In plain terms

The claim being assessed: In humans, prespecified changes in gut microbial community composition are primary upstream causes of disease across distant-organ disease classes.

Where the evidence stands: Γ — Plausible mechanism. There is a believable biological story and partial support, but the causal step has not been demonstrated.

What this mechanism does: This makes the process worse or keeps it going, but something else starts it.

Strongest evidence for it: Gut communities can causally modify selected animal phenotypes and recurrent CDI, with mixed context-specific human metabolic, hepatic, and Parkinson signals. No evidence establishes community composition as a primary cause across human distant-organ disease classes.

What decisive evidence would look like

In plain terms, researchers would change the proposed cause, confirm it changed as intended, and measure a meaningful health outcome while checking other explanations. The result should be able to support or count against the claim.

Relative explanatory fit: Alternative ranked ahead

What this means for claims that the microbiome controls whole-body health

Gut microbes can make biologically active chemicals and can affect disease in tightly controlled animal experiments. In people, however, the same community can be shaped by diet, medicines, illness, and bowel transit. That makes a broad claim that microbes are the main upstream cause of disease throughout the body much harder to prove.

Implications for health and treatment: The review supports a narrower, more useful view: microbiome-based treatment is clearly established for selected recurrent gut infections, while most claims about treating distant-organ disease, metabolic disease, cancer, or brain conditions remain experimental or mixed. A promising biological signal is a reason for well-controlled trials, not a reason to replace evidence-based prevention or treatment with supplements, consumer tests, or unsupervised microbial interventions.

Limits for individual care: There is no validated general-purpose 'healthy microbiome' target, and a result from mice, a short-term symptom measure, or an association study cannot prescribe an intervention for an individual. The review cannot determine whether a particular food, probiotic, or transplant will help a specific person; safety and suitability depend on the condition and clinical context.

Medical information notice: This review provides general health information, not medical advice. It cannot diagnose a condition or replace an individual discussion with a qualified clinician.

This assessment rests on 39 sources and 39 individually checked claims. 38 of those claims were verified against the original paper.

A low grade is not an accusation. It means the causal work has not been done yet, which is the ordinary state of most actively studied mechanisms in biology.

At a glance

Sub-claims assessed separately

ID	Sub-claim	Type	Level	Pivotal card
SC1	Human systemic disease cohorts show reproducible disease-associated differences in gut community composition.	existence	Β	—
SC2	Defined microbial manipulations can causally move disease-relevant extraintestinal phenotypes in selected rodent models.	necessity / sufficiency	Ζ	Card ↗
SC3	Restoring gut microbial colonization resistance after antibiotics reduces recurrent C. difficile infection in humans.	sufficiency	Ε	—
SC4	Microbiota-directed interventions can alter some distant-organ disease-related endpoints in selected human treatment contexts.	sufficiency	Ε	—
SC5	In humans, gut microbial community composition is an autonomous primary upstream cause of disease across multiple distant-organ disease classes.	primacy	Γ	Card ↗

Visual Evidence Cards · Pivotal Subclaims

Pivotal subclaim illustrations

Dark-theme simulated-micrograph renders and concise causal evidence summaries for the pivotal subclaims that govern this review's grade. Click any card to inspect the master 2400×2400 render.

Master 2400×2400 render ↗

Master 2400×2400 render ↗

Figures

Explanatory figures summarizing this review. They carry no evidential weight of their own; where a figure and the review differ, the review governs.

Full assessment

From here the review is written for a technical reader: caps and thresholds, competing models, the independence basis, and the citation ledger.

Sources: 39 (37 Tier A) · Load-bearing claims: 39 · Verified: 38 · Unverified: 1 · Contradicted: 0 · Gate: not passed — 1 unresolved claim (C35, finding F1)

Adversarial review: This edition records a red-team synthesis by Claude (Anthropic) over two OpenAI Codex gradings; adversarially reviewed by Claude Opus 5.5 (Anthropic) on 2026-10-14. 3 corrections applied (see ADVERSARIAL REVIEW AND APPROVAL). Status: Approved as revised.

In plain terms

The gut is home to a vast community of bacteria and other microbes. The claim graded here is the strong one: that the particular mix of microbes a person carries is a main cause of disease in organs far from the gut — the heart, the liver, the brain, the body's metabolism — rather than something that shifts after a person is already ill, or that simply reflects what they eat, the medicines they take, and how fast food moves through them.

The evidence does not establish that. It sits on the third rung of a nine-rung scale: the mechanism is plausible and works in simplified systems, but the causal dependency has not been shown in people at the breadth claimed. What is solid is narrower. In mice, adding a single bacterial species, or blocking one bacterial chemical reaction, changes arthritis, liver tumours and movement problems — real causation, in animals. In people, replacing the gut community cures a stubborn gut infection that follows antibiotics, which is causation too, but it is an infection of the gut itself, not of a distant organ. Transplant trials aimed at distant organs — in Parkinson's disease, in liver disease, in metabolic disease — are mixed: some find a benefit, several find none, and where a benefit appears it is usually confined to one measure and fades within months. Every one of those trials treated people who were already ill, so none can show that a microbial change started the disease.

The missing experiment is specific, and nobody has done it: choose the microbial feature in advance, change it in people before they are ill, and see whether the disease that follows is changed, in more than one organ, while accounting for diet, medication and how fast the gut moves. Until that exists, the honest reading is that gut microbes can modify some illnesses in some settings, not that they govern health across the body. That is a statement about how much work has been done, not an accusation against the people doing it, and not a reason to discount the parts that are established. A formal survey of the field itself reports considerably more confidence than this review does, and that gap is the single most useful thing on this page.

AT A GLANCE

Gut communities can causally modify disease in specific settings, but human evidence does not establish composition as a primary, cross-system cause of distant-organ disease.

• The experiment that decides it: A replicated, pre-onset human perturbation of a prespecified strain-defined community with distant-organ disease endpoints and controls for host, diet, transit, and medication.
• What the field implies vs. what the evidence shows: No single implied level can be established on the checkable record, so Divergence is Contested rather than a named gap. The 2025 Delphi consensus statement that this edition read as implying Ζ is paywalled, and its accessible abstract does not support that reading (see finding F1 in ADVERSARIAL REVIEW AND APPROVAL); the one other field-facing source in the ledger frames the field below this review's own assessment. The dispersion is real, and Γ and Moderate do not depend on it.
• Best-supported rival model: Host-, disease-, and exposure-first reciprocal ecosystem (Ε).

Rubric: 04-microbiome-systemic.md · HSEF v5.4 · 2026-09-27 · Previous level: Γ (both input gradings) Entry type: conjunction · Calibration entry: no · Review mode: revision (red-team synthesis of two versions)

Scorecard detail

Subclaim score matrix

Aggregation rule (pre-registered before scoring). Scope-matched: the overall score follows the broad human multi-system primacy scope. Animal, recurrent-CDI and selected human treatment evidence remain separately scoped and are reported, not aggregated.

Derivation. SC5 is Γ and sets the overall letter. SC2 is Ζ only inside selected rodent models; SC3 is Ε only inside the antibiotic-depleted CDI niche, which is not a distant organ; SC4 is Ε for selected human treatment contexts. None of the three higher subclaims shares the graded scope, so none can raise the letter.

