# HSEF Review: Serotonin deficiency hypothesis of MDD ## In plain terms For decades the most familiar explanation of depression has been that it is caused by too little of a brain chemical called serotonin, and that the antidepressant pills known as SSRIs work by topping that chemical back up. This review asks how much of that explanation the scientific evidence actually supports. The short answer is: the mechanism is plausible and partly supported in simplified settings, but the central claim has not been demonstrated in people. Large reviews have not found that people who are depressed reliably have less serotonin than people who are not, and the genetic studies that were supposed to link the serotonin system to depression have failed in very large samples. What has been shown is narrower and real: in people who have been depressed before, an experiment that temporarily strips the body of the raw material for making serotonin can bring symptoms back, and can reverse a recovery achieved on medication. The same experiment does nothing to the mood of people who have never been depressed. SSRIs do outperform dummy pills on average, but the difference is small, and no trial has compared an SSRI against a dummy pill designed to mimic its side effects while checking whether patients could tell which one they were taking. So the grade here is one step above a bare association: the evidence establishes a plausible mechanism with partial support, not a demonstrated cause. A grade like this is a statement about which experiments have and have not been done, not an accusation against any researcher, and it is not a claim that antidepressants do not work. It would move up if a properly blinded trial with a matched comparator showed the drug effect survives, or if an experiment could separate correcting a shortage from simply pushing the system in one direction. --- ## AT A GLANCE **Lowering serotonin availability reinstates symptoms in people who have been depressed before and does nothing to people who have not — the serotonin system is causally load-bearing for relapse, while the claim it is graded on, that a serotonin deficit causes depression and that SSRIs work by correcting it, remains Mechanistically plausible with partial system support (Γ).** | HSEF Level | Confidence | Role of Mechanism | Momentum | Stability | Divergence | |---|---|---|---|---|---| | **Γ** — Mechanistically plausible, partial system support | Moderate | reinforcing / stabilizing factor | Associational only | Stable | Severe overstatement | - **The experiment that decides it:** A blinded tryptophan-depletion challenge in drug-naive, currently depressed patients, run against a control arm matched for everything the procedure does except lowering tryptophan. - **What the field believes vs. what the evidence shows:** The headline Divergence uses the public and patient-facing reference class, which implies Ζ against an assessed Γ (+3, severe overstatement); measured instead against the current specialist literature, which no longer defends the deficit claim, the implied level is Δ and the divergence is +1 (mild overstatement). - **Best-supported rival model:** Neuroinflammation hypothesis (Γ) *Rubric: 07-psychiatry-neuromodulation · HSEF v5.4 · 2026-09-27 · Previous level: Β* *Entry type: restated from conjunction · Calibration entry: no* --- ## Scorecard detail | Field | Value | |---|---| | **Overall HSEF level** | Γ — Mechanistically plausible, partial system support | | **Confidence** | Moderate | | **Role of mechanism** | reinforcing / stabilizing factor | | **Role derivation step** | Step 5: amplifies or maintains a phenotype initiated upstream | | **Momentum (M)** | Associational only — resting on the absence of any registered depletion study or subjective-effect-matched SSRI trial as of 2026-09-27 | | **Stability (S)** | Stable — no pending result can be named that would move the letter; the framework's default applies | | **Divergence (D)** | headline uses the public and patient-facing reference class: implies Ζ; assessed Γ; gap +3 → Severe overstatement. Specialist-literature reference class: implies Δ; gap +1 → Mild overstatement | | **Primary rubric** | 07-psychiatry-neuromodulation.md | | **Secondary rubrics** | none | | **Subclaims scored** | 4 | | **Aggregation rule** | weakest-link — pre-registered | | **Other rules would return** | weakest-link Γ · primary-subclaim Γ · scope-matched Γ | | **Competing models evaluated** | 5 (including nominal) | | **Tournament rank of nominal** | 2 of 5 | | **Decisive-perturbation studies** | 5 | | **Load-bearing claims verified** | 21 verified / 0 unverified / 0 contradicted | | **Review mode** | revision | | **Review date** | 2026-09-27 | | **Evidence base** | canonical | ### Subclaim score matrix | ID | Subclaim (one sentence) | Type | Scope | Rubric | Score | Confidence | Governing cap or threshold | |---|---|---|---|---|---|---|---| | SC1 | Deficient serotonin neurotransmission is a proximate cause of the initiation or maintenance of major depressive disorder. | primacy | adults with MDD (DSM criteria); disease-class claim about initiation and maintenance | 07-psychiatry-neuromodulation.md | Γ | Moderate | Rubric Γ cap — "Mechanistic plausibility with in vitro, rodent, or unblinded human support caps at Γ" — with boundary clause 5 blocking Δ: the qualifying depletion experiment is null in never-depressed volunteers | | SC2 | SSRIs relieve the symptoms of major depressive disorder to a greater extent than placebo. | existence | adults with MDD in short-term randomised placebo-controlled trials | 07-psychiatry-neuromodulation.md | Γ | High | Δ threshold not met: no SSRI trial with "active comparator matched for acute subjective effect, blinding integrity measured and reported" exists [11][12] | | SC3 | SSRIs relieve MDD specifically by correcting a pre-existing serotonin deficit rather than by any other mechanism. | mechanism | SSRI responders with MDD; the mediating mechanism of response | 07-psychiatry-neuromodulation.md | Γ | Moderate | Rubric Γ cap; Δ not crossed because the available mediator manipulation tests serotonin-dependence, not deficit correction | | SC4 | Serotonin availability is causally required to maintain the recovered state in people with a history of MDD. | necessity | adults with a documented history of MDD, currently in remission, drug-free or medicated | 07-psychiatry-neuromodulation.md | Δ | Moderate | Δ crossed on "Depletion or challenge experiment in humans whose outcome is symptom trajectory, not a biomarker", positive result [2][4][5]; Ε not taken because the boundary is not cleanly mapped | **Aggregation rule (pre-registered before scoring).** Weakest-link — carried over unchanged from the input grading, which pre-registered it for the conjunctive form: the graded claim joins a causation clause to a mechanism-of-treatment clause with an "and", so the conjunction is only as strong as its weakest clause. **Derivation.** Weakest-link over SC1=Γ, SC2=Γ, SC3=Γ, SC4=Δ gives Γ. **What the other rules would return.** Weakest-link Γ, primary-subclaim Γ, scope-matched Γ. All three now agree, which was not true of the input grading, where the letter was rule-dependent: weakest-link and scope-matched returned Β while primary-subclaim returned Γ. Under primary-subclaim the primary is SC3, the clause that supplies the hypothesis's distinguishing content. Under scope-matched the central claim is advanced at disease-class scope, and the highest subclaim at that scope is Γ; SC4 is context-specific and is reported separately. --- ## Competing mechanistic models *A hypothesis is scored against its rivals, not in isolation. Ranked by strength of causal evidence.* | Rank | Model | Core claim | HSEF | Standing vs. the nominal | |---|---|---|---|---| | 1 | Neuroinflammation | MDD is initiated and maintained by elevated inflammatory mediators | Γ | Has the only human exposure followed by incident episodes in people who were not depressed [17] | | 2 | **Serotonin deficiency (nominal)** | Deficient serotonin causes MDD and SSRIs correct that deficiency | **Γ** | — | | 3 | Neuroplasticity / BDNF | MDD reflects impaired plasticity and reduced BDNF signalling | Γ | Strong rodent causality, no human perturbation; rodent-proxy rule caps at Γ [14] | | 4 | Psychosocial stress | Adverse stressors on a vulnerable substrate initiate MDD | Γ | Robust epidemiology, associational; interventional leg tests treatment not model | | 5 | Default mode network / circuit | Aberrant DMN connectivity produces MDD | Β | Observational imaging throughout; the rubric's Β cap names exactly this | ### Model 1 — Neuroinflammation hypothesis MDD is initiated and maintained by elevated peripheral and central inflammatory mediators that dysregulate monoamine metabolism and impair neuroplasticity. It explains the comorbidity of depression with inflammatory illness and the appearance of depressive episodes in people given cytokines therapeutically. Its strongest evidence is a meta-analysis of 26 prospective studies in which about a quarter of patients free of depression at baseline developed a major depressive episode during interferon-alpha therapy [17]. That is the only line in this tournament in which an exposure administered to people without depression is followed by incident episodes, which is why it ranks first. It caps at Γ nonetheless: the exposure is exogenous cytokine administration rather than the endogenous inflammatory state the model posits, the design is a cohort rather than a randomised perturbation of inflammation, and the remaining human evidence is case-control biomarker work to which universal cap 2 applies. ### Model 2 — Serotonin deficiency hypothesis (nominal) Deficient serotonin neurotransmission causes MDD and SSRIs relieve it by correcting that deficiency. Its strongest support is not, as usually presented, SSRI efficacy; it is the depletion literature. Blinded, within-subject tryptophan depletion reverses antidepressant-induced remission [5], lowers mood in drug-free remitted patients [2][4], and does nothing to the mood of volunteers who have never been depressed [1][4]. A challenge-PET study additionally reports reduced serotonin release capacity in antidepressant-free patients in an episode [6]. Against this: no consistent evidence that serotonin concentration or activity is lowered in depression [1], null candidate-gene and gene-by-stress findings in the largest samples assembled [7][8], and rival mediators with prospective predictive support [14]. The nominal ranks second, up from third in the input grading, which did not count the depletion-reversal literature at all. Ranking second does not rescue the graded claim, and the tournament does not change its HSEF level. ### Model 3 — Neuroplasticity / BDNF hypothesis MDD reflects impaired synaptic plasticity and reduced BDNF signalling, with antidepressants acting by restoring plasticity rather than by correcting a monoamine deficit. It explains the weeks-long delay between acute pharmacological action and clinical improvement, and the rapid action of ketamine. In rodents the behavioural actions of typical antidepressants are blocked by BDNF deletion and BDNF infusion into prefrontal cortex or hippocampus is sufficient to produce antidepressant-like effects; ketamine's synaptic and behavioural actions are blocked in BDNF null and BDNF Met knock-in mice [14]. Every causal element is in animals with behavioural-despair readouts, which the rubric's rodent-proxy rule caps at Γ, and rodent BDNF deletion does not by itself produce depressive behaviour [14]. It is demoted from rank 1 in the input grading, which credited it with "the ketamine natural experiment": ketamine's rapid effect in humans is a drug effect, not a perturbation of BDNF, and the BDNF dependence is shown only in mice. ### Model 4 — Psychosocial stress model MDD is primarily initiated and maintained by adverse psychosocial stressors acting on a vulnerable substrate, with biological