Understated Interventions

Identity restructuring and cognitive augmentation in 2026, after the psychedelic and nootropic maps have been drawn.

Psychedelic-assisted therapy and the standard nootropic catalog are well-explored territory by 2026, and this audit looks past them at what remains underdeployed in the shadow of patent cliffs, specialty silos, and inherited stigma. Off-patent molecules with clean mechanism rarely reach Phase 3 because no sponsor recoups the investment. Findings cross journal boundaries slowly and sometimes not at all, so that a cautionary result in narcolepsy pharmacology never reaches cognitive sleep medicine. Compounds carrying social baggage repel the IRBs that would otherwise run the studies. The resulting underdeployment costs tens of thousands of US lives per year; the regulatory posture producing it is miscalibrated rather than broken. I grade the interventions, put numbers on the ledger, and hold "what to take" and "what to allow" as separate questions. Scott Alexander covered the FDA caution asymmetry in "My IRB Nightmare" and the sponsor-push counter-shape in "Adumbrations of Aducanumab"; Zvi covers Type I bias in institutional behavior on LessWrong.

1. What "understated" means here

An intervention counts as understated here if it has plausible mechanism and usable human data but remains underdeployed for reasons other than the evidence. The usual reasons are some combination of economics (off-patent, no sponsor willing to fund Phase 3, as with metformin in non-diabetics, low-dose lithium, and transdermal estradiol), inherited stigma attached to the molecule or its cultural framing (nicotine, testosterone, ibogaine), and silo effects where a finding sits in one specialty's literature and never reaches the specialty that would use it. The cleanest example of the last category is the narcolepsy-pharmacology result that oxybate-induced slow-wave sleep is functionally dissimilar to physiological SWS; the finding should govern how cognitive sleep medicine thinks about pharmacologic SWS enhancement, and it does not.

The standard psychedelic-assisted therapy menu (ketamine, psilocybin, MDMA) and the standard nootropic catalog (racetams, modafinil, noopept) are outside scope because the map is already drawn there. The question running through the audit is what else produces durable self-concept change or measurable cognitive augmentation, and how much of that is worth doing for a cognitively healthy adult with no metabolic dysfunction. Identity restructuring here means trait-level change sustained for six months and measured on personality inventories rather than symptom scales; cognitive augmentation means working memory, processing speed, or learning rate in healthy adults, measured directly rather than through dementia-risk proxies. Sleep and GLP-1 pharmacology sit beneath both as substrate questions.

2. The regulatory asymmetry

Regulatory frameworks for drug approval and post-market surveillance are designed to track harms from interventions. FAERS, pharmacovigilance, and post-market surveillance systems collect data on adverse events associated with approved drugs. No comparable surveillance system tracks excess mortality or disability associated with the natural history of conditions for which a plausible intervention is denied, delayed, or unfunded. The "safe and effective" standard, as codified, concerns intervention harms and efficacy relative to an implicit baseline of non-intervention; the baseline itself is not separately measured.

The structure of the two error classes differs. When an approved drug later shows harm, the exposed population is identified, the excess events are attributable, and institutional accountability follows. When a drug that would have reduced mortality is not approved or not reimbursed, the counterfactual patients are not identified in clinical or administrative data, and excess events accrue to a distributed baseline rather than to a decision. Thalidomide and Vioxx produced congressional review. The three-year delay on misoprostol approval did not, despite Gieringer's 1985 estimate of roughly eight to fifteen thousand preventable ulcer deaths per year during the bottleneck (Gieringer, RAND). The Gieringer figure is drawn from a policy-advocacy tradition and has been contested by critics offering lower estimates; it is retained here as the most-cited historical estimate of a Type II harm of this class, with the caveat that its magnitude is disputed.

The historical record is consistent with this structure. Thalidomide in 1961 produced the 1962 Kefauver-Harris Amendments and the "safe and effective" standard in its modern form. The amendments addressed the failure mode that had just occurred. Vioxx in 2004 added a second layer of review stringency following cardiovascular safety concerns; plausible excess-death estimates range from roughly 27,000 to 55,000, with Graham's 38,000 figure near the top end. Post-Vioxx prescribing practice deprioritized COX-2 inhibitors even in populations over 65 without prior MI, a subgroup in which observational risk-benefit estimates had shifted in favor of treatment. Excess pain, reduced mobility, and NSAID-induced GI bleeding among patients who were not offered coxibs have not been systematically quantified.

Aducanumab (2021) and lecanemab (2023) illustrate the reverse pattern. FDA granted accelerated approval to aducanumab on biomarker grounds over a 10-1 advisory committee vote against approval, with a failed ENGAGE Phase 3 and efficacy of roughly 24% of placebo decline over 18 months on the favorable reading (PMC10193636). Subsequent prescriber surveys reported that 67% of respondents had reduced confidence in the accelerated pathway. The two sets of decisions are difficult to reconcile on a single implicit tolerance for uncertainty.

Sunstein (2002) observes that a symmetrically applied precautionary principle is self-defeating, since inaction also carries risk (Sunstein 2002). Selectively applied, it operates as a preference ordering over error classes rather than a neutral stance toward uncertainty.

Hansson, Steele, and Stirling give the precautionary principle its stronger formulation as a rule for allocating the burden of proof under deep uncertainty about long-tail harms. Read that way, regulatory conservatism prices in model uncertainty, the cost of a later-failed high-profile approval, and the distinction between a reversible prescription and a durable precedent. Those considerations are real. The question raised here is whether the multiple at which they are currently weighted against the mortality on the no-intervention side is correctly calibrated, and whether that multiple is made explicit in decision documents.

"Regulatory" in this section abbreviates at least six gatekeepers: FDA approval, DEA scheduling, CMS and commercial-payer coverage, specialty-society guidelines (AAN, ACOG, NAMS), prescriber malpractice exposure, and manufacturer economics under Hatch-Waxman. A new indication on a generic confers at most 3-year 505(b)(2) exclusivity, which does not price-protect a commodity. The cases in §11 are rate-limited by different components of this stack and require different fixes.

Chalmers and Glasziou estimated in 2009 that 85% of biomedical research funding is avoidably wasted across design, conduct, and reporting (PMID 19525005). A material fraction is structural: Phase 3 trials don't happen for off-patent drugs because no sponsor recoups the investment, even when the mechanism is established and the Phase 2 signal is clean. Prasad and Cifu's Ending Medical Reversal documents the complementary failure. Interventions become standard of care on weak evidence and persist after disproof; the same apparatus is permissive at the entry point when a sponsor exists and resistant at the exit point when removing the intervention would require admitting a prior mistake.

Off-patent drugs get roughly 30% of the regulatory-authorization success rate of on-patent ones. That isn't because the molecules are worse. Nicotine patches for MCI have a positive 2012 Phase 2 (Newhouse 2012) and remain off-label two decades later because transdermal nicotine is generic and no manufacturer will fund Phase 3 on a commodity patch.

The HRT story compounds the pattern with a social layer. WHI (2002) studied combined CEE+MPA initiated in women averaging 63, a decade postmenopause, and found modestly elevated breast cancer and cardiovascular risk. Prescribing culture summarized this as "hormones are dangerous," producing a 20-year decline that affected perimenopausal women for whom the original trial data didn't apply. The 2024 Medicare reanalysis of estrogen-only therapy in women over 65 (N=10 million, Baik/Baye/McDonald) found a 19% all-cause mortality reduction, with reductions in dementia, breast cancer, lung cancer, colorectal cancer, CHF, and MI (PMID 38595196). The counterfactual (how many excess dementia cases, fractures, and cardiovascular deaths are attributable to the post-WHI prescribing decline) has never appeared in a regulatory decision document.

Every intervention in this document can be read as an instance of the asymmetry. Nicotine for MCI: economic plus social compound. Estradiol in perimenopause: the WHI overcorrection written onto a population the original trial didn't enroll. Low-dose lithium: cheap mineral salt with suicide-prevention epidemiology, no sponsor (and, as LATTICE now shows, no positive RCT on cognition either; see §11 and §12). Ayahuasca NEO-PI data: trapped in ethnobotany because the Western psychiatric pipeline has no receptor for ceremonial use in Peru. Methylene blue: mechanism-solid, human RCTs thin, no one will fund the trial because the compound costs thirty dollars at chemistry suppliers.

The reframe splits the bottom line. What a healthy adult should do for identity and cognitive work in 2026 has a private answer: which interventions are worth the individual time and risk. What society should allow has a different answer, shaped by whether you accept the asymmetry as given or count the deaths on the other side.