What the other rules would return. Weakest-link Β, primary-subclaim Γ, scope-matched Γ. The aggregation choice changes the result by one rung because the purely associational SC1 is weaker than broad primacy. The rule was pre-registered before scoring in both input gradings and is the rule both applicable index entries prescribe, so there is no sign of rule-shopping in either.

COMPETING MECHANISTIC MODELS

Model 1 — Host-, disease-, and exposure-first reciprocal ecosystem

Host physiology, diet, transit, medications, genotype, and established disease shape community composition, while microbes feed back on selected phenotypes. Human population studies show how stool consistency and medication influence measured community [25, 26]; drug treatment itself can change the microbiome [24, 27]; and the direction can be demonstrated rather than inferred — switching volunteers between entirely animal-based and entirely plant-based diets reconfigured the community and its gene expression within days [32], while colonic transit tracks composition and the fermentation-to-catabolism balance [33]. Animal transfer and human FMT trials show community interventions can alter outcomes in selected settings. This is a reciprocal, context-dependent rival, not a claim that microbes are inert.

Model 2 — Niche-specific microbial causality

Microbial functions can be causal when the host state makes them limiting, as in antibiotic-depleted recurrent CDI or selected susceptible rodent models. This explains strong perturbation results without claiming gut composition is primary across human disease classes. It is the model the human genetic instrument most naturally supports: a defined microbial function, not community composition, tracking a metabolic trait [36]. The same intervention logic has not generalized reliably to other systemic outcomes.

Model 3 — Gut microbiome as autonomous multi-system master regulator

This nominal model proposes that community composition is upstream of and broadly governs human metabolic, oncologic, psychiatric, and immune health. It has plausible gut-organ pathways, causal animal results, and narrow human intervention signals. Null and transient outcomes, host and exposure confounding, failed transfers in rodents [29, 30], and the absence of any defined pre-onset human perturbation contradict the autonomous primary framing. Γ reflects partial support, not broad human causation.

Model 4 — Diet-substrate control

Diet supplies substrates for microbial functions and alters composition and systemic endpoints. Diet-dependent transfer effects [2], a human dietary-switch experiment [32], and trials where dietary change dominated [10] support this model. Microbial interventions still move selected phenotypes, so diet is not a complete substitute for microbial causality.

Model 5 — Host-genotype and physiology gating

Host genotype, immune state, barrier function, and transit affect colonization and phenotype expression. Genotype-specific mouse outcomes [6], variable engraftment [7, 9], and a hepatic phenotype that reversed with housing sector [30] support gating. This leaves room for causal microbial effects but challenges a uniform autonomous determinant.

Tournament result

The shared evidence favors reciprocal host- and context-gated causation, with strong niche-specific effects. The nominal model ranks third and loses on its broad human primacy claim; that loss informs the mechanism's role but does not itself lower the HSEF level. No completed study decisively separates the broad model from the leading rival. A defined perturbation before disease progression, with hard organ outcomes and independent replication, is still needed.

HYPOTHESIS STATE EVALUATION FRAMEWORK (HSEF v5.4)

Consensus consistency validation — 9 October 2026

Current grade: Γ
Grade changed: no

Scope: Retrospective, transparently unblinded consistency check of load-bearing primary records already in the topic inventory. This is not a new peer review and does not establish historical Consensus use.

Consensus queries: - gut microbiota causal systemic disease human randomized fecal microbiota transplantation autoimmune arthritis type 2 diabetes cardiovascular disease mechanisms primary studies - primary human randomized fecal microbiota transplantation trial Kootte 2017 Vrieze 2012 Yu FMT-TRIM autoimmune arthritis segmented filamentous bacteria Wu 2010

Saved raw Consensus records: - D:\HypothesisCritic\deliverables\consensus_consistency_audit_2026-10-09\checks\gut-microbiome-systemic-causality\consensus-search-01.json - D:\HypothesisCritic\deliverables\consensus_consistency_audit_2026-10-09\checks\gut-microbiome-systemic-causality\consensus-search-02.json - D:\HypothesisCritic\deliverables\consensus_consistency_audit_2026-10-09\checks\gut-microbiome-systemic-causality\consensus-fetch-c46f8701dfcf5ac1911e15b130428f46.json - D:\HypothesisCritic\deliverables\consensus_consistency_audit_2026-10-09\checks\gut-microbiome-systemic-causality\consensus-fetch-9df9e85d89ea5addbfae4573b28db55c.json

Original-primary comparisons: - PMID 20620945 (claims C1) (PubMed abstract_metadata in baseline inventory): Segmented filamentous bacteria drive Th17-associated autoimmune arthritis in a mouse model; bounded animal causality, not broad human systemic primacy. - PMID 24009397 (claims C2) (PubMed abstract, reopened 2026-10-09): Fecal microbiota from obesity-discordant twins transferred body/fat-mass and metabolic phenotypes to germ-free mice; diet-dependent and animal transfer. - PMID 32150549 (claims C9) (PubMed abstract_metadata in baseline inventory): FMT-TRIM placebo-controlled pilot in obesity did not improve the primary metabolic endpoint; weighs against assuming human systemic metabolic effects from transfer. - PMID 28978426 (claims C8) (PubMed abstract_metadata in baseline inventory): Lean-donor FMT effect on insulin sensitivity depended on baseline microbiota and was short-term; not disease initiation or general systemic cause.

Discovery/reconciliation versus source incorporation: Initial Consensus query surfaced mostly reviews. Targeted query returned an atopic-dermatitis RCT (irrelevant), a FMT-TRIM meeting abstract, and reviews; fetch verifies FMT-TRIM concerns an obesity pilot, not proof of cross-system human causation. The field Delphi consensus mentioned by the page is not service use. Original primary lines support selected animal causal effects and mixed/limited human treatment evidence.

Conclusion: No grade change: Γ for microbial composition as a broad primary cause across distant human organ diseases. Historical mention is field-level Delphi consensus, not Consensus service provenance; fresh searches are retrospective only.

This dated supplement adds source reconciliation to the original review. Its original narrative, registered scope, source inventory and grades remain historical records. No earlier adversarial approval has been verified for this edition at the supplement date. This provenance supplement has no separate approval and is not an independent or blinded adjudication. New sources described here are supplemental and are not counted in the original inventory.

Restated claim and testable boundary

The original sentence combines a broad primacy claim, a causal claim, and an underspecified outcome class. I score the strongest interpretable version: defined shifts in gut community composition are a primary cause of incident or progressive disease across human distant-organ systems. Here, "primary" means upstream and materially causal, with effects not explained mainly by host state, diet, transit, medication, or disease itself. "Defined shifts" must be specified before outcome observation by taxa/strains or reproducible functional consortium; it cannot mean whichever community difference is later found in people who became ill. "Distant organs" means extraintestinal clinical disease outcomes, not gut symptoms or microbial biomarkers. This scope makes the claim falsifiable and prevents defining "causal communities" by the outcome they are meant to explain.

A falsifying result would be replicated engraftment of a prespecified community change without altering a prespecified hard organ outcome, or apparent effects disappearing after controlling for host state, transit, diet, medication, or disease stage. A result that changes symptoms after diagnosis can support a treatment-stage modifier claim; by itself it cannot establish that composition initiated disease or is the primary determinant of systemic health.