changes downstream. The exposure-outcome association is among the most robust in psychiatry, and randomised psychotherapy trials show symptomatic benefit. Both legs fall short of a causal demonstration in this rubric's terms: the epidemiology is associational and universal cap 2 binds it, and the trials test a treatment rather than the causal model. Scored as a preliminary expectation rather than from a dedicated ledger. ### Model 5 — Default mode network / neural circuit hypothesis MDD reflects aberrant connectivity in the default mode network and frontal-limbic circuits, normalised by effective treatment. Case-control functional-imaging differences and their normalisation after treatment are exactly what the rubric's Β cap names — "Case-control imaging or biomarker differences" — and connectivity normalisation after treatment cannot distinguish cause from consequence of symptom change. Scored as a preliminary expectation rather than from a dedicated ledger. ### Tournament result On the shared evidence, neuroinflammation has the strongest human causal line bearing on the initiation of depression and the nominal has the strongest human causal line bearing on the maintenance of a treated state; the nominal ranks second of five. The study separating rank 1 from rank 2 is the interferon-alpha incidence meta-analysis [17]: an exposure given to people without depression is followed by incident episodes, which the depletion literature cannot match, because depletion does not induce depression in people who have never been depressed [1][4]. Four of the five models sit at Γ and the fifth at Β, and that is itself the finding — no model of depression in this tournament has an uncontested human causal demonstration, so the nominal's position says more about the state of the field than about the hypothesis. Ranking second does not raise the nominal's HSEF level; the tournament informs the role of mechanism and never modifies the level (Rule 9). --- ## HYPOTHESIS STATE EVALUATION FRAMEWORK (HSEF v5.4) ### Consensus provenance reconciliation — 9 October 2026 Case-specific Consensus discovery records were located and linked to this final edition. This supplements the review's provenance; it leaves the original assessed claim, grades and source counts intact. Provider summaries and full-text badges do not establish examination of original primary passages by this reviewer. Source access and incorporation retain the limits stated in the original methods and any dated research update. Recorded evidence artifacts: - `deliverables/prepublication-enrichment-2026-10-05/website-series/mdd-serotonin/consensus-browser-links.json` - `deliverables/prepublication-enrichment-2026-10-05/website-series/mdd-serotonin/consensus-browser.txt` - `deliverables/prepublication-enrichment-2026-10-05/website-series/mdd-serotonin/consensus-targeted-check.txt` The located artifacts do not state a reliable source-comparison count. No count is inferred. **Existing source reconciliation:** Yes; 7 distinct primary identifiers, abstract/metadata checks documented; excludes unrelated PMID from verification. No independently readable body verified; update explicitly avoids padding with meta-analyses. Matched primary identifiers in the preserved dated ledger: PMID 2184795, PMID 7944875, PMID 12869766, PMID 15289275, PMID 19755572, PMID 30845820, PMID 36635177. These are historical checks, not newly credited complete-paper examinations. Shared template wording about an amyloid-specific Deep follow-up is not credited as a case-specific run for this topic. The actual raw case records above establish provider use. No earlier adversarial approval exists for this case; this provenance supplement is not a new independent or blinded adjudication. No score or confidence change is made by this reconciliation. ### Provenance header > **Sources:** 17 (12 Tier A) · **Load-bearing claims:** 21 · **Verified:** 21 · **Unverified:** 0 · **Contradicted:** 0 · **Gate:** passed > **Adversarial review:** Generated by Sol 6.1; adversarially reviewed by Claude Opus 5.5 (Anthropic) on 2026-10-14. 3 corrections applied (see ADVERSARIAL REVIEW AND APPROVAL). Status: **Approved as revised.** ### Claim restatement (Step 0) **As indexed:** not in corpus index; fresh entry — serotonin deficiency hypothesis for major depressive disorder. **Entry type:** conjunction **Graded claim:** Deficient serotonin neurotransmission is a proximate cause of major depressive disorder, and SSRIs relieve major depressive disorder by correcting that serotonin deficit. Two changes were made to the input grading's restatement, both recorded rather than made silently. "The proximate cause" became "a proximate cause", because the definite article asserts uniqueness that the subclaim it maps to (SC1, "is a proximate cause") never claimed, and a restatement stronger than every subclaim under it is not the claim being scored. "Specifically by correcting that serotonin deficit" became "by correcting that serotonin deficit", with the specificity requirement moved wholly into SC3, where it is tested — leaving it in the parent sentence and also in SC2 double-counted it (pitfall 10). The harder question is whether this strong form is what proponents defend today, or a form nobody defends. Very few researchers now argue that a serotonin deficiency is the cause of depression; the current specialist literature argues at most that the serotonin system is implicated [2], and one of the most-cited reviews of antidepressant action treats the monoamine account as inadequate on its own [14]. The strong form is nonetheless the claim that carries the hypothesis's public influence: it is what direct-to-consumer advertising asserted [15], what most of the general public is reported to believe [1], and what patients most often hear from a clinician [16]. It is also falsifiable as stated, which the weaker "implicated" form is not. This review therefore grades the strong form and says plainly, here and in the divergence arithmetic, that the specialist field has largely moved off it. A reader who wants the weaker claim graded should read SC4, which is that claim's defensible core, and which reaches Δ. ### Hypothesis parsing The central causal claim is that reduced 5-HT availability at postsynaptic receptors impairs mood regulation in MDD, and that SSRIs relieve depression by blocking the serotonin reuptake transporter to restore adequate 5-HT signalling. The proposed mechanism covers both initiation and maintenance of MDD and is claimed at disease-class scope. The conjunctive note is where this edition differs most from the input grading. The input recorded that "the 'SSRIs correct the deficit' clause has never been isolated from the efficacy clause; no trial has independently manipulated serotonin status as an independent variable". The second half of that is false on the record: serotonin status has been manipulated as an independent variable in humans, under blind conditions, with the antidepressant left in place and symptom trajectory as the outcome, repeatedly since 1990 [4][5]. What has never been isolated is something narrower and more interesting — no experiment distinguishes correction of a pre-existing deficit from pharmacological displacement of a system that was not deficient in the first place. The hypothesis originated in pharmacological backward inference — SSRIs elevate synaptic serotonin and relieve depression, therefore depression results from serotonin deficiency — and was popularised through marketing that overstated its scientific grounding [15]. ### Field classification and rubric routing The primary rubric is 07-psychiatry-neuromodulation.md, whose scope names "antidepressant mechanism claims (monoamine, neuroplasticity, neuroinflammation)" and "the monoamine hypothesis" explicitly. The fit is good and no secondary rubric is required; the rival models are governed by the same rubric. No index note was applied, because none exists. The input grading recorded two "index adjudication notes" — that the functional-unblinding cap applies to all inert-placebo SSRI RCTs, and that tryptophan depletion failure in healthy subjects does not cross Δ. Neither appears in `hypothesis-index.md`, which carries no serotonin or monoamine entry at all; its only psychiatry entry is the psilocybin dual-phase hypothesis. The second proposition is sound, but it is boundary-check clause 5 of the framework, not an index note, and it is applied here as such. The first is an extension of a rubric rule beyond the interventions that rule enumerates, and is treated below as the open adjudication it is rather than as settled institutional memory. A stub index entry is recommended in the reviewer notes. ### Evidence inventory The evidence base spans seventeen sources across six design classes: an umbrella review of the serotonin literature and the published critique and reply it generated; a meta-analysis of monoamine depletion studies and the primary depletion-reversal experiment; a challenge-PET study of serotonin release capacity; two very large null genetic studies; four syntheses of antidepressant trial data including the active-placebo comparison; a patient-level re-analysis of severity dependence; a mechanistic review; two sources bearing on how the claim is communicated; and one meta-analysis used only in the tournament. **For SC1 (a serotonin deficit causes MDD).** The umbrella review [1] examined six areas — serotonin and 5-HIAA in body fluids, 5-HT1A receptor binding, transporter binding, depletion studies, transporter-gene association and gene-by-stress interaction — and concluded that there is no convincing evidence that depression is associated with, or caused by, lowered serotonin concentration or activity. It is the strongest single item against SC1 and it is also contested: a 35-author critique in the same journal and issue argues the conclusion is overstated, citing selective reporting of the depletion literature, omission of meta-analyses of circulating tryptophan, and over-simplified interpretation of receptor and transporter imaging [2]; the review's authors replied maintaining their conclusion [3]. This review records both and adopts neither wholesale: the umbrella review's own certainty ratings, not its headline, are what bear on the letter. The genetic leg is the one place where the evidence is not merely inconsistent but null at scale — no significant stress-by-5-HTTLPR interaction across 31 datasets and 38,802 subjects [7], and no support for any of 18 historical candidate genes in samples up to 443,264 [8]. The one positive direct measurement is a challenge-PET study reporting reduced serotonin release capacity in 17 antidepressant-free patients against 20 controls [6]; its endpoint is receptor binding, not symptoms, so universal cap 4 and the rubric's acute-pharmacodynamic rule both apply, and its sample is small, age-imbalanced and unreplicated. **For SC2 (SSRIs outperform placebo).** The network meta-analysis of 21 antidepressants across 522 double-blind trials and 116,477 participants found every drug more efficacious than placebo, with odds ratios from 2.13 (95% CrI 1.89–2.41) to 1.37 (1.16–1.63) [9]. The effect is small on the sum score: 1.80 HRSD points in the complete FDA registration dataset for the four new-generation antidepressants with full data (35 trials, 5,133 patients), reaching the NICE three-point criterion only in trials whose mean baseline exceeded about 28, and doing so because placebo response declines with severity rather than because drug response rises [10]; a separate synthesis that included 131 trials in 27,422 participants found −1.94 HDRS points, pooled over the 49 of those trials reporting the outcome, below its own pre-defined threshold, with every trial at high risk of bias [11]. That severity dependence is itself contested: a patient-level, item-based re-analysis of 8,262 participants from 28 trials found no difference between non-severe and severe depression in the reduction of depressed mood, and attributes the apparent dependence to the 17-item sum score [13]. The decisive structural fact for the level is simpler and is reported in the same synthesis: none of the 131 trials used an active placebo [11]. The nearest thing the field has is a Cochrane review of nine antidepressant trials against active placebos in 751 participants, pooling to 0.39 standard deviations and falling to 0.17 (0.00–0.34) when one strongly positive trial is removed, whose authors concluded that unblinding may inflate efficacy estimates from inert-placebo trials [12] — a tricyclic-dominated, unregistered set that measured no blinding integrity. **For SC3 (SSRIs work by correcting the deficit).