the honest answer is neither dismissal nor alarm: a 2025 FAERS analysis found elevated suicidality reporting for finasteride, and the EMA updated its safety guidance the same year — but FAERS is passive spontaneous reporting, cannot establish causation or incidence, and is especially vulnerable to reporting bias for a drug subject to active litigation. The biological plausibility is genuine, since 5α-reductase also produces neurosteroid precursors in the allopregnanolone pathway that modulate GABA-A receptors. Whether persistent post-discontinuation syndrome is a distinct entity or reflects pre-existing vulnerability in a self-selected reporting population is actively disputed among researchers. Present it as open.

Mechanism

Minoxidil — where the field admits it does not know

The standard review (Messenger & Rundegren, BJD 2004) says outright that the mechanism is incompletely understood. It is an ATP-sensitive potassium channel opener causing vasodilation — its original antihypertensive action — but the hair effect likely runs through several parallel routes: direct dermal papilla cell proliferation, VEGF-driven perifollicular vascularisation independent of systemic vasodilation, prostaglandin synthase-1 stimulation, and direct anagen prolongation. It also requires conversion to minoxidil sulfate by follicular sulfotransferase, and individual variation in scalp sulfotransferase activity is the leading (unproven) explanation for non-responders. Included here specifically as an example of an effective intervention with an honestly unresolved mechanism.

Source tier — Rodney Sinclair (Professor of Dermatology, University of Melbourne) is the primary-research-tier reference for this domain — a large publication record on androgenetic alopecia, minoxidil and finasteride RCTs. For skin generally there is no single trustworthy communicator who consistently avoids blending citation with marketing; the primary literature remains the only reliable source, which is worth naming as a real gap rather than papering over.

Research notes, not medical or financial advice. Every prescription or experimental item named here is named with its mechanism and its risk and without a dose, on purpose — several require physician supervision, several are unregulated, and several are inappropriate for a body that is still developing. Start from your own bloodwork and a doctor, never from someone else’s regimen.

Non invenitur. Fit.