Which index entry is being graded. Two entries overlap — Gut microbiome as master regulator and Microbiome causality in human systemic disease — and the index says to establish which is claimed before scoring. Grading A parsed the first, revision B the second. The commissioned topic matches the second entry's central claim, but both prescribe scope-matched aggregation and near-identical subclaims, differing mainly over whether recurrent C. difficile is listed separately. The five subclaims below are a superset of both prescriptions: the four the master-regulator entry lists, plus a separated human-treatment subclaim, which Step 2 permits where the evidence base warrants and which the size of the human interventional literature warrants here.

Subclaim adjudication

SC1 — Human disease-associated signatures (Β, moderate). Many conditions have reported stool-composition differences, but this supports association only. Metformin can reshape the microbiome and confound T2D signatures [24, 27]; stool consistency/transit and medications are major sources of community variation [25, 26]; and the direction of the diet–community relationship can be shown experimentally, not merely inferred [32, 33]. Disease physiology and treatment can therefore generate or alter the signature. The claim does not specify diseases, taxa, or prospective temporal order. SC1 neither demonstrates causality nor earns credit for cross-system primacy. Universal cap 2 — "Pure association evidence — however large, however replicated — cannot exceed Β" — binds.

SC2 — Causal effects in selected animal systems (Ζ, moderate, tightly bounded). Rubric 04's Ζ threshold reads, verbatim: "Requires: Defined consortium or strain-level causal evidence · Rescue logic · Replication across at least two relevant models or cohorts." All three are met inside the rodent scope, by four lines that do not share a failure mode.

Introducing a single species — segmented filamentous bacteria — into germ-free K/BxN mice reinstated the Th17 compartment, autoantibodies and arthritis [1]: one defined organism, an otherwise sterile host, add-back logic. In conventionally raised obese mice exposed to a chemical carcinogen, blocking production of the microbial metabolite deoxycholic acid, or reducing the gut bacteria producing it, prevented hepatocellular carcinoma [5]: a resident microbial function, removed and restored, in animals that were never gnotobiotic and into which nothing was transferred. In α-synuclein-overexpressing mice, antibiotics ameliorated and recolonization or short-chain fatty acid administration promoted motor deficits and microglial activation [28]: a third mediator class, a third organ system. And human twin-derived communities transferred adiposity to germ-free recipients, with cohousing rescue that was diet-dependent [2].

Different organisms, different mediators, different organs, different modalities, and — decisively — different relationships to the germ-free state. The framework's test is whether "a single systematic error could not produce both"; no single error produces all four. Cap 5 does not bind here: the cap "governs the inference from that system to the hypothesis's claimed scope", and this subclaim's scope is the rodent system. It binds SC5 instead, which is where it is applied.

Η is refused on its own quoted requirement — "Defined consortium or factor rescue in multiple contexts, a narrowed mechanism, and strong exclusion of diet, housing, batch, and handling confounds." No narrowed mechanism is replicated across models, and the second clause is demonstrably false of this literature: transfer of obesity-donor communities into antibiotic-pretreated mice on a low-fat diet shifted taxa without the phenotype [29]; a hepatic phenotype following one gnotobiotic colonization reversed direction between the germ-free and SPF sectors of a single facility [30]; and human-to-mouse transfer efficiency is itself a documented source of experimental variability [7]. Bloom's internal negative control belongs here too: commensal Bacteroides colonized susceptible and non-susceptible mice equivalently and induced disease only in the susceptible, while a disease-enriched Enterobacteriaceae isolate failed despite robust colonization [6] — the rodent boundary is host genotype, not abundance.

Those failures bind the human-donor whole-community line specifically, which is why it is marked the weakest of the four; they do not touch the single-species add-back or the resident-function removal. One refinement on the reading offered in revision B: the failed obesity transfer [29] used a low-fat chow diet, and the twin-transfer study itself reported that its rescue "was diet-dependent" [2]. The later null therefore falls inside the boundary the earlier study mapped rather than contradicting it — which is a stronger claim about context-dependence, not a weaker one.

SC3 — Recurrent CDI after antibiotic depletion (Ε, high confidence within this niche; out of scope for the graded claim). Randomized microbiota restoration reduces recurrence in the depleted post-antibiotic gut niche, including a large phase III purified-spore trial (12% vs 40%; RR 0.32, 95% CI 0.18–0.58) [15], alongside earlier FMT trials [14]. This is compelling human evidence that microbial restoration can causally restore local colonization resistance, and it is the one human context in which microbial necessity is demonstrated.

It is also, for this review, out of scope. CDI is an intestinal infection in a deliberately depleted niche; the graded claim is about organs distant from the gut. Under scope-matched aggregation it cannot contribute to the letter, and the index says so in terms: recurrent C. difficile "is a depleted-niche special case and does not generalise to multi-system regulation — do not let SC3 carry SC4." It is reported here because deleting the field's strongest human causal result would understate the evidence as surely as letting it carry the claim would overstate it.

SC4 — Selected human treatment-stage outcomes (Ε, moderate). Human trials provide real but bounded signals. Lean-donor infusion improved insulin sensitivity at six weeks in men with metabolic syndrome [31], and a later trial by the same group found the same six-week improvement absent by eighteen weeks, with response predicted by baseline faecal microbiota composition [8]. Other randomized trials show no benefit despite measured microbiota changes and persistent engraftment [9, 13], no incremental benefit over a Mediterranean diet [10], and short-lived effects that fade by twelve weeks [11, 12]. In cirrhosis, an open-label randomized trial found fewer further hepatic-encephalopathy episodes after transplant from a rationally selected donor [38] and a phase II trial found an HE recurrence signal on a secondary endpoint [16], whereas a 2026 active-comparator trial found similar minimal-HE resolution with FMT and lactulose (68.3% vs 66.1%; p=.80) [17]. A single-arm melanoma trial reported clinical benefit in 6 of 15 patients whose disease had progressed on anti-PD-1 [39] — a distant-organ endpoint, but uncontrolled.

Parkinson disease illustrates why the positive human signal should be neither ignored nor generalized. GUT-PARFECT found improvement on its off-medication motor scale at 12 months, while most other scales did not differ [18]. A Finnish multicenter RCT found no primary motor benefit at six months [19]. A 2025 30-person trial reported short-term improvement [20]; a 2026 phase II drug-naïve trial reported a week-35 motor-scale effect [21]; another 2026 double-blind trial found no primary motor benefit at 12 months [22]. These are treatment trials after disease onset using whole-community transfer, with different products, endpoints, timing, and control procedures. They make a Parkinson-specific gut-modifier hypothesis plausible, but do not identify the responsible community component or establish disease initiation. A 67-person systemic-sclerosis trial of cultured microbiota infusion also missed its primary lower-GI symptom endpoint [23]; it did not directly test extraintestinal organ causation.

Why Ε and not Δ. Step 10 is explicit: "Failed generalization is Ε, not a reason to stay at Δ… Ask whether the context boundary has been mapped — ideally by internal negative controls showing where the effect stops — or whether generalization simply was never tested. Mapped boundary means Ε; untested means Δ." The limits here are mapped, and mapped from inside the positive trials. In GUT-PARFECT the primary motor score and objective colon transit separate while twelve prespecified secondary instruments do not, one scale moves against the intervention, and no separation is present before six months [18]. The metabolic benefit is present at six weeks and gone by eighteen, with responders predicted by baseline composition determined before the outcome [8], and a second trial reproduces the same fade independently [12]. Engraftment without effect shows the boundary is not engraftment [9]; no incremental benefit over diet shows part of it is diet [10]; non-superiority to lactulose shows part of it is existing standard of care [17]. Rubric 04's Ε reads "context-specific causal effect with defined limits. A transferable phenotype exists, but microbial identity or generalizability remains incomplete" — which is this evidence, incomplete identity included.