** The mechanistic review [14] argues that acute monoamine potentiation cannot explain a clinical onset measured in weeks, and advances neuroplasticity and early change in negative affective bias as candidates; it reports that early drug-induced change in emotional processing predicts later clinical response and behaves as a specific treatment effect rather than as a general expectancy mediator [14]. It does not say, and the input grading's evidence card wrongly had it saying, that no trial has manipulated serotonin as an independent variable; it says the opposite of the strong anti-serotonergic reading, noting that all currently licensed antidepressants are believed to relieve depression by increasing serotonin or norepinephrine availability, at least initially [14]. **For SC4 (serotonin availability is required to maintain recovery).** This subclaim is new in this edition and it carries the hypothesis's only positive field-valid perturbation. In the primary experiment, 21 patients in antidepressant-induced remission underwent double-blind, placebo-controlled crossover depletion; total and free plasma tryptophan fell 87% and 91%, 14 of the 21 relapsed, mood returned within 24 to 48 hours of resuming normal intake, and the control drink produced no significant behavioural effect [5]. The meta-analysis of the wider depletion literature is stratified and its strata are the finding: no mood decrease in healthy controls, a slight decrease in healthy controls with a family history of MDD, a moderate decrease in drug-free patients with MDD in remission, and relapse in remitted patients taking serotonergic antidepressants [4]. The critique reports that for the drug-free remitted stratum the pooled effect was Hedges' g = −1.9 (95% CI −3.02 to −0.78) across 8 samples, falling to −1.06 (−1.83 to −0.29) with one outlier removed [2] — a figure the umbrella review did not present, and the omission of which is the critique's strongest specific charge. Two caveats travel with all of this and are why SC4 sits at Δ rather than higher: several of those samples come from overlapping research groups, and the umbrella review notes that depletion also alters amino-acid balance, neurotrophic signalling and cerebrovascular parameters, so it is not established that the procedure manipulates serotonin specifically [1]. **Decisive-perturbation studies.** Five studies are marked decisive perturbation: - **[4] Ruhé et al. 2007** — meta-analysis of human monoamine depletion; endogenous: yes; direction: supports SC4, null for SC1's population; favours Model 1 within a defined context. Decisive because it is the field's only stratified synthesis of a human manipulation of the nominal mediator, and because its negative stratum is the internal comparison that bounds the positive ones. - **[5] Delgado et al. 1990** — blinded within-subject depletion in antidepressant-induced remission; endogenous: yes; direction: supports SC4; favours Model 1. Decisive because it is the single cleanest human mediator manipulation in the literature: the drug stays, the mediator goes, symptoms return, and they return again to baseline when it is restored. - **[9] Cipriani et al. 2018** — network meta-analysis of antidepressant RCTs; endogenous: no; direction: supports SC2; favours Models 1, 2 and 3 equally, which is the point. Decisive for the efficacy clause; silent on mechanism. - **[11] Jakobsen et al. 2017** — systematic review with trial sequential analysis; endogenous: no; direction: null for SC2's clinical-significance threshold. Decisive because it establishes the structural fact that no trial in the SSRI base used an active placebo. - **[12] Moncrieff, Wessely & Hardy 2004** — Cochrane review of antidepressants against active placebos; endogenous: no; direction: null. Decisive because it is the only attempt in the record to run the comparator the rubric asks for, and because its sensitivity analysis is the field's best estimate of how much of the drug-placebo gap survives partial matching. **Verification.** All 21 registered claims were checked against PubMed records on 2026-09-27; full text was read for sources 1, 2, 10 and 14 and the accessible portion of source 3, and the remaining records were verified at abstract and metadata level, with sources 4, 5, 6, 13 and 17 behind paywalls. Every ledger row and every claim of the input grading was independently re-checked rather than accepted. The one figure in this review that was not read in the record that generated it is the depletion subgroup effect size in claim C4: it is reported in the full text of the critique [2], which attributes it to the meta-analysis it is criticising [4], whose own full text is paywalled. The claim is written as a statement about what the critique reports, which is what was verified, and the wording difference between the two sources — the critique calls the stratum "depressed people not taking antidepressants" where the meta-analysis's abstract calls it drug-free patients with MDD in remission — is recorded rather than smoothed over. **Factual issues found.** Four of the input grading's eight PMIDs resolve to unrelated papers, and one ledger row is a composite of three separate records. The full table is in `redteam/identifier-audit.md`; in summary, the Ruhé depletion meta-analysis was cited as PMID 17389903 (a fronto-limbic MRI study in the same journal and volume), the Harmer review as PMID 28237360 (a paper on potassium channels and blood pressure), the Delgado experiment as PMID 2268388 (a flow-cytometry methods paper) — the last being the input's only Tier A source — and the Jakobsen review under the title and authors of a 2017 BMC Psychiatry paper, the journal, volume and pages of nothing, the DOI of a different 2019 BMJ Evidence-Based Medicine paper by an overlapping author group, and a PMID belonging to a veterinary article about mandibular malformation in calves. In every case the DOI was right or nearly right and the PMID was wrong, which is the signature of identifiers assembled rather than retrieved. The input grading nonetheless recorded all eight claims as "verified against PubMed abstracts and metadata on 2026-09-27". Four of its eight claims were also materially wrong, and three of those were load-bearing. The assertion that no trial has independently manipulated serotonin correction, attributed to the Harmer review, is not in that review and is false as a statement about the literature; it was the stated trigger for the SC3 cap. The assertion that SSRI trials "did not include blinding-integrity assessment", attributed to the Jakobsen review, is not in that review either, and could not be verified against any source; it was the stated trigger for the SC2 cap. The Delgado experiment was credited with a healthy-control and drug-naive comparison it does not contain. And the three "verbatim" rubric quotations used to fire the SC2 and SC3 caps, together with all three "field-valid causal perturbation" definitions in the boundary check, appear nowhere in 07-psychiatry-neuromodulation.md; they are paraphrases written in the register of quotations. The score here was derived on the corrected understanding throughout, and the full claim-by-claim table is in `redteam/claim-audit.md`. ### Cap rules applied **SC1 — a serotonin deficit causes MDD** Universal cap 2 — "Pure association evidence — however large, however replicated — cannot exceed Β" — fires on the associational leg and is recorded as binding there: metabolite and plasma studies, receptor and transporter imaging, and the candidate-gene and gene-by-stress literature [1][6][7][8] cannot carry SC1 above Β however many of them accumulate. It is **not** recorded as the subclaim's ceiling, and this is the single point on which this edition departs from the input grading. SC1's evidence base is not association-only. It contains a human depletion-challenge programme with symptom-trajectory endpoints [4] and an unblinded human pharmacological challenge comparing serotonin release capacity in patients and controls [6]. The framework is explicit that a cap "binds the inference from a single evidence class, not from a set that was assembled to cover each other's weaknesses", and pitfall 21 names the error of identifying the weakest class, applying its cap, and assigning the ceiling to the whole claim. What binds SC1 instead is the rubric's own Γ cap: "Mechanistic plausibility with in vitro, rodent, or unblinded human support caps at Γ." Its unblinded-human limb describes the sourced evidence: an open-label challenge-PET comparison with a positive result [6], and the pharmacology of the drugs themselves, all of which are believed to act at least initially by increasing serotonin or norepinephrine availability [14]. The blinded depletion programme adds mechanistic plausibility but returns its positive results outside SC1's population [4], so it supports the mechanism without crossing Δ for this subclaim. No rodent evidence is relied on for SC1: none is in the ledger, and the letter does not need it. Universal cap 4 — "Biomarker change without functional/phenotypic change caps at Γ" — fires additionally on the challenge-PET release-capacity finding [6], as does the rubric's acute-pharmacodynamic rule, "Acute molecular, occupancy, or imaging change with no link to symptom trajectory". Ceiling: Γ. **SC2 — SSRIs outperform placebo** The rubric's Γ cap binds. The trial base is large, randomised and positive, but it consists entirely of inert-placebo comparisons whose blinding was never verified [9][11]. The rubric classes "Inert-placebo RCT that does report blinding-integrity data" as Moderate evidence; these trials do not reach even that description, and the Δ threshold asks for something none of them is. The functional-unblinding hard rule is recorded as **not triggered as written**, against the input grading, and the reason is given rather than assumed. The rule reads: "**Functional-unblinding rule.** Inert-placebo trial of an intervention with unmistakable acute subjective effects — psychedelics, ketamine, entactogens, suprathreshold stimulation, ECT — with no blinding-integrity check reported. Such a trial did not control what it claims to control." SSRIs are not among the interventions it enumerates, and their side-effect profile, while noticeable, is not the unmistakable acute psychoactive experience the rule was written for. Extending it by analogy would be defensible; doing so silently, behind a quotation that does not exist, is not. The concern the rule expresses is live here in any case, and it is handled where the evidence actually speaks to it: in the Δ threshold and in the active-placebo comparison [12]. The ceiling is Γ either way, which is worth stating plainly — the cap attribution changed and the letter did not. **SC3 — SSRIs work by correcting the deficit** The rubric's Γ cap binds. The mediation-by-correlation rule — "**Mediation-by-correlation rule.