3. The withheld-intervention harm ledger

Concrete numbers for the thesis. Each row is a condition where effective interventions exist but are underdeployed, with the best-available epidemiological estimate of the excess mortality or morbidity absorbed by the "no intervention" baseline. These are headline findings from large cohorts and registries that the regulatory and prescribing apparatus has not translated into changed practice.

Caveats on this table matter. Most rows are observational hazard ratios with the usual treatment-selection confounds (healthy-user, treatment-seeker). The 37,000-deaths row is a modeled counterfactual with large assumption sensitivity on uptake, retention, diversion, and substitution; the paper itself reports bounds. The Baik 2024 HRT row is a Medicare claims reanalysis carrying the healthy-user confound that drove the original pre-WHI observational enthusiasm, the same methodological class that produced the error WHI subsequently overturned.

The conditions in this table are confirmation-selected. Each row is a case in which the "withheld intervention" framing has already gained purchase inside mainstream medicine. A symmetric audit would include cases in which the ledger favors the cautious default, including gabapentinoid overprescription, SSRIs in adolescents with an emergent suicidality signal, benzodiazepine dependence cascades, and bariatric overuse. The signal is real and confirmation-selected.

Sources: Swedish PTSD cohort PMC7758737; ADHD mortality Trans Psychiatry 2024; OUD/MAT NBK541393; insomnia/sleep PMC12518473; TRD J Affect Disord 2024; HRT PMID 38595196; X-waiver PMID 36624606.

At global scale, GBD 2021 places mental disorders at 12% of global DALYs and 35% of years lived with disability. Depressive disorders alone contribute 56.3 million YLDs, an increase of 36.5% since 2010 (Lancet Psychiatry 2024). WHO treatment-gap estimates run at 50% untreated in high-income countries and 75-95% in LMIC. Median government spending on mental health is 2% of the health budget.

These numbers define scale rather than prescribe action. Some of the gaps close through expanded prescribing within existing frameworks (MAT, HRT, ADHD stimulants). Others require regulatory changes (Schedule I review, off-patent repurposing pathways). Others require shifts in prescribing culture (nicotine for MCI, psychedelics for existential distress). None appear in the standard calculus, which scores visible actions rather than counterfactual baselines.

4. Identity restructuring beyond the default

Ketamine, psilocybin, and MDMA are outside scope. The question for this section is which remaining interventions produce trait-level personality change sustained at six-month follow-up, measured on validated personality inventories rather than symptom scales.

Pharmacological candidates

Ayahuasca (C+). Frecska 2020 (N=24) measured NEO-PI changes at six months post-ceremony and found neuroticism d=1.45, openness d=2.20 (Frecska 2020). Single-trial, uncontrolled, tourist cohort. An earlier draft cited a 2025 naturalistic replication; I couldn't verify that citation against a public source and removed it. There is no independent replication of the trait-level signal at twelve months. The d=2.20 openness is large enough to trigger a measurement-artifact prior on its own: most effective psychotherapy shifts openness 0.3-0.5 SD. Ceremonial tourism selects hard on openness and expectancy, both of which inflate self-reported NEO-PI scores. A plausible reframe is that ayahuasca opens a months-long window of reduced rigidity during which narrative restructuring happens, and the NEO-PI is catching a sustained state rather than a trait shift; even that reading assumes the effect is real, which an N=24 uncontrolled study cannot establish.

Safety floor. Ayahuasca's active principle is an MAOI-DMT combination. Contraindications: any SSRI, SNRI, tramadol, dextromethorphan, MDMA, bupropion, stimulant, or tyramine-rich diet within the documented washout window (2-6 weeks depending on agent, 5 weeks for fluoxetine). Absolute contraindications: personal or first-degree-relative history of psychosis or bipolar I, cardiac conduction disease, uncontrolled hypertension, any serotonergic agent. "Serious container" in the ranking at the end of this section means MAPS/ICEERS-level medical screening: pre-ceremony psychiatric evaluation, ECG, medication review by a clinician who has read the pharmacology. Deaths during ceremonies are documented, usually from undisclosed cardiac conditions or interactions.

Ayahuasca is among the most-cited candidates for durable personality-layer change outside the standard psychedelic menu, despite an evidence base consisting of a single small uncontrolled study. A properly-powered controlled replication has not been run, at least partly because the data sit at the ethnobotany / trait-psychology interface rather than inside a single specialty. The expectation here is that the effect will shrink substantially under controlled replication. In the interim, the grade reflects mechanism plausibility and the absence of a better-replicated alternative rather than established effect.

Ibogaine (C). Williams/Stanford 2024 (Nature Medicine, N=30 SOF veterans with TBI/PTSD): 88% PTSD reduction and 87% depression reduction at one month, with fMRI theta and cortical complexity changes (Williams et al. 2024). Noller 2017 opioid-dependence cohort (N=14, 8 completed): ASI-Lite and BDI-II reductions sustained to twelve months (Noller 2017). Nobody measured NEO-PI, SRET, or any identity-specific instrument. The framing of ibogaine as "identity restructuring" is user phenomenology projected onto what the trials measured, which is symptom burden. Mechanism (NMDA antagonism, 5-HT2A agonism, BDNF surge, noribogaine sigma-1 activity) is consistent with a large neuroplastic window, but the studies don't demonstrate identity-layer change.

Safety floor. Ibogaine's documented mortality signal is cardiac: Torsades de pointes from QT prolongation, deaths at doses within the "therapeutic" range, frequently in patients with occult hypokalemia, CYP2D6 poor-metabolizer status, or concomitant methadone. The Stanford protocol screened with 12-lead ECG, echocardiogram, electrolytes (K⁺ >4.0, Mg >2.0), CYP2D6 genotyping, continuous telemetry, on-site ACLS capability, and overnight post-dose monitoring. Most Mexico-based ibogaine clinics do not meet that bar. A clinic that cannot produce pre-dose ECG with QTc <450 ms, electrolyte panel, CYP2D6 genotype, at least 24 hours of continuous telemetry, on-site defibrillator and ACLS-trained provider, and documented serotonergic/methadone washout does not meet the screening floor of the cited protocol and should not be treated as equivalent.

The evidence supports use in TBI+PTSD where a clinic meets the Williams protocol floor. Evidence for identity work in a cognitively healthy adult does not exist at present, and the cardiac risk profile is non-trivial.

5-MeO-DMT (C). Beckley/atai BPL-003 Phase 2b (N=193, treatment-resistant depression, 2024-25): MADRS drop of 11-12 points at day 29 with persistence through eight weeks, quadruple-masked dose-finding RCT, methodologically the cleanest 5-MeO trial to date (BPL-003). No identity outcomes measured. A single case report described 75% PCL-5 reduction sustained to twelve months with reported ego dissolution, which is anecdote-grade. The current evidence supports further investigation in TRD. Identity-restructuring claims await trials that measure NEO-PI or SRET alongside symptom scales.

Salvinorin A, nitrous, DMTx (D). Salvinorin A (Mendelson 2011, N=4): dose-dependent acute ego dissolution, no durability data (Mendelson 2011). Nitrous oxide has Nagele's pooled RCTs on 24-hour suicidal-ideation reduction (54% vs 12%) (Nagele), but no identity measures and acute action only. DMTx (Luan 2024, Imperial, N=11) produced acute psychological tolerance with ego remaining intact at high extended-infusion doses (Luan 2024). The Luan finding distinguishes DMTx from the classical-psychedelic dissolution profile and rules it out as an identity-restructuring tool.

Non-pharmacological protocols

MBCT / MBSR (A). Network meta-analysis puts depression-relapse RR at 0.73 (0.54-0.98) with 12-15 month durability (Kuyken 2016). More interesting for identity work: self-referential encoding task (SRET) shows reduced negative self-bias in trials where depression-symptom outcomes are null (PMC10600813), which separates the mechanistic claim from the clinical one. Brewer and Garrison show default-mode deactivation in trait meditators (Brewer 2011). Effect sizes on depression are modest (d=0.13-0.24). Mechanism data for reduction of negative self-referential processing and DMN rumination are stronger than for any non-pharmacological intervention in this section, with the confound that DMN deactivation in meditators may reflect self-selection. The supported intervention is an 8-week MBCT course with a trained teacher; consumer app-based implementations have not been shown to replicate the effect.