A second, genuinely independent human evidence class supports the same reading. Bidirectional Mendelian randomization in 952 normoglycaemic individuals found host-genetics-driven gut butyrate production associated with improved insulin response, and abnormalities in propionate production or absorption causally related to increased type 2 diabetes risk (P = 0.004) [36]. That is the framework's first-named field-valid equivalent, in humans, pointing at a systemic metabolic endpoint. It instruments a microbial function rather than community composition and is partly confounded with host absorption, so it strengthens SC4 and does not reach SC5. Positive therapeutic effects and a genetic instrument are evidence of modifiability and of contribution; they are not evidence of primacy.

SC5 — Human multi-system primacy (Γ, moderate). The claim's decisive scope is not "can microbes ever affect a host?" It is whether defined community shifts are a primary causal determinant of systemic health, causing disease in distant organs. No reviewed study prospectively defines a community shift, demonstrates it precedes disease, perturbs it in humans, and shows replicated incident or progressive disease effects across distant-organ systems while controlling plausible host and exposure causes.

The carnitine/TMAO line is the clearest worked example of why an apparently convergent case stops short, and it is the reason it was added to this edition. Suppressing the intestinal microbiota in human volunteers abolished production of the proatherogenic metabolite TMAO from dietary carnitine, and it returned as the microbiota recovered; in mice, suppressing the microbiota abolished the carnitine-driven acceleration of atherosclerosis; and in a cohort of 2,595 people undergoing cardiac evaluation, carnitine predicted cardiovascular events only among those with concurrently high TMAO [37]. Three components, and a different cap binds each. The human perturbation's endpoint is a circulating metabolite, so universal cap 4 binds — "Biomarker change without functional/phenotypic change caps at Γ. This applies even when the biomarker change is mechanistically informative." The human clinical endpoint is observational, so universal cap 2 holds it at Β. The disease endpoint is in mice. Nothing in the chain is a human perturbation with a human disease outcome.

Forced-transfer paradigms carry the same ceiling for the inference that matters here: "Seeding or forced-transfer paradigms cap at Γ for endogenous propagation claims. Forced spread is not observed spread." The animal and CDI scores therefore remain high within their scopes and cannot lift SC5 above Γ.

Aggregation and caps

I preregistered scope-matched aggregation: the overall score follows SC5, the broad human primacy scope stated by the hypothesis. Therefore overall Γ. Weakest-link aggregation would return Β because SC1 is association-only; primary-subclaim and scope-matched aggregation both return Γ. SC2 is Ζ in selected rodent models; SC3 and SC4 are Ε in their bounded settings. The aggregation rule does not average across incompatible systems.

The rubric's thresholds are applied within each tested scope. Rodent and recurrent-CDI evidence cross causal thresholds in their own systems; human treatment trials and a human genetic instrument provide direct, though bounded, evidence in theirs. Two cap-versus-threshold collisions are recorded rather than resolved silently. The first is the standing one: in-vivo transfer crosses the rubric's Δ while universal cap 5 restricts the inference from forced transfer to endogenous human biology — resolved by scope, with cap 5 binding SC5 and the threshold binding SC2, SC3 and SC4. The second is new to this edition: the framework names Mendelian randomization among the field-valid equivalents that make the upper ladder reachable, while rubric 04's own list of domain equivalents does not mention it. The framework's enumeration is universal and governs reachability; the rubric's list is illustrative and not exhaustive, so the instrument is admitted at the scope it tests. Rubric 04's list is flagged as needing revision.

The convergence rule is retained and then tested against pitfall 21. Animal, human-genetic, human-interventional and observational classes are assessed together and no cap is re-applied to the set. But they converge on different claims at different scopes — one microbial function and a metabolic trait; a circulating metabolite; treatment-stage symptom change; rodent phenotypes — not on the graded claim. The set is not re-capped; it simply does not cover the graded scope, and there is no defined, pre-onset human perturbation establishing primary causation across distant-organ systems. The broad claim remains Γ.

Causal evidence count and independence

Recounted against this ledger: 28 intervention or perturbation records, of which 18 report a favourable direction on at least one stated endpoint inside their own bounded system, 8 are null on a relevant primary endpoint, and 2 are context-dependent, with the direction of effect depending on housing or diet. That is not 18 independent lines for SC5. Null trials are part of the causal audit. Observational studies — including the prospective cardiovascular association in the TMAO report — are excluded from positive causal counts, and the single-arm melanoma trial is counted as a record but not as a controlled perturbation.

The four rodent lines share only the species. The two CDI lines share the post-antibiotic depleted niche, which is why SC3 is Ε and not Ζ. Neither group carries SC5, the graded claim, which has no independent causal line at all — and that, rather than any cap, is the reason the aggregate is Γ.

Competing explanations and exclusions

The host-first reciprocal model fits the evidence: host physiology, disease, medication, diet, transit, and genotype shape microbial composition; microbial functions can then feed back in selected contexts. Metformin is both treatment and a microbiome perturbation [27], while stool consistency and medication explain substantial microbiome variation [25, 26] and a controlled dietary switch reconfigures the community within days [32]. Inflammation-dependent transfer behavior and genotype-specific disease phenotypes show that recipient context matters [6, 7], and a housing sector can reverse the sign of a transfer phenotype [30]. This model predicts causal effects without treating the microbiome as the primary upstream determinant.

The most serious alternatives have not been excluded for the broad claim: (1) reverse causation from disease or its treatment; (2) diet and medication as common causes of composition and outcome; (3) transit or bowel preparation altering composition and symptoms; (4) host-genotype or immune-state gating; and (5) intervention effects from many organisms, metabolites, or procedures rather than a defined compositional shift. Whole-stool FMT cannot identify which component is necessary or sufficient. Conversely, these limitations do not erase the context-specific causal results.

Where the two gradings disagree, and what was adopted

Dissent is recorded, not discarded. Both input gradings returned Γ with Moderate confidence and this synthesis holds both. Four disagreements sit beneath the letter.

Three of revision B's contributions are adopted. Its lay summary, which grading A lacked and which was A's only validator failure. Its Divergence method. And three sources showing that rodent transferability is less robust than A's ledger implied [28–30] — the last of which is the named reason Η is refused for SC2.

Calibration, uncertainty, and evidence quality

This is a red-team synthesis of two gradings of the same claim, both produced by the same model family, assessed by an independent model. Primary studies were sampled through 2026-09-27; the ledger is the union of both inputs plus four omission-audit additions. Every ledger row was resolved against PubMed or the publisher and every registered claim checked against its source, with full text read where a figure or a direction was load-bearing; thirty-four of thirty-nine rows were read at abstract depth, and no supplement audit is claimed. It is not an exhaustive systematic review or a pooled effect estimate. No sealed expected-band target was supplied, so this is not a calibration entry and no blind calibration score is claimed.

Confidence is moderate because four independent rodent perturbations, randomized CDI interventions, a human genetic instrument and emerging human treatment trials support causal effects in bounded settings, while direct tests of human pre-onset multi-system primacy are missing, human trial results are mixed, and the field's reproducibility problems are demonstrated rather than hypothetical [7, 29, 30]. That combination is what the framework's Moderate band describes: multiple independent studies, some replication, assay-level concerns remaining. No specific pending result has been identified that is expected to directly resolve the broad claim, so stability is Stable; "stable" describes the absence of an identified decisive pending result, not immutability of the evidence base. Momentum is Mechanistic only because recent trials perturb microbiota after disease is established.