** Candidate mediator correlated with outcome, never manipulated" — is recorded as **not triggered**, because the mediator has been manipulated in humans with the intervention intact [4][5]. The input grading fired this rule on the strength of an assertion that no such manipulation exists. SC3 is nonetheless held at Γ, for a better reason: the manipulation that exists answers a different question. Depletion reversing a drug-induced remission is exactly as consistent with a system pushed above its own baseline by a drug as with restoration of a prior deficit, and nothing measured in any of these studies distinguishes them. **SC4 — serotonin availability maintains recovery** No cap triggers. The subclaim does not rest on in vitro, rodent or unblinded human support; it rests on a blinded, within-subject human mediator manipulation with a symptom endpoint [2][4][5]. **Cap precedence.** Two collisions, both recorded rather than resolved privately. First, universal cap 2 against rubric 07's Γ cap and Δ threshold on SC1: the cap governs the inference from the associational class to the disease-class claim, the rubric threshold governs the subclaim within the system its perturbations were run in, and by the framework's rule — "A perturbation crosses a threshold within the system it was actually run in. A universal cap governs the inference from that system to the hypothesis's claimed scope" — the rubric's Γ cap is what fixes SC1. Second, the rubric's Δ sentence requires "a pre-registered primary outcome", and the depletion literature that satisfies its field-valid perturbation list predates trial registration. That clause is read here as attaching to the trial-design limb of the sentence rather than to the mediator-manipulation limb; the contrary reading would hold every depletion study below Δ on a publication-era technicality, which the framework's own upper-ladder section warns against. Under the contrary reading SC4 would be Γ and the overall letter would be unchanged. Both readings are flagged for the rubric's maintainer. ### Β/Γ/Δ boundary check **SC1.** The rubric's field-valid causal perturbation list includes "Depletion or challenge experiment in humans whose outcome is symptom trajectory, not a biomarker". Qualifying evidence is present: the acute tryptophan depletion literature [1][4]. Γ-cap hard rules: the acute-pharmacodynamic rule is triggered by the challenge-PET finding [6]; the rodent-proxy rule — "Forced-swim, tail-suspension, sucrose-preference, or head-twitch alone; head-twitch is pharmacodynamic, not therapeutic" — is not triggered, because no rodent evidence is relied on for this subclaim; the mediation-by-correlation, open-label and unblinded-rater rules are not triggered. The dependency is disease-relevant, not merely biomarker-relevant. Clause 5 decides it: the qualifying perturbation's result is **not positive in the population this subclaim needs**. Depletion produces no mood effect in most healthy volunteers, the family-history signal is rated very low certainty and was not confirmed in subsequent studies [1][4], and no depletion experiment in drug-naive, currently depressed patients is on the record. A negative field-valid perturbation does not cross Δ. Ceiling: Γ. **SC2.** The relevant field-valid perturbation is "Active comparator matched for acute subjective effect, blinding integrity measured and reported". Qualifying evidence is **absent**: none of the 131 trials in the largest systematic review used an active placebo [11], and the nine active-placebo trials that exist are tricyclic-dominated, unregistered and measured no blinding integrity [12]. The functional-unblinding rule is not triggered as written; the open-label and unblinded-rater rules are not triggered. The dependency is phenotype-relevant. The perturbation result is positive — SSRIs do beat inert placebo, which is why SC2 sits at the Γ ceiling rather than below it — but Δ fails on the comparator, not on the result. Ceiling: Γ. **SC3.** The relevant field-valid perturbation is "Mediator manipulation with the intervention intact: antagonist blockade of the acute experience with the drug still given, or dosing under general anaesthesia". Qualifying evidence is present in form [4][5]. No Γ-cap hard rule is triggered. The dependency is phenotype-relevant. The result is not positive for the proposition SC3 states: the manipulation shows the drug response depends on serotonin availability, while SC3 claims the drug works by correcting a pre-existing deficit, and the deficit premise fails at SC1. For the proposition SC3 actually asserts, no qualifying perturbation has been run. Ceiling: Γ. **SC4.** Same field-valid perturbation as SC1: "Depletion or challenge experiment in humans whose outcome is symptom trajectory, not a biomarker". Qualifying evidence is present [2][4][5]. No Γ-cap hard rule is triggered: the endpoint is symptom trajectory, the design is blinded and placebo-controlled within subject, the raters were not the mediator's advocates, and the evidence is human. The dependency is phenotype-relevant. The perturbation result is positive, and there is an internal comparison group in which the effect is absent. Ceiling: Δ or higher. ### Threshold testing **SC1 — Δ threshold.** Requirement: "Requires a controlled human interventional study with a credible active comparator, or a mediator manipulation in humans, with a pre-registered primary outcome." Satisfied: no. The manipulation exists but returns a null in the population the subclaim needs [1][4]. What would cross it: a blinded depletion or challenge experiment in drug-naive people with a current depressive episode, or in never-depressed volunteers, producing clinically meaningful worsening against sham, with a manipulation check showing the intervention changed serotonergic transmission specifically. **SC2 — Δ threshold.** Same requirement. Satisfied: no. What would cross it: a pre-registered SSRI trial against a comparator matched for acute subjective effect, with allocation-guess data collected from participants and raters and reported, showing the drug effect survives adjustment for guessed allocation. **SC3 — Δ threshold.** Same requirement. Satisfied: no. What would cross it: an experiment separating deficit correction from pharmacological displacement — response stratified by an independently measured baseline index of serotonergic function, or a design in which the serotonergic effect is delivered to people shown not to be deficient. **SC4 — Δ threshold.** Same requirement. Satisfied: yes, on the mediator-manipulation limb. Fourteen of 21 remitted patients relapsed under blinded depletion and recovered on resumption, with no effect of control testing [5]; the meta-analysed drug-free remitted stratum shows a moderate to large mood decrease [2][4]; and never-depressed volunteers show nothing [1][4]. The one element of the threshold sentence this literature does not meet is pre-registration, addressed under cap precedence. **SC4 — Ε threshold.** Requirement: "Requires causal support restricted to a defined context — one diagnosis, one severity stratum, one treatment-resistant subgroup, or only with the accompanying psychotherapeutic scaffold." Satisfied: partial, and Ε is not taken. The context is defined and determinable at enrollment — a documented history of MDD with current remission — and the never-depressed stratum marks where the effect stops. What holds it back is the quality of the mapping rather than its existence: the strata come from heterogeneous study sets rather than one design with internal controls, several drug-free-remitted samples share research groups, the outlier sensitivity is large [2], recent crossover studies in depressed patients showed no convincing effect [1], and the manipulation's specificity for serotonin is contested. What would cross it: independent replication of the drug-free remitted result, and a manipulation check separating depletion's serotonergic from its non-serotonergic effects. **Generalization.** Generalization was tested and it fails: the depletion effect does not extend to people who have never been depressed [1][4]. Pitfall 16 warns that failed generalization is Ε's definition rather than a reason to sit below it, and that the question is whether the boundary was *mapped* or merely unexamined. Here it was examined, and it is partly mapped — which is why SC4 sits at Δ with Ε named as the next rung and its conditions stated, rather than at Γ. The pitfall cuts both ways, and this is the direction in which it cuts here. ### Falsification patterns and pitfalls The rubric lists seven falsification patterns for this field. **None is met**, and saying so is more informative than the input grading's list, which recorded three "patterns" that are not in the rubric — they were cap names, entered with `downgrade_applied: true`, which double-counted the caps as downgrades. - **The effect disappears against a subjective-effect-matched comparator** — not met. No subjective-effect-matched SSRI comparison exists. The partially matched active-placebo set attenuates the effect to 0.17 (0.00–0.34) on sensitivity analysis but does not eliminate it, and is tricyclic-dominated [12]. - **Allocation guessing is near-perfect and outcome tracks guessed allocation** — not met; allocation guessing has not been measured in this trial base at all, which is a gap rather than a falsification. - **Blocking the acute experience, or dosing under anaesthesia, leaves the effect intact** — not applicable to an agent with no acute subjective experience to block. - **A non-subjective congener reproduces the therapeutic effect** — not applicable in the same way. - **Blinded on/off discontinuation produces no symptom return** — not met, and its contrary is observed: blinded removal of the mediator produces symptom return [4][5]. This is the pattern that would have refuted the serotonergic account most directly, and it is the one the evidence goes against. - **Measured expectancy predicts outcome as well as the nominal mediator** — not met; expectancy has not been measured against the mediator in these trials. The one relevant datum points the other way: early drug-induced change in emotional processing behaves as a specific treatment effect rather than as a general expectancy mediator [14]. - **Multi-site replication with blinded central raters fails to reproduce a single-site effect** — not met for SSRI efficacy, which replicates across hundreds of sites [9]. No downgrade is applied from this section. **Pitfalls from `pitfalls-library.md` and the rubric:** - **Pitfall 21 — a universal cap re-applied to a convergent evidence set** (SC1): the error this revision corrects. Cap 2 was applied to a subclaim whose base includes a human mediator manipulation, and returned Β. - **Pitfall 1 — mistaking presence for primacy** (SC1): serotonin-system differences in imaging and metabolite studies are read as causal deficiency when they are equally consistent with consequence, with medication history, or with nothing [1]. - **Pitfall 17 — negative field-valid perturbation read as a crossed threshold** (SC1): avoided here and correctly avoided in the input grading. The healthy-volunteer depletion null is of exactly the right design and is recorded as blocking Δ, not as supporting it. - **Pitfall 10 — mechanism conflation** (SC2): "specifically" has been removed from the efficacy subclaim and returned to SC3. - **Pitfall 13 — therapeutic effect misread as mechanistic proof** (SC3): the hypothesis's historical warrant is backward inference from SSRI efficacy, which constrains mechanism only weakly [14]. - **Pitfall 11 — intervention specificity assumed rather than demonstrated** (SC4): depletion also alters amino-acid balance, neurotrophic signalling and cerebrovascular parameters, and it is not established that it reliably lowers brain serotonin [1]. This is the main reason SC4 is Δ at Moderate confidence rather than higher. - **Pitfall 16 — failed generalization mistaken for weak evidence** (SC4): applied in both directions, as set out above. - **Pitfall 18 — causal-evidence tallies that do not survive an audit** (SC1): see below. - **Pitfall 19 — unlinked claims** (throughout): the input grading's tournament and upper-ladder sections name a ketamine trial, BDNF knockout work, interferon-alpha depression, genetic instruments and SSRI dose-response, with no ledger row behind any of them. Those that survive here carry a source; those that could not be resolved were removed or restated generically. **Causal-evidence tally.