IFS (B). PARTS trial 2024 (Harvard-affiliated, N=60 PTSD): d=−0.9 at 24 weeks, with self-compassion and decentering as the hypothesized mechanism (PMID 38934934). Only two of 27 reviewed IFS studies met RCT criteria; the base is thin relative to the clinical enthusiasm. The trial used a waitlist comparator, and waitlist-controlled psychotherapy trials systematically inflate effect size relative to active comparators by roughly 0.4-0.5 SD (Cuijpers 2014). The shrunk estimate against an active comparator is likely closer to d=−0.4 to −0.5. The mechanism framing (parts, decentering) is testable; the evidence base remains thin relative to clinical enthusiasm and active-comparator data is absent.

Coherence Therapy (D). No RCTs; case-series only. Ecker's memory-reconsolidation framing rests on real neuroscience (Nader's reconsolidation literature), but the therapy itself has not been tested under conditions that would establish clinical evidence. Public profile and the "reconsolidation-based therapy" branding exceed what the studies support.

Expressive writing (C). Guo 2023 meta of 31 RCTs (N=4,012): d=0.07-0.16, inconsistent at 6 months (Guo 2023). Identity-specific outcomes rarely measured. Effect size is small and the cost is negligible; evidence does not support use as a primary identity intervention.

Flotation-REST (C). Feinstein/LIBR 2024 PLOS One safety RCT (N=75) confirmed tolerability (Feinstein 2024). 2018 pilot (N=50) showed GAD symptom reduction sustained at 6 months (Feinstein 2018); 2023 anorexia RCT sustained body-dissatisfaction reductions at 6 months. No identity-specific instruments have been used. The anxiety signal is real; identity-change claims are unsupported, and the mechanism is more plausibly described as interoceptive training.

Wim Hof, SSP, taVNS, Holotropic (D). Wim Hof RCT 2024 (N=84): depression and anxiety outcomes at parity with active control at 3 months rather than superior (Wim Hof 2024). No personality measures. Holotropic Breathwork has one Czech study and no RCT-grade personality data. Safe & Sound Protocol and transcutaneous auricular VNS produce measurable acute autonomic effects but minimal trait-level evidence. Claims of "polyvagal trait change" are not supported by the current literature. Evidence supports use for state regulation rather than trait-level identity change.

Identity ranking

Ordered by evidence per unit of effort for trait-level identity change over a one-year horizon: MBCT or MBSR (A) is the best-supported intervention, with network-meta evidence and replication; IFS with an experienced practitioner (B) has one large-effect trial and a thin base and is promising but not replicated; ayahuasca in a medically screened container (C+) rests on one uncontrolled trial and the grade reflects novelty and mechanism plausibility rather than replicated effect, with the safety floor above as a prerequisite; expressive writing, flotation, breathwork, and SSP have evidence only as adjuncts and do not support use as primary identity interventions.

5. General cognitive augmentation

The target population for this section is cognitively intact healthy adults with no clinical indication, and the outcome variables of interest are working memory, processing speed, and learning rate measured directly.

Creatine 5g/day (A). Xu 2024 meta of 16 RCTs (N=492): memory SMD=0.31, processing speed SMD=−0.51 (faster), null for executive function (Xu 2024). Effect was larger in females and in adults aged 18-60, consistent with lower baseline dietary creatine. Sleep-deprivation rescue via elevated cerebral phosphocreatine is the best-characterized mechanism in the supplement literature, with MRS-verifiable changes in brain bioenergetic substrate. Effect size is modest, replication is good, and the adverse-event profile is negligible at the 5g dose. The evidence supports monohydrate at 5g daily; alternative forms (HCl, buffered) have no demonstrated advantage.

HIIT and resistance training (B+). Twenty-minute HIIT outperforms 30-minute for acute BDNF, with I²>75%, meaning the pooled SMD hides enormous between-study variance and the BDNF claim probably crosses zero on prediction intervals; the cognitive-outcome signal is more robust. Zone-2 cardio does not have the supporting evidence it has been granted culturally; the 2025 narrative review explicitly failed to support zone-2 as optimal for mitochondrial or neurogenic gains in the general population.

Resistance training shows a domain-specific cognitive effect that has been undersold in the exercise literature. A 2025 meta of 17 RCTs (N=739, mean N ≈43 per trial) reports working-memory SMD=0.44, spatial SMD=0.63, and null effects on executive function and processing speed (Front Psychiatry 2025). The domain specificity is consistent with a targeted PFC-hippocampal neurogenic mechanism rather than a generalized exercise-benefit handwave. The principal weakness is structural: 17 small RCTs with a selective pattern of positive and null domains matches the signature of publication bias, and a trim-and-fill correction has not been run. The tDCS entry in this section was downgraded to F on the same grounds (publication-bias-corrected effect near zero), and the same standard applies here. The working-memory and spatial signals are likely real; the effect-size point estimates are likely inflated. The grade was revised from A to B+ pending a bias-corrected meta.

The protocol consistent with current evidence is HIIT at 20 minutes, 2-3 times per week, with cognitively demanding work scheduled within the 2-4 hour post-exercise plasticity window, combined with resistance training 2-3 times per week for the working-memory and spatial signal. Post-regrade, this remains the strongest-evidenced cognitive-augmentation stack in this section, exceeded only by creatine among the supplement entries.

Neurofeedback (C). SMR neurofeedback for anxiety: d=0.7-0.9 with sham control. Roughly 50% of trials do not use sham, and the sham-controlled effect is smaller than the pooled estimate. Lab-grade protocols with operator expertise and appropriate EEG produce a real signal; consumer devices (Muse, Narbis) have not demonstrated effects above placebo and regression to the mean. Evidence supports use in research-grade clinics with sham-controlled protocols rather than in consumer hardware.

Working-memory training (D). Near-transfer effects on the trained task are real. Far-transfer to fluid intelligence and general attention has not replicated across approximately fifteen years of study, including dual n-back protocols.

tDCS / tACS (F). 2024 Brain Research analysis across 134 interventions, correcting for publication bias, put the tDCS effect on working memory at roughly zero (Brain Research 2024). Executive function effects η²p=0.058-0.075 pre-correction collapse further under bias adjustment. Consumer tDCS/tACS devices are not supported by the bias-corrected evidence base.

L-theanine + caffeine (C). Two-hour window attention-switching SMD=0.33 (theanine+caffeine meta). Small acute effect, no evidence of durable benefit. Compatible with routine morning caffeine use.

Exogenous ketones (C). 2026 meta of 38 studies, N=1,602: pooled SMD=0.29, I²=91% (Ketones meta 2026). The pooled SMD at I²=91% is a fiction. When 91% of observed variance is between-study, the meaningful quantity is the prediction interval, which crosses zero. The literature does not agree on what this does. Acute dose may help under sleep deprivation or metabolic stress. Treat it as N-of-1 self-test, not a population effect of SMD=0.29.

Within this section, the strongest-evidenced interventions for cognitive augmentation in a healthy adult are combined HIIT and resistance training (B+ after publication-bias-aware regrading) and daily creatine at 5g (A). L-theanine with caffeine and situational exogenous ketones have small real effects and are compatible adjuncts. Consumer tDCS, tACS, consumer neurofeedback, dual n-back, and Cogmed do not survive bias-corrected analysis and are not recommended on current evidence.

6. Additional interventions

The following interventions were outside the scope of the initial audit but have sufficient 2023-2026 trial data to grade.

Semaglutide for alcohol use disorder (B+). Hendershot 2025 Phase 2 RCT (N=48, 9 weeks, 0.25→0.5→1.0 mg/week) in adults with AUD. Drinks per drinking day β=−0.41 (p=.04); lab self-administration Cohen's d=0.46-0.48; weekly craving reduced; heavy drinking days favored semaglutide (JAMA Psychiatry 2025). Mechanism: GLP-1 receptor agonism attenuates dopaminergic reward salience in VTA/NAc, generalizing across palatable substances. Observational cohort N=83,825 replicates: 50-56% lower AUD incidence/recurrence in semaglutide users vs other weight-loss drugs (Nat Commun 2024).

Klausen's exenatide-for-AUD reanalysis is the caveat: alcohol-suppression concentrated in BMI >30, attenuated at healthy BMI. The GLP-1-for-addiction mechanism may require metabolic dysfunction as substrate. A healthy-BMI AUD patient is different pharmacology.