Divergence is Contested (revised 2026-10-14). This edition computed Severe overstatement, a gap of +3, from a single field-facing source: a 2025 Delphi consensus statement of the Human Microbiome Action Consortium [35], read as reporting that the evidence strongly supported a causal role for the gut microbiome in inflammatory bowel disease, obesity, type 2 diabetes and liver diseases. That paraphrase is not supported by the accessible record. The PubMed abstract describes a Delphi survey of the utility of preclinical models in elucidating a causal role, says disruption in gut microbiome composition “is implicated in” various conditions, and states that “determining the underlying mechanisms and establishing cause and effect is extremely difficult”. The full text is paywalled with no PMC copy, so this review cannot establish what the consensus statements themselves say about field belief, and does not assert that they say the opposite — only that the reading is not checkable and that the abstract's framing is in tension with it. Claim C35 is therefore marked unverified and the implied level Ζ is withdrawn. The only other field-facing source in the ledger is a 2024 narrative review that restricts established human causality to recurrent C. difficile infection, ulcerative colitis and immunotherapy response and reports the obesity trials as unsuccessful [34] — a framing at or below this review's own assessed Γ, which would make Divergence “Aligned” or “Understated” rather than an overstatement. One narrative review is not a field-wide instrument, and two sources pointing in opposite directions is precisely the case the framework assigns to Contested: the field is genuinely split and no single implied level applies, so no rung arithmetic is available. The dimension is recorded as unresolved pending full-text access to [35]. Divergence modifies neither the level nor the confidence band, so Γ and Moderate are unaffected.

Upper-ladder check

Two of the five field-valid equivalents that grading A recorded as unavailable are in fact available, and correcting that is the largest evidential change in this edition — in the hypothesis's favour.

Mendelian randomization: available. Bidirectional MR in 952 normoglycaemic individuals reports host-genetics-driven butyrate production associated with improved insulin response and abnormal propionate handling causally related to type 2 diabetes risk (P = 0.004) [36]. It does not carry the claim further because it instruments a microbial function rather than community composition, is partly confounded with host absorption, and covers one trait class in one cohort.

Withdrawal and rechallenge: available. Antibiotic suppression abolished human carnitine-derived TMAO production and it returned on recovery [37]. It does not carry the claim further because the human endpoint is a metabolite — cap 4 — and the human clinical endpoint in the same report is observational — cap 2.

Therapeutic perturbation against a hard endpoint: available. Recurrent CDI is a randomized success against a hard endpoint, but it is a local intestinal infection in a deliberately depleted niche. Parkinson, hepatic and metabolic interventions are context-specific, mixed, use undefined products and begin after diagnosis; the melanoma result is single-arm.

Dose-response across populations: available but observational, and capped at Β. Natural experiment: not available — antibiotic exposure is the obvious candidate and is not clean, because it perturbs the host directly and is prescribed for conditions that independently predict outcomes.

Evidence converges on context-specific microbial effects, not on broad human primacy. The cap is not reapplied to the convergent set; the set does not cover the graded scope.

The steelman

The case for Δ, put at full strength. Two field-valid equivalents now reach human biology directly: a human genetic instrument pointing at a systemic metabolic endpoint [36], and a withdrawal-and-rechallenge experiment in people [37], whose mechanism accelerates atherosclerosis in mice and whose metabolite predicts cardiac events in thousands. Add randomized human transplant trials that move motor scores in Parkinson disease [18, 20, 21] and insulin sensitivity in metabolic syndrome [8, 12, 31], and four rodent lines that move arthritis, hepatic tumours and motor phenotypes by defined strains and defined microbial functions [1, 2, 5, 28]. That is human genetics, human pharmacological withdrawal, human randomized intervention and animal perturbation — four classes, each excluding a confounder the others cannot, which is the convergence pattern the framework says Ζ exists for and which certainly clears Δ.

The case for Β. The broad human evidence is overwhelmingly observational; diet, transit and medication are demonstrated determinants of the composition offered as the cause, with stool consistency the largest single source of variation in population cohorts and metformin sufficient to manufacture a type 2 diabetes signature; randomized transplant results are small, mixed, mostly transient and never use a defined product; and the one unambiguous human success is an intestinal infection in a niche emptied by antibiotics.

Rejected, and why. Δ is rejected for the graded primacy scope, on three named blockers. Cap 4 — the human withdrawal experiment ends at a circulating metabolite, and its clinical endpoint is observational, which cap 2 holds at Β. Scope — the genetic instrument tests one microbial function against one trait class in one cohort, so it cannot speak to primacy across disease classes, and a set of results pointing the same way at different scopes is not the convergent set the framework's non-recapping rule protects. Boundary-check clause 5 — no field-valid perturbation of a prespecified human community feature has returned a positive incident-disease result at all, because no such experiment has been run before onset; cap 5 then blocks forced-transfer evidence from filling that gap by inference. Β is rejected because it cannot accommodate four converging rodent perturbation lines, two randomized CDI trials, several positive randomized distant-organ results and a human genetic instrument. Δ or higher is retained in the narrower scopes that earned it.

What would change the assessment

A material upward update would require a preregistered, replicated intervention that (a) defines the strain or consortium before outcomes, (b) verifies engraftment/function, (c) precedes incident or progressive extraintestinal disease, (d) includes add-back/removal or rescue logic, (e) controls diet, transit, medications, genotype, procedure, and baseline disease, and (f) reproduces hard organ outcomes in an independent cohort or second organ system. A successful after-onset symptom intervention alone would raise a disease-specific modifier subclaim, not broad primacy. A well-powered null trial despite target engagement would count against causality in its prespecified scope. A Mendelian-randomization study instrumenting community composition rather than a single microbial function, replicated across cohorts against more than one distant-organ endpoint, would be the cheapest available route to Δ at the graded scope.

Proposed decisive experiments

1. Pre-onset test of defined-community primacy. In a prespecified high-risk population, perturb a defined microbial feature — a strain-resolved consortium or a named microbial function — before disease onset, verify target engagement longitudinally, and follow incident distant-organ disease in at least two organ systems. Controls: pre-outcome eligibility, blinded endpoint adjudication, diet, transit, medication and genotype measured throughout, and independent replication. Falsifies the nominal model if engagement is verified and incident outcomes do not move, or if host and disease changes precede the microbial shift. Supports it if a defined shift precedes and reproducibly changes incident disease across organ classes.

2. Independent replication of a human treatment-stage transfer effect, with add-back logic. Take the strongest existing signal — the Parkinson motor result or the six-week metabolic result — and replicate it multicentre and blinded, with a defined product rather than whole stool, prespecified durability endpoints beyond the point at which current effects fade, and a removal or add-back arm. Controls: co-transferred metabolites, bowel preparation, concomitant treatment, donor variability. Falsifies the narrow model if target engagement occurs without the predicted clinical effect across sites. Supports it if phenotype movement tracks the prespecified feature with rescue evidence.

3. Mendelian randomization on composition rather than function. Instrument community-composition features directly, with sensitivity analyses against host pleiotropy at the known loci, across independent cohorts and against more than one distant-organ endpoint. This is the one route to the graded scope that does not require an intervention no ethics committee would approve.