** The input grading audited its own tally to zero creditable causal items. Recounting against the cap rules returns **two**: the depletion meta-analysis [4] and the depletion-reversal experiment [5], both positive, endogenous, human and phenotype-relevant — and both bearing on SC4 rather than on SC1, which is why the aggregate does not move with them. Items rejected: the challenge-PET release-capacity finding [6], under cap 4 and the acute-pharmacodynamic rule; the three antidepressant trial syntheses [9][11][12], which are decisive perturbations for the efficacy clause but not perturbations of the endogenous mechanism; the genetic studies [7][8], observational and null; and the interferon meta-analysis [17], which is a causal item for a rival model and is non-randomised. ### The steelman #### Upper-ladder check The final level is Γ (below Ζ); the check is required. **Mendelian randomization** — available and null: the obvious instrument, 5-HTTLPR and the historical candidate-gene set, fails in the two largest analyses available [7][8]; a null and weak instrument cannot carry the claim upward and cannot be read as refuting it either. **Natural experiment** — not available: no population-scale exposure alters serotonergic transmission cleanly and independently of the psychosocial and medical determinants of depression. **Cross-population dose-response** — available and unhelpful: SSRI dose-response is flat above the therapeutic range and indexes drug exposure rather than serotonergic state. **Withdrawal and rechallenge** — available, and counted: the depletion-reversal paradigm is this design, and it is what carries SC4 to Δ. It cannot carry the graded claim further, because withdrawing serotonin from a drug-elevated state cannot show the pre-drug state was deficient. **Therapeutic perturbation against a hard endpoint** — available and bearing on SC2 only: the trials perturb the system against a symptom endpoint with no matched comparator [11][12] and with the mechanism unidentified. Convergence across classes is not present: the classes here do not cover each other's weaknesses, they fail in the same place, which is the deficiency premise. No universal cap has been re-applied to a convergent set; the correction in the opposite direction — a cap wrongly applied to a mixed set — is what moved SC1. #### The case for one rung higher (Δ) SC4 crosses Δ on the rubric's own field-valid perturbation list, and a reader could reasonably argue that the aggregation is being used to bury it. The hypothesis's second clause predicts that removing serotonin should remove the benefit; that prediction was tested in humans under blind conditions, and it held [2][4][5]. Add reduced release capacity in unmedicated patients [6], the lower plasma tryptophan reported in the critique literature [2], and the fact that every licensed antidepressant acts at least initially on monoamine availability [14], and the picture is of a system that is perturbed in depression and causally load-bearing for recovery. On that reading the weak clause is not the mechanism but the deficiency framing, and the hypothesis restated as "serotonergic transmission is causally involved in depressive states" would sit at Δ or Ε. #### The case for one rung lower (Β) This is the input grading's answer and it deserves its full weight. Every direct measurement of serotonin in depression is associational, and the largest synthesis of them finds no consistent signal [1]. The genetic instruments are null in the largest samples ever assembled for the question [7][8]. The depletion literature's own meta-analysts conclude that depletion studies "fail to demonstrate a causal relation" and instead mark a vulnerability trait [4]. And the manipulation may not be serotonin-specific at all [1]. On this reading, the human challenge and pharmacological evidence that lifts SC1 to Γ supports serotonergic involvement in mood regulation in general, not the deficiency claim actually being graded, and importing it is charity in scoring — which Rule 3 forbids. #### Rejected, and why **Rejected the case for Δ; accepted Γ.** The blocker is the aggregation rule operating on a genuinely conjunctive claim. SC4 is not the graded claim; it is a necessary condition of the graded claim's second clause, and each of SC1, SC2 and SC3 independently fails the Δ threshold — SC1 on boundary-check clause 5, the qualifying perturbation being null in never-depressed volunteers [1][4]; SC2 on the absence of any comparator matched for acute subjective effect with measured blinding integrity [11][12]; SC3 on the absence of any experiment distinguishing deficit correction from pharmacological displacement. **Rejected the case for Β; accepted Γ.** The blocker is the cap-precedence rule. Rubric 07's Γ cap, not universal cap 2, is what binds a subclaim whose evidence base includes a human mediator manipulation with a symptom endpoint; treating one associational leg as the ceiling for the whole is pitfall 21, and it is the specific error that produced Β. The charity objection is met not by generosity but by the rubric's own text: "Mechanistic plausibility with in vitro, rodent, or unblinded human support caps at Γ" describes this evidence base, and Β — "Association-only psychiatric work" — does not. ### Role-of-mechanism derivation The role is **reinforcing / stabilizing factor**, matched at Step 5. Step 1 does not match: nothing shows the serotonin change is produced by the depressive state rather than producing it. Step 2 — epiphenomenon — does not match, and this is where the edition departs from the input grading, which selected it. Step 2 requires "no demonstrated causal dependency in *either* direction, only co-occurrence", and a dependency has been demonstrated in one direction: lowering serotonin availability reinstates symptoms in people with a history of depression and reverses antidepressant-induced remission under blind conditions [4][5]. Step 3 fails on all three of its tests — removing serotonin does not prevent depression, engaging it does not initiate the disorder in never-depressed people [1][4], and nothing shows it is rate-limiting. Step 4 fails: serotonin is not shown to be required for the phenotype to arise. Step 5 therefore matches: the mechanism maintains and modulates a state initiated upstream. The nominal's second place in the tournament played no part in this derivation. ### Momentum, Stability, Divergence **Momentum — Associational only.** The artifact is the absence of the decisive designs from the registries. A ClinicalTrials.gov search on 2026-09-27 for tryptophan-depletion studies in depression returns only completed observational records registered in the early 2000s; no registered SSRI trial uses a comparator matched for acute subjective effect, and where the active-placebo design does appear in current depression trials it is midazolam against ketamine rather than anything serotonergic. The last systematic synthesis of the depletion literature remains the 2007 meta-analysis, a gap the umbrella review itself notes [1]. Output since 2022 has been dominated by re-analysis and by the umbrella-review controversy [1][2][3]. Causal work on this hypothesis has stalled while commentary has grown. **Stability — Stable.** This is a change from the input grading's Eroding, and the framework's rule forces it: "If no trigger can be named, the answer is **Stable** by definition." No pending result can be named that would move the letter. Replication of the reduced-release-capacity finding [6] would strengthen an association that cap 4 already holds at Γ. A further umbrella review would re-synthesise the same primary literature. The Moncrieff–Jauhar exchange, which the input grading named as its trigger, has already been published in full [2][3] without adding data. A named trigger would be either of the two decisive experiments proposed below; neither is registered. **Divergence — Severe overstatement.** Field framing implied level: **Ζ**, measured against how the claim is presented outside the specialist literature, which is where this hypothesis has its influence. The umbrella review records that surveys indicate most of the general public believes it is established that depression is caused by a chemical imbalance, and that many general practitioners share that view [1]; direct-to-consumer advertising asserted serotonin correction outright [15]; and hearing the explanation from a healthcare provider is the one source that uniquely predicts endorsing it [16]. A reader taking that framing at face value would infer multiple converging causal lines, which is Ζ. Assessed level: Γ. Gap: +3 rungs. Divergence = Severe overstatement. The choice of reference framing decides that answer, so it is stated rather than assumed. Measured against the **current specialist literature** the arithmetic is different and much smaller: contemporary reviews do not defend the deficiency claim — one of the most-cited treats the monoamine account as inadequate to explain antidepressant action [14], and even the 35-author defence of the field's research programme claims only that the serotonin system is "implicated" [2] — so specialist framing implies about Δ, giving +1, mild overstatement. The severe value therefore measures the gap between the evidence and clinical or public communication, not a gap between the evidence and what the field now argues in print. That distinction matters for how this review should be read: the overstatement being measured is largely historical and popular, and the field has substantially corrected itself in the specialist literature without the correction reaching the consulting room. ### Final adjudication The overall HSEF level is **Γ (Mechanistically plausible, partial system support)**, confidence **Moderate**, role **reinforcing / stabilizing factor**. Weakest-link aggregation over SC1=Γ, SC2=Γ, SC3=Γ, SC4=Δ gives Γ, and all three aggregation rules now return the same letter. Moderate rather than High confidence reflects demonstrable reproducibility problems in the serotonin evidence base — a candidate-gene programme that collapsed in the largest samples [7][8], imaging findings inconsistent across reviews and confounded by medication history [1], a heterogeneous depletion literature partly concentrated in overlapping groups [2][4], and a single direct measure of release capacity in 17 patients [6]. The level is not higher — not Δ — because the graded claim is a conjunction and each of its three disease-class clauses independently fails the Δ threshold. The specific missing experiment for the causation clause is a blinded depletion or challenge study in drug-naive, currently depressed patients; for the efficacy clause it is a trial against a comparator matched for acute subjective effect with blinding integrity measured; for the mechanism clause it is any design that separates correcting a deficit from displacing a system that was not deficient. The level is not lower — not Β — because SC1's evidence base is not association-only. A human manipulation of the nominal mediator, with the intervention left in place and symptom trajectory as the endpoint, has been run repeatedly under blind conditions [4][5], and an unblinded human challenge study and the pharmacology of the drugs themselves lend the mechanism partial human support [6][14]. "Association-only psychiatric work caps at Β" is not a description of this evidence base; "mechanistic plausibility with in vitro, rodent, or unblinded human support caps at Γ" is. What would move it up: a positive depletion result in drug-naive depressed patients would cross Δ for SC1; an active-comparator trial with measured blinding integrity would cross Δ for SC2; stratification of response by an independently measured baseline serotonergic index would cross Δ for SC3, and all three together would put the conjunction at Δ or above. What would move it down: a matched-control depletion study showing the mood effect survives without tryptophan depletion — that is, that the paradigm is not acting through serotonin — would remove SC4's positive perturbation and return the whole claim to the associational base, which is Β. ### What this review changed **Β → Γ.