Safety additions for the obese-plus-AUD population. GLP-1 agonists strip lean mass at about 25% of total weight loss; in AUD patients with pre-existing alcohol myopathy and nutritional deficits, this matters more than in general-obesity populations. Monitor grip strength or DEXA; co-prescribe resistance training; replete B1/B12/folate as standard. Gastroparesis is real and can persist after discontinuation; the ASA has issued guidance on perioperative airway aspiration risk (7-day hold for elective surgery, anesthesia consult for emergent). No causal suicidality link in 2024-25 regulatory review. Real Phase 2 signal in obesity plus AUD with the safety additions above. Watch for Phase 3 and for gambling and compulsive-behavior trials. Don't extrapolate to healthy-BMI-only AUD without waiting for trials that enroll them.

Acupuncture for combat PTSD (B). VA trial, N=93 combat veterans, 15 weeks, verum vs minimal-needle sham. CAPS-5: verum d=1.17, sham d=0.67, between-group d=0.63 (p=.005). Fear-potentiated startle recovery enhanced in verum, a biological endpoint that points to extinction-learning modulation rather than pure expectancy. Retention 77% across arms (PMC10882512).

Known weakness: one trial, one site, one PI (Hollifield has been on most of the PTSD-acupuncture signal). The sham-acupuncture literature as a whole is a decade-long cautionary tale about sham design: minimal-needle sham is not inert, and verum-vs-sham effect sizes historically shrink toward null as sham fidelity improves. An earlier draft graded this A−; that was single-trial-treated-as-replicated, the failure mode this document was supposed to avoid. B until independent replication. Generalizability to non-combat PTSD is untested; anxiety/depression sub-scale effects were sham-sensitive. For combat PTSD with access to a protocol matching the trial's verum design, defensible. Not the top identity-adjacent intervention; IFS and MBCT have comparable or stronger evidence per unit of design quality.

Urolithin A / Mitopure (B). Singh 2022 RCT in middle-aged adults (4 months): muscle strength +12%, VO₂ gains of clinical magnitude, acylcarnitines and CRP decreased (PMC9133462). Mechanism: urolithin A is a gut-microbiome metabolite of ellagitannins that induces mitophagy and clearance of damaged mitochondria. Conflict-of-interest note: the primary RCT and the upcoming CLARITY cognitive trial are both funded by Amazentis (the Mitopure manufacturer); independent replication has not been conducted. Current evidence supports the muscle and endurance signal subject to sponsor-independence caveats. Cognitive endpoints are pending CLARITY (N=650, readout expected March 2026), and claims of cognitive benefit should wait on that readout and on an independent replication.

Transdermal estradiol in perimenopause (C+). Canadian Longitudinal Study of Aging 2025, N=7,251 postmenopausal women. Transdermal users had better episodic memory (d=0.303) than never-users; oral users had better prospective memory (d=0.283). Executive function null (PMC12380479). No RCTs in perimenopause proper; that's the gap. Small controlled trials of transdermal E2 plus micronized progesterone in early perimenopause show depressive-symptom rates of 17.3% vs 32.3% placebo.

For perimenopausal women with cognitive or mood symptoms, the observational effect size for transdermal estradiol exceeds that of most cognitive interventions in this document. Current prescribing practice remains shaped by the WHI-era combined-therapy data in a population the original trial did not enroll, and this indication is the one most directly affected by the asymmetry described in §2.

SGLT2 inhibitors (C). Preclinical: BBB penetration 0.3-0.5 brain:serum, BDNF upregulation, AChE inhibition, protection against Aβ₁₋₄₂ toxicity. Dapagliflozin early-AD RCT (N=48, 12 weeks): Stroop improved (p=.046 as secondary among many) but NAA/Cr null and ADAS-Cog/MMSE unchanged (PMC12198475). Mechanism is promising; human cognitive signal is thin. For patients with a metabolic indication, use of the class is reasonable. Evidence does not currently support use as a primary cognitive intervention in healthy adults.

Methylene blue (B for mechanism, C for human data). Mechanism is among the most clearly characterized in the document: methylene blue acts as a redox mediator donating electrons to Complex IV, bypassing Complex I/III dysfunction, which constitutes a rescue pathway for mitochondrial inefficiency. The dose-response is U-shaped, with lower doses in the characterized range producing enhancement and higher doses increasing oxidative stress. A small healthy-subject RCT (N=26) showed fMRI activation and memory improvement; a Phase 2 AD trial (N=321, 24 weeks) improved ADCS-CGIC and MMSE. No large 2024-2026 RCTs have been conducted, as no sponsor has funded one (ADDF review).

Safety: methylene blue is a potent MAO-A inhibitor at doses characterized in the cognitive-enhancement literature. Serotonergic contraindications include SSRIs, SNRIs, tramadol, dextromethorphan, MDMA, bupropion, St John's wort, and any MAOI; co-administration has produced serotonin syndrome with documented fatal cases in surgical anesthesia. G6PD deficiency produces hemolytic anemia on exposure and requires pre-screening. Non-pharmaceutical (dye-grade, chemistry-supplier) product is contaminated with heavy metals and uncharacterized for human use; only pharmaceutical-grade USP/EP product is appropriate for human administration. Specific dosing is not published in this document; dose-ranging belongs in a clinician-supervised context.

The case for methylene blue is strong on the economic side of the asymmetry (the absence of a sponsor explains the absent Phase 3) and weaker on the efficacy side: the Phase 2 AD signal has not been independently replicated, and Phase-2-to-Phase-3 attrition base rates in this space favor null results. The current evidence supports continued academic and funding interest; it does not support self-administration given the MAOI interaction surface and sourcing constraints.

Lion's mane (C). Docherty 2023 RCT (N=41, 1.8g): Stroop p=.005 at 60 min post-dose; 28-day stress trend p=.051 (Docherty 2023). Acute effect only; no sustained cognitive-domain improvement. Replication in larger samples is pending. Evidence supports a small acute attention signal; no durable cognitive effect has been demonstrated.

Bromantane / Ladasten (C). N=728 multicenter 28-day trial (50-100 mg/day): CGI-S 76% responders, CGI-I 90.8%, AE 3%, discontinuation 0.8% (PMID 21322821). Single trial, Russian-language primary publication, no Western replication, no independent AE adjudication. CGI-S/CGI-I are rater-impression endpoints and vulnerable to sponsor effects in open-label or single-blind designs. Mechanism: atypical psychostimulant with anxiolytic profile. Regraded from B− to C on grounds that a single uncorroborated CGI-driven trial does not support a B grade regardless of N. Use as a personal intervention is not supported: no regulated US supply exists, gray-market sourcing is unassayed, and the adverse-event profile in naive users is uncharacterized. The signal is hypothesis-generating.

Psilocybin microdosing (D). Szigeti 2021 self-blinding RCT (N=191), the largest placebo-controlled psychedelic microdose study to date: outcomes improved in both microdose and placebo arms without significant between-group differences, with over 50% blind-breaking (eLife 2021). The current evidence is consistent with expectancy effects rather than pharmacological action.

Spermidine (C). Kiechl Bruneck cohort showed an observational dose-response on cognitive-impairment risk. Schwarz 2022 RCT (N=100, 12 months, 0.9 mg/day): Mnemonic Similarity Task null (−0.03, p=.47) (Schwarz 2022). Wirth 2018 and Pekar 2021 were also null. Current randomized data does not support supplement-level cognitive claims despite the preclinical autophagy mechanism.

Cold water immersion (C). PLOS ONE 2025 meta (11 RCTs, N=3,177): stress reduction observed at 12 hours post-immersion only (not immediately, at 1 hour, 24 hours, or 48 hours); inflammation increases transiently at the 1-hour mark (PLOS One 2025). No durable mood or cognitive effect was observed.

Selegiline (D). Knoll's original aging hypothesis from 1988 has not been replicated in recent trials; 2023 Semmelweis BPAP-derivative work was negative. No active trial program is pursuing the cognitive-aging indication at present.

MDMA-PTSD, post-FDA rejection (D). MAPP1 and MAPP2 had functional unblinding over 90%, and three papers were retracted by Psychopharmacology in August 2024 over site violations. Headline efficacy (88/87% clinically meaningful response, 67/71% no longer meeting PTSD criteria) is large, but the trial-quality ceiling is compromised. Further evidence from Lykos's revised Phase 3 or an independent replication is required.

Within this section, the interventions best supported on current evidence are: acupuncture for combat PTSD where the sham-controlled verum protocol is available, semaglutide for AUD in patients with metabolic substrate, transdermal estradiol in perimenopause (the case most directly illustrating the asymmetry in §2), urolithin A for healthspan endpoints with cognitive data pending (CLARITY), and methylene blue as a mechanism-plausible but trial-limited case of the unfunded-generic pattern. Current evidence does not support use of psilocybin microdosing, spermidine, cold water immersion for cognition, selegiline, or the currently-submitted MDMA-PTSD protocol.