Knowledge gaps

• Human temporal direction. Every human interventional record enrolled people already diagnosed or already in a defined risk state. Nothing tests whether a community shift precedes onset.
• Defined causal identity. Most clinical trials transfer whole communities; the responsible organisms or functions remain unresolved, and whole-stool transfer cannot identify them in principle.
• Competing upstream factors. Diet, transit, medication, host physiology, genotype and established disease all reshape composition, and the first two have been shown to do so experimentally.
• Generalization boundaries in humans. The limits are mapped on the time and engraftment axes and partly on the diet axis; why the Parkinson trials disagree with each other is not explained by any mapped variable.
• Rodent reproducibility. Housing sector can reverse the sign of a transfer phenotype and transfer efficiency varies between recipients; this is why the rodent subclaim stops below Η.

Reviewer notes

Two structural notes on this edition. The plain-language summary leads the page, ahead of AT A GLANCE: the framework requires it "before any technical content", and the compact scorecard is technical content, so on this site the lay text comes first and AT A GLANCE follows immediately. And the subclaim-adjudication table in the input carried seven columns of data under a four-column header and rendered as garbage; it has been rebuilt as prose sections, which is where the argument was already being made.

Red-team record. The final stage of this review was an independent adversarial pass by Claude (Anthropic), a different model family from the two OpenAI Codex gradings it synthesises. All 42 ledger rows across both inputs were resolved in PubMed or at the publisher and all 36 registered claims were checked against their sources; no confabulated identifier and no citation inversion was found in either input, which in a corpus with this history is worth stating plainly. The level is held at Γ / Moderate, which both inputs also returned. Beneath the letter: the rodent subclaim is held at Ζ against revision B's Ε, because B applied Η's "narrowed mechanism" requirement at Ζ and because differing mechanisms establish independence rather than defeating it; the human distant-organ treatment subclaim is held at Ε against B's Δ, because the context boundary is mapped by internal negative controls inside the positive trials and B had removed five of the six trials that map it; the recurrent-CDI subclaim is restored and explicitly labelled out of scope, per the index's own adjudication note; the ledger is reconciled to the union of both inputs plus four omission-audit additions; and Divergence is recomputed from field framing, with an implied level of Ζ and a gap of +3 — rejecting A's method and B's value. Two claims in grading A's upper-ladder check that no Mendelian-randomization instrument and no human withdrawal experiment existed were overturned by evidence that runs for the hypothesis, and both are now assessed on their merits rather than denied. Revision B's lay summary, its Divergence method, and its three records showing that rodent transfer is less robust than A's ledger implied were adopted. One WARN-class judgement is disclosed here rather than fixed: thirty-four of thirty-nine ledger rows were read at abstract depth, which is recorded honestly on each row and is sufficient for every figure carried, but is not a full-text audit of the corpus. The complete trail — identifier audit, claim audit, numbered findings with dispositions, and the change log — is in PROCESS.md and redteam/.

ADVERSARIAL REVIEW AND APPROVAL

Authorship of record: This edition records a red-team synthesis by Claude (Anthropic) over two OpenAI Codex gradings. Specific model versions are not inferred from a house label. Adversarial reviewer for this round: Claude Opus 5.5 (Anthropic). Date: 2026-10-14.

Scope of checks. PMIDs verified 38/38 against PubMed, plus one deliberately DOI-only row (reference 17, not indexed in PubMed as of this date), checking title, first author, journal, year, DOI and publication-type flags; no retraction, expression of concern or erratum-of-record flag on any row, and every ledger row carries an explicit none_found status flag with a check date. Load-bearing claims spot-checked against the retrieved abstracts: 14 (PMID 35045228, 33542131, 38686220, 39900089, 30778224, 40033063, 39944364, 20620945, 24009397, 32150549, 28978426, 28586116, 39800192, 27912057, 26069274). Internal counts recomputed from the ledger itself: 39 rows, 37 Tier A, 39 registered claims, 39 crosswalk rows, access depths 34 abstract / 4 partial full text / 1 complete full text. Scoring logic checked against the HSEF framework and SKILL.md: the six universal caps and the three cap-versus-threshold collisions this edition records, the pre-registered scope-matched rule and its two declared alternatives, the independence table required for the Ζ subclaim, the upper-ladder field-valid equivalents, and the M/S/D artifact requirements. Validator re-run: PASS, 0 fail, 0 warn, 34 pass under the v5.4 validator this edition declares (31 pass under the archived v5.3 validator).

Findings

Rejected findings (checked, no change)

• Headline Γ. This is the strongest-argued letter in the group and I could not move it. The steelman for Δ is put at full strength and rejected on three named blockers — cap 4 on the human carnitine withdrawal experiment, whose endpoint is a circulating metabolite; cap 2 on the same report's observational clinical endpoint; and the absence of any field-valid perturbation of a prespecified human community feature before disease onset, which cap 5 then blocks forced-transfer evidence from filling by inference. The pre-registered scope-matched rule follows SC5, and the three higher subclaims (Ζ rodent, Ε recurrent C. difficile, Ε human treatment-stage) are each explicitly scope-restricted relative to multi-system human primacy.
• Scope-matched aggregation. Checked against the known misreading. Both Ε subclaims and the Ζ subclaim are scope-restricted; the derivation states this and the two alternatives (weakest-link Β, primary-subclaim Γ) are reported. No rule-shopping: both input gradings pre-registered the same rule, and both applicable index entries prescribe it.
• The Ζ on SC2. The independence table names four rodent lines and the distinct systematic error each excludes, and Η is refused on its own quoted requirement using three named reproducibility failures. This satisfies the framework's shared-failure-mode test; "the four rodent lines share only the species" is stated, not assumed.
• Recurrent C. difficile scored at Ε but excluded from aggregation. Checked and upheld: the index instructs that it must not carry the primacy claim, and reporting it while refusing to aggregate it is that instruction followed rather than evaded.
• Momentum "Mechanistic only". The artifact names the work — five randomized Parkinson transfer trials 2024-2026, the hepatic-encephalopathy trials, and the metabolic transplant series — and states why it is not "Decisive work underway": the trials perturb microbiota after disease is established. Checked, no change.
• Divergence. Originally checked and accepted on the ground that the arithmetic and the implied level were both named. That acceptance is withdrawn: the lead retrieved the underlying record and the implied level turned out to rest on an unverifiable paraphrase, so this is now finding F1 above rather than a rejected finding.
• "Decisive-perturbation studies: 32" against "28 intervention or perturbation records" in the text. Rejected as a contradiction: these are two different quantities, both defined. 32 is the count of ledger rows flagged decisive_perturbation in the score card; 28 is a narrower recount that the text defines on the spot and whose components (18 favourable, 8 null, 2 context-dependent) sum correctly. The labels sit four apart and the lead may wish to harmonise one of them.
• "All 42 ledger rows across both inputs" and "all 36 registered claims". Checked against the external process log, which records the same input-level totals before the union was reconciled to this edition's 39 rows and 39 claims. Not an error.
• Reference 17 carries a DOI and no PMID. Rejected: the reference preamble discloses the convention, the record is not indexed in PubMed (two targeted searches returned nothing), and the ledger row records full_text_partial access against the publisher. See the residual limitation below.
• Collective-author reference defect. Checked specifically, because this is the recurring builder bug. Reference 35's consortium author is rendered correctly — "Metwaly A, Kriaa A, Hassani Z, et al. (Human Microbiome Action Consortium)" — with no dropped author and no leading comma. No other row is affected.
• Rendering defects. None found. This edition was hand-assembled rather than produced by the reference builder, and carries none of the missing-space, empty-DOI or "ErratumIn" artifacts seen elsewhere in the corpus.
• Revision splice contradiction. Checked for the systematic Oct-9 artifact in which a spliced "Consensus consistency validation" block contradicts preserved reviewer notes asserting that no Consensus work was done. Absent here: this document contains no such assertion, and the spliced block states its own limits ("does not establish historical Consensus use"). No scoping edit was needed.
• Score card review_metadata.reviewer. Left as issued; corpus-wide house provenance.