** SC1 moved from Β to Γ because universal cap 2 had been applied to a subclaim whose evidence base is not association-only; the cap binds the biomarker and genetic leg, and rubric 07's Γ cap binds the subclaim (pitfall 21; framework, cap precedence). Under weakest-link the aggregate followed. A fourth subclaim was added at Δ — that serotonin availability is required to maintain the recovered state — which is the proposition the input grading omitted and where the hypothesis's only positive human perturbation lives; it does not move the aggregate but it corrects a causal tally the input had audited to zero. SC2 and SC3 held at Γ with their governing rules replaced, both having been fixed by rubric caps quoted in words that appear nowhere in the rubric, and SC3's cap being contradicted by the record. Confidence moved High → Moderate, role moved epiphenomenon → reinforcing / stabilizing factor, and Stability moved Eroding → Stable, each for the reason given in its section. The level was **not** derived before the prior score was consulted: this pass was commissioned as a revision and the prior letter was supplied in its brief, so Rule 12's anchoring safeguard could not be met. That is disclosed rather than asserted away; the mitigation used was to derive each subclaim from the rubric text and the retrieved records first and compare letters afterwards. Where this edition disagrees with the input grading, the input's case is written out in the steelman and in `PROCESS.md` rather than removed. --- ## Proposed decisive experiments ### Experiment 1: Depletion challenge in drug-naive, currently depressed patients with a specificity control - **Experiment:** Blinded, within-subject acute tryptophan depletion versus a balanced amino-acid control in adults with a current, first-episode depressive episode who have never taken a psychotropic drug, with a parallel arm receiving a procedure matched for amino-acid load and cerebrovascular effect but not for tryptophan. - **Readout:** Change in a clinician-rated depression score at 6 and 24 hours; plasma tryptophan to large-neutral-amino-acid ratio as the manipulation check; a serotonergic neuroendocrine or imaging index to show the manipulation acted where it is claimed to. - **Critical control:** The matched non-tryptophan-depleting arm, which separates serotonergic depletion from the procedure's other effects — the specificity gap that currently caps SC4 at Δ. - **Result that would falsify the nominal model:** No difference between depletion and control in drug-naive depressed patients with a confirmed manipulation check, which would leave SC1 at Γ permanently and locate the depletion effect wholly in the previously-treated or previously-depressed state. - **Result that would support the nominal model:** Clinically meaningful worsening under depletion but not under the matched control, which would cross Δ for SC1 and be the first positive field-valid perturbation for the causation clause. ### Experiment 2: SSRI against a subjective-effect-matched comparator with measured blinding integrity - **Experiment:** Pre-registered randomised trial of an SSRI against an active placebo chosen to reproduce its early side-effect profile, with allocation-guess and confidence ratings collected from participants and raters at every visit. - **Readout:** Pre-registered primary symptom outcome, analysed unadjusted and adjusted for guessed allocation, with the guess-accuracy distribution reported. - **Critical control:** The allocation-guess data themselves; a trial that does not report them reproduces the gap this review identifies rather than closing it. - **Result that would falsify the nominal model:** No drug-comparator difference among participants whose guessing was at chance, which would place SC2 below its current ceiling and trigger the rubric's first falsification pattern. - **Result that would support the nominal model:** A drug effect surviving adjustment for guessed allocation, which would cross Δ for SC2 and remove the largest single objection to the efficacy clause. ### Experiment 3: Stratification of SSRI response by an independently measured baseline serotonergic index - **Experiment:** A prospective stratum within a randomised SSRI trial in which baseline serotonergic function is measured independently — for example by a challenge paradigm of the kind used in [6] — before randomisation, with response analysed by baseline stratum. - **Readout:** Pre-registered interaction between the baseline serotonergic index and the drug-placebo difference. - **Critical control:** Blinding of the baseline index to allocation and of outcome raters to the index, with the stratum boundary pre-registered so it cannot be fitted after outcomes are known (pitfall 22). - **Result that would falsify the nominal model:** Equal benefit regardless of baseline serotonergic function, which would refute the deficit-correction clause directly and move SC3 from "not demonstrated" to "shown not to be". - **Result that would support the nominal model:** Benefit concentrated in a pre-defined low-function stratum, which would be the first evidence that SSRIs correct a deficit rather than displacing a system that was not deficient, and would cross Δ for SC3. --- ## Knowledge gaps - **No validated in vivo measure of synaptic serotonin in untreated depression** — there is no specific, validated, scalable measure of synaptic serotonin availability in living untreated patients; the nearest index is release capacity inferred from radioligand displacement in samples of a few tens of people. Without it SC1's premise cannot be tested directly and SC3's stratification experiment cannot be run. - **Specificity of tryptophan depletion** — it is not established that the procedure lowers brain serotonin specifically rather than altering amino-acid balance, neurotrophic signalling and cerebrovascular parameters at the same time. This is the single constraint holding SC4 at Δ, and it is resolvable with a matched non-depleting control arm. - **Blinding integrity has never been measured in the SSRI trial base** — no trial in the syntheses examined here reports allocation-guess data, so the magnitude of functional unblinding is unknown rather than known to be small. It is the reason SC2 cannot cross Δ despite resting on the largest randomised evidence base in psychiatry, and it is cheap to close. - **The deficit-correction question has never been posed as an experiment** — no study has stratified antidepressant response by an independently measured baseline serotonergic index. SC3 is untested rather than refuted. --- ## Reviewer notes **Red-team record.** This edition is the red-team synthesis of the single input grading archived unaltered at `sources/A_v5.4_grading/`, performed on 27 September 2026 by Claude (Anthropic), an independent model from a different family than the grader. What changed: the overall level moved Β → Γ on a cap-precedence correction at SC1; a fourth subclaim was added at Δ for the proposition the input omitted; SC2 and SC3 held with their governing rules replaced; confidence moved to Moderate, role to reinforcing / stabilizing factor, Stability to Stable; the source base grew from 8 to 17 records, four of the original eight requiring identifier correction and one being a composite of three separate records; and four of the original eight claims were found materially inaccurate, three of them load-bearing. Why: the input's identifiers were assembled rather than retrieved, and two of its caps were fired against rubric text that does not exist in the rubric. After an independent consistency check the same day, SC1's Γ was re-rested on sourced human evidence only, after a clause citing rodent work that no ledger row carried was removed; the letter is unchanged. The full trail — the identifier table, the claim-by-claim audit, the numbered findings with their dispositions, and the change log, including its post-verification section — is in `PROCESS.md` and in `redteam/`. **Conjunction adjudication.** The aggregation rule is unchanged from the input grading and remains weakest-link, pre-registered for the conjunctive form. What changed is the decomposition: four subclaims rather than three, with the specificity requirement moved out of the efficacy clause. All three aggregation rules now return the same letter, which they did not before. **Cap precedence.** Two collisions are recorded openly in the cap-rules section: universal cap 2 against rubric 07's Γ cap on SC1, and the Δ sentence's pre-registration clause against the mediator-manipulation limb of its own field-valid list on SC4. Both are resolved with the framework's rule and both alternative readings are stated. The second is flagged for the rubric's maintainer: if "with a pre-registered primary outcome" is intended to govern the depletion and challenge paradigms as well as trial designs, no study in that literature can ever cross Δ, and the rubric should say so explicitly. **A rubric question raised by this review.** The functional-unblinding rule enumerates psychedelics, ketamine, entactogens, suprathreshold stimulation and ECT. Whether that list is exhaustive or exemplary decides whether the rule fires on antidepressant trials, which is not a small question for this rubric's main subject matter. This review read it as written and reached the same ceiling by the Δ threshold instead; a maintainer should settle it rather than leaving each reviewer to extend it privately. **Divergence reference class — for the owner to settle corpus-wide.** The headline Divergence (+3, severe) is computed against public and patient-facing communication; against the specialist literature the same arithmetic gives +1, mild, and both are published. The framework asks for the level implied by how "the field itself presents this hypothesis — its reviews, abstracts, grant framing, and press". Read strictly, "the field itself" favours the specialist class and would give Mild; read to include press and clinical communication, it supports Severe. This review keeps Severe because the hypothesis's influence runs through the second channel, but the choice decides the headline number a reader takes off the page, and it should be made once for the corpus rather than per review. **Source tiering.** Tier A is assigned here by the provenance of the number cited rather than by publication genre: a meta-analysis that computed the pooled estimate being cited is the record that produced that estimate, and is tiered A for it, while narrative reviews, commentaries and the umbrella review — which explicitly did not synthesise its included meta-analyses — remain Tier B and carry no experimental numbers. This reconciles authoring standard §5.11, which permits a named meta-analysis to carry its own pooled estimate with its N, with validator 2.4's requirement that a numeric claim rest on a Tier A row. It is recorded here because it is an adjudication, not a convention the corpus has previously stated. **Competing models.** The two lowest-ranked models are scored as preliminary expectations from the framing literature rather than from dedicated ledgers, and their supporting evidence is characterised generically for that reason. The input grading named specific studies for every model with no ledger row behind any of them; that is pitfall 19 and it has been removed rather than reproduced. A future update should give the tournament its own source set. **Recommendation.