7. Exploratory premises, pressure-tested

The following theoretical frames recur in the identity-change and cognitive-augmentation literatures; each is evaluated on its current empirical support.

Predictive-coding destabilization (flotation, fasting, novelty, sleep deprivation) is a coherent framing that has motivated serious research. The empirical base is thinner than the framing implies: flotation has anxiety durability but no identity-scale measures (Feinstein 2018); a 6-week fasting/keto RCT returned non-significant cognitive endpoints; extreme novelty exposure has no controlled trials; and sleep deprivation produces state effects without durable trait-level change. "Prior-loosening" is not currently operationalized as a measured variable, and protocols organized around it do not have corresponding empirical support.

Vagal and autonomic trait-change protocols (SSP, taVNS, HRV biofeedback) produce measurable acute autonomic effects. A durable shift in autonomic setpoint has not been established. Polyvagal theory, as currently applied in consumer protocols, makes claims stronger than the available evidence supports. Use is reasonable for state regulation; trait-level claims are not supported.

Somatic and exteroceptive interventions (cold exposure, Feldenkrais, interoceptive training) show acute effects on norepinephrine and dopamine under cold exposure, with claimed durability resting on self-report. Feldenkrais-adjacent work lacks RCT-grade data on cognition or identity outcomes. The interoceptive-training signal is concentrated in MBCT and flotation.

Language-as-reconsolidation frameworks (Coherence Therapy, IFS, expressive writing) are grounded in reconsolidation biology (Nader; Ecker's clinical translation). The RCT base is uneven: IFS has one well-conducted trial (d=−0.9, PARTS 2024), Coherence Therapy has none, and expressive writing has 31 RCTs with d=0.07-0.16. Among therapies claiming reconsolidation mechanisms, IFS has the strongest current empirical support.

Metabolic substrate shifts (ketones, fasting, metformin, rapamycin) have minimal human RCT data with cognitive endpoints in healthy adults. The PEARL trial for rapamycin showed a mood and well-being signal at 5mg/week (ηp²=0.108-0.166 at 48 weeks) with no cognitive endpoint formally assessed (PEARL). Metformin in non-diabetics rests on observational data (8% cognitive-impairment reduction, 10% dementia reduction) confounded by the healthy-user effect; TAME is delayed (TAME). Taurine (Singh 2023) is preclinical, without human cognition data, and with a conflicting 2025 follow-up on the aging link (Singh 2023). The category is mechanism-heavy and data-light on cognitive endpoints in healthy adults; use on current evidence should follow the stated metabolic indication.

Morning bright light phase-shifts the circadian axis. Same-night PSG architecture effects are weak or null. The evidence supports use for sleep-timing stability rather than as a direct cognitive intervention.

8. Sleep as the consolidation substrate

Sleep functions as the consolidation substrate through which the other interventions in this audit operate rather than as a parallel category. The 2023-2025 literature converges on a specific negative finding: pharmacologic enhancement of slow-wave quantity does not reliably produce corresponding gains in memory consolidation, and the relevant target variable is the architecture of slow-oscillation / spindle coupling rather than time-in-SWS.

Closed-loop auditory stimulation, CLAS (B, young adults only). Ngo 2013 established the mechanism: acoustic coupling to endogenous slow oscillations, enhanced SO-spindle coupling, hippocampal consolidation, d=0.99 on declarative memory (Ngo 2013). MDPI 2024 three-night continuous CLAS protocol: +35% language learning, +26% discovery learning in young adults (MDPI 2024). Frontiers Sleep 2023 meta: effect sizes declining about 91.8% per year across replications, many null, age-stratified. Auditory evoked responses are around 50% weaker at 55+, and memory consolidation fails or worsens in middle-aged cohorts. The mechanism requires intact SO amplitude and thalamocortical excitability, both of which degrade with age. The evidence supports use under 45; over 55 the signal is null or reversed. Home devices (e.g. Dreem) have uncertain comparability to research-grade protocols.

DORAs (B). Daridorexant PSG data N=1,466: wake transitions reduced 6.9%, N3 and REM preserved (Mignot 2022). Lemborexant: sleep onset SMD −0.43, less next-day impairment than zolpidem (Rosenberg 2019). Mechanism: OX1R/OX2R antagonism removes wake-promoting drive without suppressing sleep-generation machinery, which is how they differ from benzodiazepines and Z-drugs. No DORA trial has measured memory consolidation as a primary endpoint. Preserved architecture on PSG does not establish preserved consolidation, as the sodium oxybate literature demonstrates. Among pharmacological sleep agents, DORAs have the most favorable architecture profile and are reasonable first-line for insomnia over Z-drugs. Use as a consolidation-enhancing agent is mechanism-plausible and empirically untested.

Sodium oxybate (F). Oxybate dose-dependently raises stage-3/4 EEG slow-wave time. Vienne 2016 and subsequent work show that pharmacologically induced slow waves are "functionally dissimilar to physiological slow waves," with memory consolidation paradoxically null in several direct tests (Vienne 2016). SO-spindle coupling is disrupted under oxybate even as stage-3/4 time increases, and consolidation in the relevant paradigms tracks coupling rather than quantity. The finding is longstanding within narcolepsy pharmacology and has not been incorporated into the cognitive-sleep-medicine framing of SWS as a consolidation target. The implication generalizes to thermal devices, magnesium, apigenin, and other interventions marketed on a "deeper sleep" mechanism: increasing SWS quantity does not reliably increase memory consolidation.

Trazodone low-dose, doxepin 3-6mg (C). Trazodone meta 2022: increased N3/SWS, reduced arousals, but only in insomnia populations and not in healthy adults (Trazodone 2022). Doxepin 3-6mg preserves architecture with minimal data in healthy subjects. Both are reasonable for insomnia with preserved architecture; no memory-consolidation data either way.

Thermal devices and sleep supplements (F). Eight Sleep and ChiliPad have no PSG data, only self-report. Magnesium formulations and apigenin rest on wearable-algorithm or subjective data without PSG-grade evidence. Glycine 3g has minimal PSG data and some reduction in SWS latency. None have a demonstrated effect on consolidation.

Morning bright light (C). Reliable circadian phase-shift; same-night PSG architecture effects weak or null. Relevant to sleep-timing stability rather than direct overnight consolidation. Low cost.

On the current evidence, the reasonable sleep protocol is architecture-protective rather than SWS-maximizing: morning bright light within an hour of waking, a DORA for insomnia when indicated, and CLAS under age 45 where declarative consolidation is the target. Thermal devices and the standard sleep-supplement category lack PSG-grade support.

9. GLP-1 at healthy BMI

Non-metabolic claims for GLP-1 agonists are drawn almost entirely from obese or overweight samples, and the one trial that directly tested the effect-size gradient across BMI showed the benefit concentrated at high BMI.

In obese samples, the non-metabolic signals are consistent. The JAMA Psychiatry 2025 Phase 2 RCT (N=48) in AUD showed grams consumed β=−0.48 and heavy drinking days d>0.80 (JAMA Psychiatry 2025). A real-world cohort (N=83,825) reported 50-56% reduction in AUD incidence and recurrence (GLP-1 AUD cohort). Parkinson's epidemiology in GLP-1 users shows 36-60% risk reduction, and preclinical work shows robust microglial M1-to-M2 polarization.

Klausen's exploratory analysis of the exenatide AUD trial found the alcohol-suppression effect concentrated in BMI>30 and attenuated at healthy BMI, suggesting that the addiction-relevant mechanism may depend on metabolic dysfunction as substrate. Extrapolating obese-sample benefits to healthy-BMI use is not supported by current data.

OxSENSE (N=60, healthy BMI 18-30) is the trial that will address the anhedonia endpoint at therapeutic doses; results are pending. Sarcopenia risk is relevant at healthy BMI: GLP-1 agonists produce lean-mass loss at approximately 25% of total weight loss, and the ratio is less favorable at normal BMI with a smaller adipose buffer. 2024-25 regulatory review found no causal link to suicidality.

On current evidence, use of GLP-1 agonists at healthy BMI for identity, cognitive, or wellness indications is not supported, and OxSENSE results should inform any reassessment. For comorbid obesity plus AUD, the evidence base is now stronger than for most dedicated AUD pharmacotherapies, and use under a combined metabolic and addiction indication is supported.