Residual limitations

Two load-bearing items could not be re-verified in this pass. The first has since been resolved as finding F1 above and is kept here because it states the limit of what is checkable. The Divergence artifact rested on a quotation from the paywalled 2025 Human Microbiome Action Consortium Delphi consensus statement — that the evidence "strongly supported" a causal role in inflammatory bowel disease, obesity, type 2 diabetes and liver diseases. The PubMed abstract of that record does not contain the disease-specific sentence and emphasises instead that establishing cause and effect is extremely difficult; the record has no PMC copy and the publisher text was unreachable here. The ledger declares full_text_partial access and the review quotes a second, more cautious sentence from the same source, so the attribution was supported by declaration alone. The lead has since confirmed that the accessible record does not support it; C35 is now marked unverified and Divergence is Contested (finding F1). A reader with institutional access to the paywalled statement could still settle what it says. Divergence modifies no letter, so nothing below the headline turns on it. Second, reference 17 — the 2026 active-comparator minimal-hepatic-encephalopathy trial supplying the "68.3% vs 66.1%; p=.80" contrast — is not indexed in PubMed and could not be opened here. The external red-team trail this edition cites (PROCESS.md and redteam/) is not present in the deliverables package; it exists at reviews/gut-microbiome-systemic-causality/, so the pointer resolves only outside this package. Access depth remains the dominant constraint: 34 of 39 rows were read at abstract depth.

Approval: Following adversarial review and the corrections listed above, the revised document is approved. — Claude Opus 5.5 (Anthropic), 2026-10-14.

REFERENCES

These primary reports ground the claims above. Links point to PubMed records where available and directly to the DOI record otherwise; evidence depth is stated in the ledger.

1. Wu HJ et al. Gut-residing segmented filamentous bacteria drive autoimmune arthritis via T helper 17 cells. Immunity. 2010;32:815–827. PMID 20620945 · DOI 10.1016/j.immuni.2010.06.001.
2. Ridaura VK et al. Gut microbiota from twins discordant for obesity modulate metabolism in mice. Science. 2013;341:1241214. PMID 24009397 · DOI 10.1126/science.1241214.
3. Hsiao EY et al. Microbiota modulate behavioral and physiological abnormalities associated with neurodevelopmental disorders. Cell. 2013;155:1451–1463. PMID 24315484 · DOI 10.1016/j.cell.2013.11.024.
4. Manfredo Vieira S et al. Translocation of a gut pathobiont drives autoimmunity in mice and humans. Science. 2018;359:1156–1161. PMID 29590047 · DOI 10.1126/science.aar7201.
5. Yoshimoto S et al. Obesity-induced gut microbial metabolite promotes liver cancer through senescence secretome. Nature. 2013;499:97–101. PMID 23803760 · DOI 10.1038/nature12347.
6. Bloom SM et al. Commensal Bacteroides species induce colitis in host-genotype-specific fashion in a mouse model of inflammatory bowel disease. Cell Host Microbe. 2011;9:390–403. PMID 21575910 · DOI 10.1016/j.chom.2011.04.009.
7. Gray SM et al. Mouse adaptation of human inflammatory bowel diseases microbiota enhances colonization efficiency and alters microbiome aggressiveness depending on the recipient colonic inflammatory environment. Microbiome. 2024;12(1):147. PMID 39113097 · DOI 10.1186/s40168-024-01857-2.
8. Kootte RS et al. Improvement of insulin sensitivity after lean donor feces in metabolic syndrome is driven by baseline intestinal microbiota composition. Cell Metabolism. 2017;26:611–619.e6. PMID 28978426 · DOI 10.1016/j.cmet.2017.09.008.
9. Yu EW et al. Fecal microbiota transplantation for the improvement of metabolism in obesity: The FMT-TRIM double-blind placebo-controlled pilot trial. PLoS Med. 2020;17(3):e1003051. PMID 32150549 · DOI 10.1371/journal.pmed.1003051.
10. Koopen AM et al. Effect of Fecal Microbiota Transplantation Combined With Mediterranean Diet on Insulin Sensitivity in Subjects With Metabolic Syndrome. Front Microbiol. 2021;12:662159. PMID 34177842 · DOI 10.3389/fmicb.2021.662159.
11. Wu Z et al. Fecal microbiota transplantation reverses insulin resistance in type 2 diabetes: A randomized, controlled, prospective study. Front Cell Infect Microbiol. 2023;12:1089991. PMID 36704100 · DOI 10.3389/fcimb.2022.1089991.
12. Piwchan S et al. Fecal Microbiota Transplantation by Rectal Enema Improves Short-Term Insulin Resistance in Metabolic Syndrome: A Pilot Randomized Controlled Trial. J Obes Metab Syndr. 2025;34(4):434–445. PMID 40916606 · DOI 10.7570/jomes25014.
13. da Ponte Neto AM et al. Fecal microbiota transplantation in patients with metabolic syndrome and obesity: a randomized controlled trial. World J Clin Cases. 2023;11(19):4612–4624. PMID 37469721 · DOI 10.12998/wjcc.v11.i19.4612.
14. van Nood E et al. Duodenal infusion of donor feces for recurrent Clostridium difficile. N Engl J Med. 2013;368:407–415. PMID 23323867 · DOI 10.1056/NEJMoa1205037.
15. Feuerstadt P et al. SER-109, an Oral Microbiome Therapy for Recurrent Clostridioides difficile Infection. N Engl J Med. 2022;386(3):220–229. PMID 35045228 · DOI 10.1056/NEJMoa2106516.
16. Bajaj JS et al. Microbiota transplant for hepatic encephalopathy in cirrhosis: The THEMATIC trial. J Hepatol. 2025;83:81–91. PMID 39800192 · DOI 10.1016/j.jhep.2024.12.047.
17. Agarwal A et al. Impact of two-session fecal microbiota transplantation on minimal hepatic encephalopathy in cirrhosis. World J Hepatol. 2026;18(4):116153. DOI 10.4254/wjh.v18.i4.116153.
18. Bruggeman A et al. Safety and efficacy of faecal microbiota transplantation in patients with mild to moderate Parkinson's disease (GUT-PARFECT): a double-blind, placebo-controlled, randomised, phase 2 trial. EClinicalMedicine. 2024;71:102563. PMID 38686220 · DOI 10.1016/j.eclinm.2024.102563.
19. Scheperjans F et al. Fecal Microbiota Transplantation for Treatment of Parkinson Disease: A Randomized Clinical Trial. JAMA Neurol. 2024;81(9):925–938. PMID 39073834 · DOI 10.1001/jamaneurol.2024.2305.
20. Wang J et al. Colonoscopic fecal microbiota transplantation for Mild-to-Moderate Parkinson's Disease: A randomized controlled trial. Brain Behav Immun. 2025;130:106086. PMID 40848995 · DOI 10.1016/j.bbi.2025.106086.
21. Zhang R et al. Gut microbiota modulation via repeated donor fecal transplantation improves motor and gastrointestinal symptoms in drug-naïve Parkinson's disease: a randomized phase 2 trial. Signal Transduct Target Ther. 2026;11(1):94. PMID 41826284 · DOI 10.1038/s41392-026-02604-9.
22. Figura M et al. Safety and Efficacy of Fecal Microbiota Transplantation in Alleviating Symptoms of Parkinson's Disease: A Randomized, Placebo-Controlled, Double-Blinded Study. Ann Neurol. 2026;100(1):10–21. PMID 41674471 · DOI 10.1002/ana.78153.
23. Fretheim H et al. Faecal microbiota transplantation in patients with systemic sclerosis and lower gastrointestinal tract symptoms in Norway (ReSScue): a phase 2, randomised, double-blind, placebo-controlled trial. Lancet Rheumatol. 2025;7(5):e323–e332. PMID 39900089 · DOI 10.1016/S2665-9913(24)00334-5.
24. Forslund K et al. Disentangling type 2 diabetes and metformin treatment signatures in the human gut microbiota. Nature. 2015;528(7581):262–266. PMID 26633628 · DOI 10.1038/nature15766.
25. Vandeputte D et al. Stool consistency is strongly associated with gut microbiota richness and composition, enterotypes and bacterial growth rates. Gut. 2016;65(1):57–62. PMID 26069274 · DOI 10.1136/gutjnl-2015-309618.
26. Falony G et al. Population-level analysis of gut microbiome variation. Science. 2016;352:560–564. PMID 27126039 · DOI 10.1126/science.aad3503.
27. Wu H et al. Metformin alters the gut microbiome of individuals with treatment-naive type 2 diabetes, contributing to the therapeutic effects of the drug. Nat Med. 2017;23(7):850–858. PMID 28530702 · DOI 10.1038/nm.4345.
28. Sampson TR et al. Gut microbiota regulate motor deficits and neuroinflammation in a model of Parkinson's disease. Cell. 2016;167(6):1469–1480.e12. PMID 27912057 · DOI 10.1016/j.cell.2016.11.018.
29. Neyrinck AM et al. Fecal transplantation from humans with obesity to mice drives a selective microbial signature without impacting behavioral and metabolic health. Sci Rep. 2025;15(1):15455. PMID 40316655 · DOI 10.1038/s41598-025-99047-z.
30. Daoust L et al. Gnotobiotic mice housing conditions critically influence the phenotype associated with transfer of faecal microbiota in a context of obesity. Gut. 2023;72(5):896–905. PMID 36881441 · DOI 10.1136/gutjnl-2021-326475.
31. Vrieze A et al. Transfer of intestinal microbiota from lean donors increases insulin sensitivity in individuals with metabolic syndrome. Gastroenterology. 2012;143(4):913–916.e7. PMID 22728514 · DOI 10.1053/j.gastro.2012.06.031.
32. David LA et al. Diet rapidly and reproducibly alters the human gut microbiome. Nature. 2014;505(7484):559–563. PMID 24336217 · DOI 10.1038/nature12820.
33. Roager HM et al. Colonic transit time is related to bacterial metabolism and mucosal turnover in the gut. Nat Microbiol. 2016;1(9):16093. PMID 27562254 · DOI 10.1038/nmicrobiol.2016.93.
34. Prosty C et al. Causal role of the gut microbiome in certain human diseases: a narrative review. eGastroenterology. 2024;2(3):e100086. PMID 39944364 · DOI 10.1136/egastro-2024-100086.
35. Metwaly A, Kriaa A, Hassani Z, et al. (Human Microbiome Action Consortium). A Consensus Statement on establishing causality, therapeutic applications and the use of preclinical models in microbiome research. Nat Rev Gastroenterol Hepatol. 2025;22(5):343–356. PMID 40033063 · DOI 10.1038/s41575-025-01041-3.
36. Sanna S et al. Causal relationships among the gut microbiome, short-chain fatty acids and metabolic diseases. Nat Genet. 2019;51(4):600–605. PMID 30778224 · DOI 10.1038/s41588-019-0350-x.
37. Koeth RA et al. Intestinal microbiota metabolism of L-carnitine, a nutrient in red meat, promotes atherosclerosis. Nat Med. 2013;19(5):576–585. PMID 23563705 · DOI 10.1038/nm.3145.
38. Bajaj JS et al. Fecal microbiota transplant from a rational stool donor improves hepatic encephalopathy: a randomized clinical trial. Hepatology. 2017;66(6):1727–1738. PMID 28586116 · DOI 10.1002/hep.29306.
39. Davar D et al. Fecal microbiota transplant overcomes resistance to anti-PD-1 therapy in melanoma patients. Science. 2021;371(6529):595–602. PMID 33542131 · DOI 10.1126/science.abf3363.