** Add a stub entry for this hypothesis to `hypothesis-index.md`: primary rubric 07; scope type initiation, maintenance and response; bundled, conjunctive; the four subclaims used here with their types and evidence anchors; aggregation weakest-link; the five competing models; and two adjudication notes — that the depletion literature is a human mediator manipulation and must be scored as one, and that universal cap 2 binds the biomarker and genetic leg of the serotonin claim rather than the claim itself. --- ## ADVERSARIAL REVIEW AND APPROVAL **Authoring model:** Sol 6.1. **Adversarial reviewer:** Claude Opus 5.5 (Anthropic). **Date:** 2026-10-14. **Scope of checks.** All 17 reference PMIDs verified against PubMed (17/17 resolve to the stated title, first author, journal and year; no retractions or expressions of concern; no empty DOI fields; no dropped collective authors). The four historical identifiers named in the Consensus provenance block (PMID 7944875, 12869766, 15289275, 19755572) were additionally checked and all resolve to on-topic serotonin/depression records, and the three linked Consensus artifacts exist at the stated paths. Because this review bears on mental-health treatment decisions, every quantitative claim about antidepressant efficacy, the serotonin-deficiency account and the depletion literature was checked against its source, including which outcome was primary and which analysis set produced each pooled number: PMID 29477251 (Cipriani: 522 trials, 116,477 participants, primary outcomes response rate and all-cause discontinuation, ORs 2.13 [1.89-2.41] to 1.37 [1.16-1.63]), 28178949 (Jakobsen), 14974002 (Moncrieff Cochrane: 9 trials, 751 participants, 0.39 SD, 0.17 [0.00-0.34] on sensitivity analysis), 18303940 (Kirsch), 31303567 (Hieronymus: 8,262 patients, 28 trials), 2184795 (Delgado: 21 patients, tryptophan down 87% and 91%, 14 of 21 relapsed, no effect of control testing), 17389902 (Ruhe strata), 28373689 (31 datasets, 38,802 subjects), 30845820 (18 candidate genes, Ns to 443,264), 36635177 (17 patients, 20 controls) and 22967776 (26 prospective studies, cumulative incidence 0.25 at 24 weeks). Scoring logic re-derived against `hsef-hypothesis-review-v5/SKILL.md` Step 3 and `references/hsef-framework.md`. No level, confidence or role change. | ID | Severity | Location | Original to corrected | Evidence | |---|---|---|---|---| | S-1 | Major | Consensus provenance reconciliation (9 October 2026) | "The earlier adversarial approval refers to its original edition" becomes "No earlier adversarial approval exists for this case" | No adversarial approval section, banner or approval line exists for this case anywhere in the deliverables tree. The sentence asserted an approval that was never issued, and was introduced by the 9 October revision -- a regression relative to the earlier edition. | | S-2 | Minor | Evidence inventory, SC2 paragraph; claim register C11; score card evidence entries (2) | "a separate synthesis of 131 trials in 27,422 participants found -1.94 HDRS points" becomes an attribution of the pooled estimate to the 49 of those trials that reported the outcome | PMID 28178949: 131 trials enrolling 27,422 participants were *included*, but the -1.94 HDRS mean difference (95% CI -2.50 to -1.37) is pooled over 49 trials, as the abstract states. The 131/27,422 figures correctly describe the included set and are retained; only the analysis set behind the number is corrected. This matters because the -1.94 estimate is load-bearing for SC2's clinical-significance finding. No direction, magnitude or confidence interval changed. | | S-3 | Editorial | Evidence inventory, SC2 paragraph | "1.80 HRSD points in the complete FDA registration dataset" becomes the same phrase qualified "for the four new-generation antidepressants with full data (35 trials, 5,133 patients)" | PMID 18303940 analysed "all clinical trials submitted to the FDA for the licensing of the four new-generation antidepressants for which full datasets were available". The unqualified phrase could be read as covering every antidepressant registered with the FDA. The review's own claim register (C10) and score card already carry the precise scope; the prose is brought into line with them. | **Rejected findings (checked, no change).** 1. **Antidepressant-efficacy statements.** Every efficacy claim matches its source exactly, including the direction and the hedging: Cipriani's finding that all antidepressants were more effective than placebo, with the stated OR range and its primary outcomes; Jakobsen's effect being below its own pre-specified 3-point threshold with all trials at high risk of bias and none using an active placebo; the Cochrane active-placebo pooled 0.39 SD falling to 0.17 (0.00-0.34) on removal of one strongly positive trial, with the authors' own unblinding conclusion. None of these is reported as ITT- or per-protocol-dependent in its source, and the review makes no ITT or per-protocol claim. The lay summary states plainly that the grade "is not a claim that antidepressants do not work", which is the correct reading of this evidence base. No change. 2. **Serotonin-deficiency statements.** The umbrella review's conclusion, the 35-author critique and the reply are all reported as positions rather than as settled fact, and the depletion strata are reproduced exactly as PMID 17389902 reports them, including its own conclusion that depletion studies fail to demonstrate a causal relation and instead mark a vulnerability trait. The never-depressed null is used to block the Delta threshold for SC1 rather than being read as support, which is the correct handling. No change. 3. **Confidence band.** Checked against the lead policy of 2026-10-14. This review's `confidence_rationale` already rests entirely on evidence density and reproducibility -- the collapse of the candidate-gene programme in the largest samples, inconsistent and medication-confounded imaging, a heterogeneous depletion literature partly concentrated in overlapping groups, and an unreplicated release-capacity measure in 17 patients -- and makes no appeal to access depth. Moderate stands and no rewrite was needed. 4. **Weakest-link aggregation and the headline.** The rule is pre-registered for the conjunctive form, the derivation over SC1, SC2, SC3 and SC4 is arithmetically and verbally correct, and all three rules are reported as returning the same letter. SC4's higher level is reported separately rather than raising the headline, which is what a scope-restricted subclaim requires. No change. 5. **"Best-supported rival" renderer artifact.** Checked the tournament ordering: rank 1 is neuroinflammation, which is not this review's nominal model (the nominal ranks 2 of 5). The prose is correct as written. 6. **Momentum "Associational only" and Stability "Stable".** Both carry the required artifact -- a dated registry search for the decisive designs, and the framework's explicit default where no trigger can be named. No "Decisive work underway" claim is made. 7. **REVIEWER NOTES versus the spliced Consensus section.** The notes describe a red-team synthesis and make no claim that Consensus work was or was not performed, so there is no contradiction with the spliced provenance block. The three artifacts the block cites exist at the stated paths and the four historical identifiers it lists all resolve to on-topic records. No change. 8. **Reference [16] year.** Printed as 2025 against a PubMed year of 2024 (Front Psychol volume 15); within the plus-or-minus-one print versus e-pub convention. Rejected. 9. **Rendering defects.** Scanned for glued numerals, empty `DOI: .` fields and "PMIDnot supplied" strings; none present. **Residual limitations.** Eleven of the 17 sources were examined at abstract and metadata level only, so body-level figures taken from full text (the Kirsch 1.80-point difference, the depletion subgroup effect size reported in the critique, the challenge-PET percentage change) rest on the review's own full-text reads of sources 1, 2, 10 and 14 rather than on an independent re-read here. The review's disclosure that the anchoring safeguard could not be met, because the prior letter was supplied in its brief, is retained and is the right call. This pass is a source, claim and scoring-logic audit, not a blinded recalibration. **Approval:** Following adversarial review and the corrections listed above, the revised document is approved. -- Claude Opus 5.5 (Anthropic), 2026-10-14. --- ## References 1. Moncrieff J, Cooper RE, Stockmann T, Amendola S, Hengartner MP, Horowitz MA. The serotonin theory of depression: a systematic umbrella review of the evidence. *Mol Psychiatry.* 2023;28(8):3243–3256. PMID 35854107. DOI 10.1038/s41380-022-01661-0. [Contradicts SC1: no convincing evidence of lowered serotonin in depression. Also carries the divergence framing and the specificity caveat on depletion.] 2. Jauhar S, Arnone D, Baldwin DS, Bloomfield M, Browning M, Cleare AJ, et al. A leaky umbrella has little value: evidence clearly indicates the serotonin system is implicated in depression. *Mol Psychiatry.* 2023;28(8):3149–3152. PMID 37322065. DOI 10.1038/s41380-023-02095-y. [Published critique of [1]; reports the depletion subgroup effect size the umbrella review did not present.] 3. Moncrieff J, Cooper RE, Stockmann T, Amendola S, Hengartner MP, Plöderl M, Horowitz MA. The serotonin hypothesis of depression: both long discarded and still supported? *Mol Psychiatry.* 2023;28(8):3160–3163. PMID 37322064. DOI 10.1038/s41380-023-02094-z. [Reply to [2]; recorded so the dissent is carried from both sides.] 4. Ruhé HG, Mason NS, Schene AH. Mood is indirectly related to serotonin, norepinephrine and dopamine levels in humans: a meta-analysis of monoamine depletion studies. *Mol Psychiatry.* 2007;12(4):331–359. PMID 17389902. DOI 10.1038/sj.mp.4001949. [Decisive perturbation, stratified: null in healthy volunteers, positive in family-history and remitted strata. The input grading cited PMID 17389903, a different paper in the same journal and volume.] 5. Delgado PL, Charney DS, Price LH, Aghajanian GK, Landis H, Heninger GR. Serotonin function and the mechanism of antidepressant action. Reversal of antidepressant-induced remission by rapid depletion of plasma tryptophan. *Arch Gen Psychiatry.* 1990;47(5):411–418. PMID 2184795. DOI 10.1001/archpsyc.1990.01810170011002. [Decisive perturbation: 14 of 21 remitted patients relapsed under blinded depletion. The input grading cited PMID 2268388, an unrelated cytometry paper.] 6. Erritzøe D, Godlewska BR, Rizzo G, Searle GE, Agnorelli C, Lewis Y, et al. Brain serotonin release is reduced in patients with depression: a [11C]Cimbi-36 positron emission tomography study with a d-amphetamine challenge. *Biol Psychiatry.* 2023;93(12):1089–1098. PMID 36635177. DOI 10.1016/j.biopsych.2022.10.012. [Supports SC1 in direction; biomarker endpoint, so capped at Γ by cap 4 and the acute-pharmacodynamic rule.] 7. Culverhouse RC, Saccone NL, Horton AC, Ma Y, Anstey KJ, Banaschewski T, et al. Collaborative meta-analysis finds no evidence of a strong interaction between stress and 5-HTTLPR genotype contributing to the development of depression. *Mol Psychiatry.* 2018;23(1):133–142. PMID 28373689. DOI 10.1038/mp.2017.44. [Contradicts SC1's genetic leg; 31 datasets, 38,802 subjects.] 8. Border R, Johnson EC, Evans LM, Smolen A, Berley N, Sullivan PF, Keller MC. No support for historical candidate gene or candidate gene-by-interaction hypotheses for major depression across multiple large samples. *Am J Psychiatry.* 2019;176(5):376–387. PMID 30845820. DOI 10.1176/appi.ajp.2018.18070881. [Contradicts SC1's genetic leg; 18 candidate genes, samples to 443,264.] 9. Cipriani A, Furukawa TA, Salanti G, Chaimani A, Atkinson LZ, Ogawa Y, et al. Comparative efficacy and acceptability of 21 antidepressant drugs for the acute treatment of adults with major depressive disorder: a systematic review and network meta-analysis. *Lancet.* 2018;391(10128):1357–1366. PMID 29477251. DOI 10.1016/S0140-6736(17)32802-7. [Supports SC2; establishes no mechanism and contains no active-placebo comparison.] 