10. Stacking: what combines

Most stacking arguments collapse on close inspection: components hit the same substrate with diminishing returns, or they interfere. Two stacks have coherent mechanism and low interference risk.

Creatine plus HIIT with cognitive work in the BDNF window (B+). Different substrates. Creatine buffers ATP via phosphocreatine; HIIT elevates BDNF and drives plasticity signaling. Additive by construction, no known interference. HIIT BDNF peaks at 15-30 min post-exercise with a 2-4 hour window for LTP-favorable learning. HIIT 20 minutes, then within 30 minutes the cognitively demanding block (learning, MBCT session, identity journaling). Creatine 5g daily, independent of timing. The cleanest real stack in the corpus.

Morning bright light plus DORA (C+). Distinct systems. Light on the circadian axis (SCN phase-shift), DORA on the homeostatic sleep-wake switch (orexin antagonism). No plausible pharmacological overlap. Used as intended (30-minute bright light within an hour of waking, DORA 30 minutes before target sleep) they complement. Low burden.

The reconsolidation case (speculative B). A high-quality NREM night following a session that opens a reconsolidation window (MBCT, IFS, or psychedelic integration) might preferentially re-encode the updated schema rather than the original. Mechanistically coherent: Nader's reconsolidation animal work is real, MBCT's active retrieval of negative self-schemas followed by reappraisal fits the reconsolidation template, and SWS/REM do handle emotional-memory integration. Zero direct RCTs. Speculation that's plausible enough to try.

Interference cases to avoid:

• GLP-1 plus psychedelic therapy. GLP-1R expression in VTA/NAc, AUD evidence of reward-circuit attenuation, psychedelic therapeutic action requires affective salience. Mechanistically you'd expect the GLP-1 to dampen exactly what the psychedelic is trying to amplify. No trial has co-administered; nobody should yet.
• Metformin plus HIIT for neuroplasticity. Metformin's Complex I inhibition raises AMPK, and exercise-induced mitochondrial biogenesis also runs through AMPK. Konopka and Burd-adjacent work shows metformin attenuates VO2max gains and mitochondrial protein synthesis from training. If you're using HIIT for cognitive or BDNF reasons and taking metformin for longevity, you're partially canceling the HIIT adaptation.
• Ketones plus HIIT. Not harmful, probably redundant. Both push FNDC5/irisin and PGC-1α. If you're already in ketosis and doing HIIT, additional exogenous ketones hit a ceiling.
• Sodium oxybate for SWS. Pharmacological SWS is not functional SWS. Net memory effect probably neutral or negative.
• CLAS at age 45+. Fails or reverses. Wrong-population application of a real mechanism.

The executive-function cost of maintaining a multi-intervention protocol is not measured in trial settings but is relevant at the individual level. A small number of correctly-stacked interventions maintained consistently is likely to outperform a larger number managed inconsistently.

11. Four cases, four different rate-limiting steps

The following four cases all involve interventions with usable data that remain out of mainstream use, but the rate-limiting step differs in each. An earlier draft treated them as a single "suppression" phenomenon, which was the document's central analytic weakness. An earlier draft also included low-dose lithium; LATTICE was null on six coprimaries at p<0.01, which is an evidence case rather than a suppression case, and the entry has been removed.

Nicotine patches for MCI: CMS coverage plus guidelines plus Hatch-Waxman

Rate-limiting step: payer coverage and specialty-society guidelines. Transdermal nicotine is already FDA-approved. Newhouse/Levin 2012 pilot (N=74 MCI): attention, memory, and psychomotor gains; MCI patients recovered 46% long-term memory vs 26% decline in placebo (Newhouse 2012). Mechanism: nAChR upregulation in hippocampus and PFC, pharmacologically distinct from tobacco's harm profile when delivered via patch. MIND has been ongoing since 2015 without a positive readout (MIND), which should update the prior toward null. The binding constraint is CMS/payer coverage for off-label MCI use, the absence of an AAN/Alzheimer's Association guideline, and Hatch-Waxman economics: a new indication on an off-patent generic confers at most 3-year 505(b)(2) exclusivity, which doesn't price-protect a commodity patch. Social stigma compounds but isn't the binding constraint.

For carefully selected MCI patients who have exhausted standard options, transdermal nicotine is defensible as a physician-run N-of-1: a prescribing neurologist or geriatric psychiatrist designs the protocol (baseline ECG, MoCA/CDR-SB, orthostatic vitals, pre-specified dose, pre-specified stop rule at a defined MoCA delta or AE threshold). Not for self-initiation. Contraindications include suspected DLB (cholinergic tone is already collapsing), symptomatic cardiovascular disease, pregnancy, and current smoking. Doses aren't published here: the amnestic-MCI Newhouse protocol does not generalize to other MCI subtypes and the indication does not generalize to cognitively healthy readers. Specifics belong in the prescribing conversation. The evidence is better than the current default of nothing; it is not better than a prescribing clinician's judgment.

Estrogen-only HRT post-WHI: malpractice exposure plus ACOG/NAMS guideline inertia

Rate-limiting step: prescriber malpractice exposure and specialty-society guideline stickiness. FDA never pulled estradiol; ACOG took roughly 14 years to reverse tone post-WHI. The 2024 Menopause reanalysis of WHI data showed estrogen-only initiated before 65: 19% mortality reduction, reduced breast, lung, and colorectal cancer, reduced dementia (Menopause 2024 reanalysis). WHI's 2002 overcorrection, driven by the combined CEE+MPA arm in older late-initiators, permanently encoded "HRT = breast cancer = bad doctor" into clinical culture. Generic estradiol is cheap and no market exists to rebuild prescribing norms. Residual liability for prescribers who move ahead of guidelines does the rest. For post-hysterectomy women initiated before 65, estrogen-only HRT has a favorable risk-benefit profile that the clinical default of tapering does not match.

Testosterone optimization in eugonadal men: nosology, not regulation

Rate-limiting step: absence of a defined clinical indication. "Low-normal with symptoms" isn't a disease state in ICD-10. TRAVERSE was funded and completed for hypogonadism; the system worked there. This is a nosological problem, not a regulatory one. TRAVERSE (N=5,246 hypogonadal): MACE noninferior; AFib 3.5% vs 2.4%, PE 0.9% vs 0.5%, AKI 2.3% vs 1.5% (Lincoff 2023). Reading the noninferiority result as dispositive while ignoring three cardiovascular endpoints trending toward harm would be the asymmetric-data handling this document critiques elsewhere; the honest read is that TRT in hypogonadal men has an acceptable risk-benefit profile with monitoring and is still undertreated, not that it's unambiguously safe. The eugonadal-optimization space remains unstudied. For symptomatic men with low-normal testosterone plus metabolic or cognitive complaints, a closely monitored trial under a prescribing endocrinologist or urologist is defensible (baseline hematocrit, PSA, estradiol, sleep-apnea screen; q3-month labs for the first year). Evidence in hypogonadal patients is better than the discourse suggests; the eugonadal case remains absent.

Silo cases (data exist, wrong specialty reads them)

Sodium oxybate SWS-quality finding: trapped in narcolepsy pharmacology, would reshape how cognitive sleep medicine targets interventions. Ayahuasca NEO-PI durability: trapped in ethnobotany and a small psychedelic cluster, invisible to personality-trait neuroscience proper. The d=2.20 openness effect size would be paradigm-shifting in trait psychology if it were cited there. GLP-1 in addiction medicine: endocrinology-owned; addiction psychiatrists are not yet deploying despite JAMA Psychiatry 2025 and the 83k cohort.

12. Evidence-grade index

Every intervention graded in this document. Scan for the grade, jump back to the section for the reasoning.

Index reflects 2026 evidence. Some rows await major trial readouts (CLARITY for urolithin A cognition, OxSENSE for GLP-1 healthy-BMI, TAME for metformin cognitive endpoints) and will move when those close.

12a. Known limitations

An adversarial review of the draft produced the following caveats, retained here so that readers who skip caveat sections still encounter them.