Evidence ledger

Claim-source crosswalk

All load-bearing claims in this review have at least one listed source. A paper may support a claim about a bounded setting while explicitly failing to support the broad claim. The limitations column and scope notes are part of the evidence interpretation, not boilerplate.

Scope and evidence-status notes

• Human observational signatures support SC1 association, not causal direction. Reverse causation, medication, transit, diet, disease stage, and host physiology remain live alternatives, and two of them have been demonstrated experimentally rather than inferred.
• Rodent work supports causal effects in tested models; "defined composition" is not interchangeable with every human microbiome state, and artificial colonization is not endogenous human spread. Transfer of the same kind of community can fail, or reverse, with diet and housing.
• Recurrent CDI supports restoration of colonization resistance in a post-antibiotic depleted niche. It is evidence of causation for that local infectious outcome, not for systemic health generally, and it is reported here rather than aggregated.
• Human FMT trials use heterogeneous and mostly undefined donor communities. Positive trials may show that a gut intervention can modify an established condition, but they do not identify a responsible composition, establish that the change preceded disease, or show that it is a primary determinant.
• The human genetic instrument and the human withdrawal experiment are real field-valid equivalents and are scored as such. Each stops at a named cap or a named scope limit, and saying where is more informative than recording that neither exists.
• Negative and positive trials are both retained. The metabolic and Parkinson evidence is mixed; effects should not be counted as mutually independent replications of a single cross-system hypothesis.
• Tier B sources appear only as field framing and as the Divergence basis, and carry no figure.
• No calibration target was provided in the schema package or prompt. This is a non-calibration review with an explicit reason, not a failed blind calibration.

Documentary record

Every grade on this site is reproducible from its record. The files below are the posted edition exactly as assessed.

• ADVERSARIAL_REVIEW.md
• SHA256.txt
• adversarial-findings.json
• codex-correction-provenance.json
• consensus-addendum.md
• consensus-validation.json
• edition-approval-audit.json
• gut-microbiome-systemic-causality.review.md
• gut-microbiome-systemic-causality.scorecard.json
• publication-correction.json
• research-update-2026-10-05.bib
• research-update-2026-10-05.html
• research-update-2026-10-05.json
• research-update-2026-10-05.md
• research-update-2026-10-05.ris
• validation-report.json
• validation-round2-2026-10-14.json
• validation.txt
• video-insert-2026-10-05.md

Sources: 39 · load-bearing claims: 39 · verified: 38 · unverified: 1 · contradicted: 0 · schema 5.4