10. Kirsch I, Deacon BJ, Huedo-Medina TB, Scoboria A, Moore TJ, Johnson BT. Initial severity and antidepressant benefits: a meta-analysis of data submitted to the Food and Drug Administration. *PLoS Med.* 2008;5(2):e45. PMID 18303940. DOI 10.1371/journal.pmed.0050045. [Attenuates SC2: 1.80 HRSD points, the NICE criterion met only above about 28 at baseline, driven by declining placebo response.] 11. Jakobsen JC, Katakam KK, Schou A, Hellmuth SG, Stallknecht SE, Leth-Møller K, et al. Selective serotonin reuptake inhibitors versus placebo in patients with major depressive disorder. A systematic review with meta-analysis and Trial Sequential Analysis. *BMC Psychiatry.* 2017;17(1):58. PMID 28178949. DOI 10.1186/s12888-016-1173-2. Erratum: *BMC Psychiatry.* 2017;17(1):162. PMID 28468655; DOI 10.1186/s12888-017-1311-5. [Establishes that none of 131 SSRI trials used an active placebo. The input grading cited this work under a different journal, volume, pages, DOI and PMID.] 12. Moncrieff J, Wessely S, Hardy R. Active placebos versus antidepressants for depression. *Cochrane Database Syst Rev.* 2004;2004(1):CD003012. PMID 14974002. DOI 10.1002/14651858.CD003012.pub2. [The nearest existing form of the SC2 threshold experiment; pooled 0.39 SD, 0.17 (0.00–0.34) on sensitivity analysis.] 13. Hieronymus F, Lisinski A, Nilsson S, Eriksson E. Influence of baseline severity on the effects of SSRIs in depression: an item-based, patient-level post-hoc analysis. *Lancet Psychiatry.* 2019;6(9):745–752. PMID 31303567. DOI 10.1016/S2215-0366(19)30216-0. [Contests the severity dependence in [10] at patient level; 8,262 participants, 28 trials.] 14. Harmer CJ, Duman RS, Cowen PJ. How do antidepressants work? New perspectives for refining future treatment approaches. *Lancet Psychiatry.* 2017;4(5):409–418. PMID 28153641. DOI 10.1016/S2215-0366(17)30015-9. [Rival mediators for SC3; also states that all licensed antidepressants are believed to act at least initially by increasing monoamine availability. The input grading cited PMID 28237360, an unrelated paper, and attributed to this review an assertion it does not make.] 15. Lacasse JR, Leo J. Serotonin and depression: a disconnect between the advertisements and the scientific literature. *PLoS Med.* 2005;2(12):e392. PMID 16268734. DOI 10.1371/journal.pmed.0020392. [Divergence artifact: advertising claims of serotonin correction unsupported by the literature.] 16. Schroder HS, Tovey J, Forer R, Schultz W, Kneeland ET, Moser JS. Where do "chemical imbalance" beliefs come from? Evaluating the impact of different sources. *Front Psychol.* 2025;15:1469913. PMID 39845538. DOI 10.3389/fpsyg.2024.1469913. [Divergence artifact: healthcare providers uniquely predict endorsement of the belief.] 17. Udina M, Castellví P, Moreno-España J, Navinés R, Valdés M, Forns X, et al. Interferon-induced depression in chronic hepatitis C: a systematic review and meta-analysis. *J Clin Psychiatry.* 2012;73(8):1128–1138. PMID 22967776. DOI 10.4088/JCP.12r07694. [Tournament only: the neuroinflammation model's human exposure-incidence line.] --- ## Scope qualifiers | Qualifier | Determinable before outcome? | Determined from | Action | |---|---|---|---| | major depressive disorder | yes | baseline diagnostic status by DSM criteria at enrollment | retained | | in people with a history of major depressive disorder (SC4 only) | yes | documented past episode and current remission status at enrollment | retained | | in responders | no | response to treatment, which is the assessed outcome | deleted | The third qualifier appeared in neither the input grading nor this edition's graded claim. It is recorded because it is the form this hypothesis most naturally attracts — "SSRIs correct a serotonin deficit *in responders*" defeats every counterexample and reduces to the outcome restated — and because SC4's scope, which is retained, looks superficially similar and is not: a history of depression and current remission are both determinable at enrollment, before any depletion is administered. ## Citation ledger | # | Citation | Tier | Kind | Identifiers | Status | Carries | Opened | |---|---|---|---|---|---|---|---| | 1 | Moncrieff et al. Mol Psychiatry. 2023;28(8):3243–3256. | B | review | PMID 35854107 · doi:10.1038/s41380-022-01661-0 | none found | C1, C2, C17 | 2026-09-27 (full text) | | 2 | Jauhar et al. Mol Psychiatry. 2023;28(8):3149–3152. | B | review | PMID 37322065 · doi:10.1038/s41380-023-02095-y | none found | C4, C20 | 2026-09-27 (full text) | | 3 | Moncrieff et al. Mol Psychiatry. 2023;28(8):3160–3163. | B | review | PMID 37322064 · doi:10.1038/s41380-023-02094-z | none found | C19 | 2026-09-27 (partial) | | 4 | Ruhé et al. Mol Psychiatry. 2007;12(4):331–359. | A | meta_analysis | PMID 17389902 · doi:10.1038/sj.mp.4001949 | none found | C3, C4 | 2026-09-27 (abstract) | | 5 | Delgado et al. Arch Gen Psychiatry. 1990;47(5):411–418. | A | primary | PMID 2184795 · doi:10.1001/archpsyc.1990.01810170011002 | none found | C5 | 2026-09-27 (abstract) | | 6 | Erritzøe et al. Biol Psychiatry. 2023;93(12):1089–1098. | A | primary | PMID 36635177 · doi:10.1016/j.biopsych.2022.10.012 | none found | C6 | 2026-09-27 (abstract) | | 7 | Culverhouse et al. Mol Psychiatry. 2018;23(1):133–142. | A | meta_analysis | PMID 28373689 · doi:10.1038/mp.2017.44 | none found | C7 | 2026-09-27 (abstract) | | 8 | Border et al. Am J Psychiatry. 2019;176(5):376–387. | A | primary | PMID 30845820 · doi:10.1176/appi.ajp.2018.18070881 | none found | C8 | 2026-09-27 (abstract) | | 9 | Cipriani et al. Lancet. 2018;391(10128):1357–1366. | A | meta_analysis | PMID 29477251 · doi:10.1016/S0140-6736(17)32802-7 | none found | C9 | 2026-09-27 (abstract) | | 10 | Kirsch et al. PLoS Med. 2008;5(2):e45. | A | meta_analysis | PMID 18303940 · doi:10.1371/journal.pmed.0050045 | none found | C10 | 2026-09-27 (full text) | | 11 | Jakobsen et al. BMC Psychiatry. 2017;17(1):58. | A | meta_analysis | PMID 28178949 · doi:10.1186/s12888-016-1173-2 | erratum: BMC Psychiatry 2017;17(1):162, PMID 28468655 | C11 | 2026-09-27 (abstract) | | 12 | Moncrieff, Wessely & Hardy. Cochrane Database Syst Rev. 2004;2004(1):CD003012. | A | meta_analysis | PMID 14974002 · doi:10.1002/14651858.CD003012.pub2 | none found | C12 | 2026-09-27 (abstract) | | 13 | Hieronymus et al. Lancet Psychiatry. 2019;6(9):745–752. | A | primary | PMID 31303567 · doi:10.1016/S2215-0366(19)30216-0 | none found | C13 | 2026-09-27 (abstract) | | 14 | Harmer, Duman & Cowen. Lancet Psychiatry. 2017;4(5):409–418. | B | review | PMID 28153641 · doi:10.1016/S2215-0366(17)30015-9 | none found | C14, C15 | 2026-09-27 (full text) | | 15 | Lacasse & Leo. PLoS Med. 2005;2(12):e392. | B | review | PMID 16268734 · doi:10.1371/journal.pmed.0020392 | none found | C16 | 2026-09-27 (abstract) | | 16 | Schroder et al. Front Psychol. 2025;15:1469913. | A | primary | PMID 39845538 · doi:10.3389/fpsyg.2024.1469913 | none found | C18 | 2026-09-27 (abstract) | | 17 | Udina et al. J Clin Psychiatry. 2012;73(8):1128–1138. | A | meta_analysis | PMID 22967776 · doi:10.4088/JCP.12r07694 | none found | C21 | 2026-09-27 (abstract) | ## Claim register | ID | Claim | Load-bearing for | Sources | Verification | |---|---|---|---|---| | C1 | No convincing evidence that depression is associated with, or caused by, lower serotonin concentrations or activity. | SC1 | 1 | verified — full text | | C2 | The same review: depletion showed no effect in most healthy volunteers, very-low-certainty weak evidence with a family history, and slightly greater mood reduction in diagnosed participants, most of whom were medicated. | SC1, SC4 | 1 | verified — full text | | C3 | Depletion meta-analysis, stratified: no decrease in healthy controls; slight decrease with family history; moderate decrease in drug-free remitted MDD; relapse in remitted patients on serotonergic antidepressants; authors conclude depletion fails to demonstrate a causal relation and marks a vulnerability trait. | SC1, SC4 | 4 | verified — abstract | | C4 | For the drug-free remitted stratum the critique reports Hedges' g = −1.9 (95% CI −3.02 to −0.78) across 8 samples, falling to −1.06 (−1.83 to −0.29) with one outlier removed. | SC4 | 4, 2 | verified — full text of [2] | | C5 | 21 patients in antidepressant-induced remission; total and free tryptophan fell 87% and 91%; 14 of 21 relapsed; recovery within 24–48 h; no effect of control testing. | SC3, SC4 | 5 | verified — abstract | | C6 | Challenge PET: frontal binding fell in 20 controls but not in 17 antidepressant-free patients (15% ± 14% vs 6.5% ± 20%, p = 0.041). | SC1 | 6 | verified — abstract | | C7 | 31 datasets, 38,802 subjects: no statistically significant stress-by-5-HTTLPR interaction under any definition. | SC1 | 7 | verified — abstract | | C8 | 18 candidate genes, samples 62,138–443,264: no main effects, no gene-by-environment effects, no excess association over non-candidate genes. | SC1 | 8 | verified — abstract | | C9 | 522 trials, 116,477 participants: all 21 antidepressants more efficacious than placebo, ORs 2.13 (1.89–2.41) to 1.37 (1.16–1.63); 46 trials at high risk of bias; certainty moderate to very low. | SC2 | 9 | verified — abstract | | C10 | 35 FDA trials, 5,133 patients: 9.60 vs 7.80 HRSD points, difference 1.80 (standardised 0.32); NICE criterion met only above about 28 baseline; driven by declining placebo response. | SC2 | 10 | verified — full text | | C11 | 131 trials, 27,422 participants included; −1.94 HDRS (95% CI −2.50 to −1.37) pooled over the 49 trials reporting that outcome, below the reviewers' 3-point threshold; no trial used an active placebo; all at high risk of bias. | SC2 | 11 | verified — abstract | | C12 | 9 active-placebo trials, 751 participants: 0.39 SD (0.24–0.54), falling to 0.17 (0.00–0.34) without one trial; authors conclude unblinding may inflate inert-placebo estimates. | SC2 | 12 | verified — abstract | | C13 | 8,262 participants, 28 trials: no difference between non-severe and severe depression in SSRI-induced reduction of depressed mood or other HDRS-6 items. | SC2 | 13 | verified — abstract | | C14 | Delayed onset not explained by acute monoamine potentiation; neuroplasticity and affective-bias accounts advanced; all licensed antidepressants believed to act at least initially by increasing serotonin or norepinephrine availability. | SC3 | 14 | verified — full text | | C15 | Early drug-induced change in emotional processing predicts later clinical response and behaves as a specific treatment effect rather than a general expectancy mediator. | SC2, SC3 | 14 | verified — full text | | C16 | Advertising claims that SSRIs correct a serotonin imbalance are not supported by the peer-reviewed literature. | SC3 | 15 | verified — abstract | | C17 | Textbooks give the theory qualified support; surveys indicate most of the public believes the chemical-imbalance account is established; many general practitioners share it. | divergence basis | 1 | verified — full text | | C18 | Survey of 1,219 students: only hearing the explanation from a healthcare provider uniquely predicted endorsing it. | divergence basis | 16 | verified — abstract | | C19 | In reply, the umbrella review's authors maintained their conclusion. | dissent record | 3 | verified — partial full text | | C20 | The critique argues the conclusion is overstated on selective reporting, omitted tryptophan meta-analyses and imaging misinterpretation, and concludes serotonin has a role in vulnerable and depressed people. | SC1 | 2 | verified — full text | | C21 | 26 prospective interferon-alpha studies: cumulative incidence of induced major depressive episode 0.25 (0.16–0.35) at 24 weeks and 0.28 (0.17–0.42) at 48 weeks. | tournament | 17 | verified — abstract | *Review generated using the HSEF Hypothesis Review skill v5.4, in revision mode, as the red-team synthesis of the input grading archived at `sources/A_v5.4_grading/`. Scoring logic is traceable via the companion score-card JSON (mdd-serotonin.scorecard.json). Validated with tools/validate_review.py v2.4.*