1. Several grades in earlier drafts rested on a single trial from a single group and were presented as replicated findings. Hollifield acupuncture-for-PTSD, Frecska ayahuasca, and PARTS IFS fall into this category. The regrade moved acupuncture from A− to B, ayahuasca from B to C+, and resistance training from A to B+. The index in §12 flags single-trial entries. Under the revised standard, A-grade requires independent replication.
2. The harm ledger in §3 is confirmation-selected. Each row is a condition where the "withheld intervention" framing already has purchase in mainstream medicine. Conditions in which the ledger would favor the cautious default are absent: gabapentinoid overprescription, SSRIs in adolescents with an emergent suicidality signal, benzodiazepine dependence cascades, bariatric overuse. The "37,000 deaths/yr from X-waiver removal" row is a modeled counterfactual rather than an observational hazard ratio; the source-class column now marks this distinction.
3. "Regulatory asymmetry" in §2 conflates at least six gatekeepers: FDA approval, DEA scheduling, CMS and commercial-payer coverage, specialty-society guidelines, prescriber malpractice exposure, and manufacturer economics under Hatch-Waxman. The six fail through different mechanisms and require different corrections. Nicotine for MCI is rate-limited by payer coverage and guideline inertia. Estrogen-only HRT is rate-limited by malpractice exposure and ACOG/NAMS guideline stickiness. Low-dose lithium was rate-limited by the evidence itself (LATTICE was null on six coprimaries at p<0.01); the entry has been removed from §11. Testosterone in eugonadal men is rate-limited by the absence of an ICD-10 indication, a nosological rather than regulatory constraint.
4. Adverse-event accounting in the original draft ran in one direction. The ledger counted deaths from underdeployment and did not count HPPD, ayahuasca-precipitated psychosis, ibogaine QTc deaths, or serotonin-syndrome risk from MAOI interactions in some of the compounds discussed. Each pharmacological recommendation now includes a contraindication and AE block. Specific doses have been removed from the public text where publishing them would have invited self-initiation.
5. Phase 2 to Phase 3 attrition disciplines the optimism in several sections. Roughly 10-20% of Phase 2 signals in neuropsychiatry survive Phase 3. LATTICE, TAME-adjacent metformin literature, and the spermidine sequence share a pattern of positive observational or Phase 2 data followed by null RCT results. The grades on methylene blue, nicotine-for-MCI, and other "real Phase 2, no Phase 3" cases are now conditioned on that base rate.
6. The central frame is weaker than the original draft argued. The defensible claim is that regulatory conservatism is miscalibrated, in some cases substantially. A stronger claim, which the prose occasionally implied, is that the asymmetry is incoherent and that any intervention with plausible mechanism is being withheld unjustly. That stronger claim does not hold: Hansson, Steele, and Stirling give a serious defense of asymmetric weighting under deep uncertainty; the aducanumab pushback indicates that Type-I detection remains functional on longer timescales; and "what society should allow" and "what an individual should take" are separate questions with different decision structures. The claim throughout this document is the weaker one.
7. The argumentative structure resembles the structure of enthusiasms that later failed. HRT-as-general-prophylaxis before WHI, SSRIs in adolescents before the black box, long-term opioid therapy before 2010, and several others had defensible mechanism, Phase 2 data, and careful advocates at the time. The possibility that the current round is similarly miscalibrated is not ruled out by the evidence assembled here. N-of-1 framing and per-intervention monitoring plans are the intended hedge; aggregate confidence is not.
8. The §2 frame is partly absorbed by the per-case diagnosis in §11. Once the six gatekeepers fail through different mechanisms and each case is rate-limited by a different component (payer coverage, guideline stickiness, malpractice exposure, Hatch-Waxman economics, nosology), "regulatory asymmetry" describes a cluster of different problems rather than a single miscalibration. The historical Thalidomide → Vioxx → Aducanumab arc in §2 carries the FDA approval-bar case but does not structurally carry the others, and §2 should be read as motivating the typology in §11 rather than as an independent unified argument.
9. The strongest defender of the asymmetric regulatory posture is not the symmetrically-applied precautionary principle Sunstein dismantles. It is the observation that approvals are precedents rather than prescriptions: once on-label, a molecule is durably available and difficult to recall, while non-approval is reversible against new evidence. Option-value and precedent-durability asymmetries are a principled reason to set different Type I and Type II thresholds. The document prices body counts across the two error classes without pricing this reversibility asymmetry, and the defender's ratio may be higher than the naive symmetric calculation implies.
10. The 37,000-deaths X-waiver figure in §3 is a modeled counterfactual with large assumption sensitivity on uptake, retention, diversion, and substitution. The source paper gives bounds; the table gives a point estimate. Treat it as a central draw from a wide distribution, not a settled scalar. The "tens of thousands of US lives per year" claim in §1 survives at the low end of that distribution but depends on multiple rows of §3 surviving their own confounds.
11. Observational healthy-user confound is not applied consistently. Metformin in non-diabetics is downgraded to D explicitly for healthy-user confound in observational cohorts. Transdermal estradiol in perimenopause (CLSA) and estrogen-only HRT beyond 65 (Baik 2024) rest on the same methodological class and are not downgraded correspondingly. The asymmetry may be defensible on mechanism and effect-size grounds, but the document does not explicitly justify it, and a parallel treatment would shift the C+ on transdermal E2 and the A on Baik toward the metformin grade.
12. The §13 recommendations are not probability-weighted by the attrition and confound priors named earlier in §12a. Under a 10-20% Phase-2-to-3 survival prior, C+ recommendations (ayahuasca, transdermal E2) should carry lower decision weight than the ordered list implies, and several B-grade entries (semaglutide-AUD, 5-MeO-DMT BPL-003, methylene blue) are Phase 2 signals that a consistent application of the attrition rule would discount further. The list is ordered by evidence per unit of effort at the current evidence level rather than at the prior-adjusted posterior.
13. Functional unblinding affects several cited psychedelic trials. BPL-003 5-MeO-DMT, Griffiths 2016 psilocybin for EOL distress, and the now-retracted MAPP1/2 MDMA trials share the difficulty that an IV or high-dose psychedelic is distinguishable from placebo by both subject and rater within minutes. Effect sizes from these trials should be discounted for expectancy in proportion to how much of the therapeutic model invokes set, setting, and integration. The document retains the headline numbers with this caveat understated in the per-entry text.
14. The recommendation set assumes a reader who can afford clinician-supervised interventions (quarterly labs, prescribing specialists, international travel for ibogaine or ayahuasca clinics, out-of-pocket GLP-1 or DORA scripts), is not currently on psychotropic medication that forecloses several compounds, is not pregnant or pediatric, and has the executive function and pharmacological literacy to run per-intervention monitoring. These constraints are implicit in §13's opening clause ("cognitively and metabolically healthy adult with no clinical indication") but are not costed. For a reader outside that profile, the list collapses to creatine, HIIT/resistance training, morning bright light, and an 8-week MBCT course with a trained teacher.

The core claim of the document is narrow: a subset of interventions is underdeployed because the ledger of harms tracks visible action more carefully than invisible inaction. The grades, recommendations, and framing carry the failure modes listed above, and the document should be read as a position paper with flagged uncertainties rather than a settled audit.

13. Recommendations

The following recommendations apply to a cognitively and metabolically healthy adult in 2026 pursuing identity and cognitive work, with no clinical indication driving the choice. Entries are ordered by evidence per unit of effort.

1. Creatine 5g/day monohydrate (A). Modest, replicated effect with negligible adverse-event profile.
2. HIIT 20 minutes, 2-3×/week combined with resistance training 2-3×/week (B+). The strongest-evidenced cognitive stack in this document after publication-bias-aware regrading. Cognitively demanding work is best scheduled within the 30-minute post-HIIT BDNF window.
3. MBCT 8-week course with a trained teacher (A). The best-supported endogenous identity intervention for reduction of negative self-referential processing and DMN activity. Consumer app implementations have not demonstrated equivalent effects.
4. Morning bright light within the first hour of waking (B). Sleep-timing stability with negligible cost.
5. Ayahuasca in a medically screened container (C+), where trait-level identity restructuring is the primary goal and standard psychedelic options are unavailable. Evidence base is one uncontrolled N=24 trial, and the grade reflects mechanism plausibility rather than replicated effect. Use requires the safety floor detailed in §4.
6. IFS with an experienced practitioner (B), where available. One large-effect trial on a thin RCT base.
7. DORA (daridorexant or lemborexant) (B) for insomnia. Not indicated as a general cognitive optimization.
8. L-theanine with caffeine (C) as a morning habit. Small acute effect.
9. Exogenous ketones situationally (C), with responder-phenotype dependence.

The current evidence does not support use of: consumer tDCS or tACS, consumer neurofeedback, dual n-back or Cogmed-style working-memory training, thermal sleep devices, Mg-threonate or apigenin as cognitive interventions, metformin/rapamycin/taurine/low-dose lithium for cognitive enhancement in a healthy adult (use for stated metabolic indications remains a separate question), GLP-1 agonists at healthy BMI for non-metabolic indications (OxSENSE pending), Wim Hof/SSP/taVNS/Holotropic as identity interventions, Coherence Therapy at a brand premium, sodium oxybate for memory consolidation, or CLAS over age 45.

Specific-indication recommendations where the indication applies: nicotine patches for MCI under clinician-supervised N-of-1 framing; estrogen-only HRT initiated before 65 post-hysterectomy; TRT in symptomatic low-normal men under monitoring; ibogaine for refractory TBI+PTSD at a clinic meeting the Williams protocol floor; GLP-1 agonists for comorbid obesity plus AUD; transdermal estradiol for perimenopausal cognitive or mood symptoms (the observational effect size d=0.303 exceeds that of most cognitive interventions reviewed, and the primary prescribing barrier is cultural rather than regulatory); and acupuncture for combat-PTSD under a sham-controlled verum protocol matching the Hollifield design.

What a non-asymmetric regulatory environment would additionally allow

Recommendations in the preceding section assume the current regulatory and reimbursement environment. If the no-intervention baseline were priced into regulatory decisions at its estimated mortality cost, the following would be expected to shift from "experimental" or "restricted" toward "available under supervision."

• Nicotine patches for MCI under a funded Phase 3 trial. The rate-limiting step is Hatch-Waxman economics on an off-patent generic, not evidence.
• Psilocybin for end-of-life existential distress. Griffiths and Ross produced replicated effect approximately a decade ago; the rate-limiting step is Schedule I status.
• Methylene blue under a properly funded dose-ranging trial. Mechanism is well-characterized, Phase 2 AD data exists, and the absence of Phase 3 tracks the unit cost of the molecule rather than the evidence.
• An off-patent repurposing pathway in general. Misoprostol, nicotine, low-dose lithium, estradiol in perimenopause, and metformin in aging share the same structural barrier: no non-industry route to Phase 3 authorization. A functioning pathway would close a substantial fraction of the gap at relatively low cost.
• Perimenopausal HRT as first-line for symptomatic cognitive and mood symptoms in women under 60, delivered transdermally, consistent with Baik 2024 and the CLSA observational data. This is already possible under current FDA labeling; the rate-limiting step is ACOG/NAMS guideline stickiness and prescriber malpractice exposure rather than regulatory prohibition.

This section separates the personal calculus from the societal one.

14. Method note

Literature search covered the 2023-2026 trial and meta-analytic record across PubMed, PMC, ClinicalTrials.gov, and the review literature. Every PMID, PMC, or DOI link was checked against the target abstract; citations that could not be verified against a public URL are retained descriptively without hyperlinks rather than fabricated. An adversarial review pass regraded approximately one third of entries. The regrades (ayahuasca B to C+, acupuncture A− to B, resistance training A to B+, bromantane B to C, psilocybin microdosing B to D, and removal of low-dose lithium from §11) are flagged inline and summarized in §12a.

15. References

1. Brewer JA et al. 2011. Meditation experience and default mode network activity. PNAS. PMID 22114193
2. Williams NR / Cherian K et al. 2024. Magnesium-ibogaine therapy in veterans with TBI. Nature Medicine. Nature Medicine | PMID 38182784
3. Feinstein JS et al. 2018. Floatation-REST in generalized anxiety disorder. PMID 29453033
4. Feinstein JS et al. 2024. Safety of Floatation-REST. PLOS One. PMID 38457451
5. Frecska E et al. 2020. Therapeutic potentials of ayahuasca revisited. PMC7524857
6. Guo L 2023. Expressive writing meta-analysis, 31 RCTs. PMID 36921878
7. Klausen MK et al. Exenatide for AUD; BMI-stratified exploratory analysis. (No verified digest URL.)
8. Kuyken W et al. 2016. MBCT in prevention of depressive relapse. JAMA Psychiatry. PMID 27119968
9. LATTICE investigators 2024. Low-dose lithium in MCI, 2-year RCT. JAMA Neurology
10. Lincoff AM et al. 2023. Cardiovascular safety of testosterone replacement (TRAVERSE). NEJM. PMID 37326322
11. Luan LX et al. 2024. Extended DMT infusion (DMTx) in healthy volunteers. J Psychopharmacol
12. MDPI Brain Sciences 2024. Three-night continuous CLAS. MDPI 2076-3425/14/11/1138
13. Mendelson JE et al. 2011. Salvinorin A dose-effects. PMID 21131142
14. Menopause 2024. WHI reanalysis, estrogen-only. PMID 38781402
15. Mignot E et al. 2022. Daridorexant in chronic insomnia. PMID 35395302
16. Nagele P et al. Nitrous oxide for TRD (pooled RCTs). PMC11097146
17. Newhouse PA et al. 2012. Nicotine patches in amnestic MCI. Neurology. PMID 22232050
18. Ngo H-VV et al. 2013. Auditory closed-loop stimulation. Neuron. PMID 23623061
19. Noller GE et al. 2017. Ibogaine for opioid dependence, 12-month follow-up. PMID 28402682
20. OxSENSE trial registration. GLP-1 anhedonia endpoint.
21. PARTS IFS trial 2024. PMID 38934934
22. PEARL rapamycin trial. NCT04488601
23. Resistance training cognition meta 2025. Frontiers Psychiatry 2025
24. Rosenberg R et al. 2019. Lemborexant phase 3 (SUNRISE-1). PMID 31819050
25. SRET / MBCT self-referential encoding. PMC10600813
26. Singh P et al. 2023. Taurine deficiency as a driver of aging. Science. PMID 37289866
27. TAME trial (metformin, delayed). AFAR/TAME
28. tDCS publication-bias-corrected meta. Brain Research 2024. PMID 38159856
29. Trazodone low-dose meta 2022. PMID 35226876
30. Vienne J et al. 2016. GHB and physiological sleep homeostasis. PMID 27253774
31. Wim Hof breathwork RCT 2024. PMID 39606690
32. Xu Y et al. 2024. Creatine and cognition meta (16 RCTs). PMID 38430492
33. L-theanine plus caffeine meta. PMID 31758301
34. Exogenous ketones cognition meta 2026. Frontiers Nutrition 2026
35. GLP-1 AUD real-world cohort (N=83,825). PMID 38806481
36. JAMA Psychiatry 2025. Semaglutide for AUD Phase 2 RCT (Hendershot). JAMA Psychiatry 2025
37. BPL-003 5-MeO-DMT Phase 2b. NCT05870930
38. MIND nicotine study. NCT02720445
39. Baik SH, Baye F, McDonald CJ. 2024. HRT beyond age 65. Menopause. PMID 38595196
40. Chalmers I, Glasziou P. 2009. Avoidable waste in biomedical research. The Lancet. PMID 19525005
41. Sunstein CR. 2002. Beyond the Precautionary Principle. Chicago Unbound
42. Gieringer DH. Drug Evaluations: Type I vs Type II Errors. RAND. eScholarship
43. Aducanumab regulatory review. PMC10193636
44. Griffiths RR et al. 2016. Psilocybin for cancer existential distress. PMID 27909165
45. Jones KF et al. 2023. Eliminate the Buprenorphine DEA X-waiver. PMID 36624606
46. Hollifield M et al. 2024. Acupuncture for combat PTSD. JAMA Psychiatry. PMC10882512
47. Singh R et al. 2022. Urolithin A (Mitopure) RCT. PMC9133462
48. CLSA 2025. Transdermal/oral estradiol and cognitive function. PMC12380479
49. Dapagliflozin in early AD RCT. PMC12198475
50. Methylene blue cognitive vitality review (ADDF). ADDF
51. Docherty S et al. 2023. Hericium erinaceus RCT. PMID 38004235
52. Szigeti B et al. 2021. Self-blinding psilocybin microdose RCT. eLife. eLife
53. Schwarz C et al. 2022. Spermidine supplementation and cognition. PMID 35616942
54. Cold water immersion meta-analysis 2025. PLOS One 2025
55. Bromantane (Ladasten) N=728 multicenter. PMID 21322821
56. GBD 2021 mental disorders. Lancet Psychiatry 2024. Lancet Psychiatry
57. Swedish PTSD suicide cohort. PMC7758737
58. ADHD mortality UK/Denmark cohort 2024. Trans Psychiatry. Trans Psychiatry 2024
59. Sleep disorders and dementia meta 2024-2025. PMC12518473
60. OUD treatment bookshelf. NBK541393
61. Treatment-resistant depression mortality 2024. J Affect Disord. Science Direct 2024