In plain terms

The claim being assessed: Helicobacter pylori infection is a major upstream cause of human peptic ulcer disease.

Where the evidence stands: Θ — Very well established. It predicts outcomes accurately and there are few unexplained exceptions left.

What this mechanism does: This is the thing that sets the process off.

Strongest evidence for it: H. pylori infection is a major upstream cause of peptic ulcer disease: Θ, High confidence — its predictions hold across removal, prevention and temporal-order evidence, with one unresolved exception in long-term aspirin users.

What is still missing: HEAT's loss of early protection against ulcer bleeding after 2.5 years in older aspirin users is unexplained; it is the one exception that keeps durable generalization (SC3) at Η.

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: Leading among models compared

What this means for ulcers and treatment

This is an example where a causal explanation has changed care: H. pylori is a major cause of many peptic ulcers, but not all of them. Anti-inflammatory pain medicines such as aspirin or ibuprofen, other illnesses, and less common causes can also matter; many infected people never develop an ulcer.

Implications for health and treatment: When testing confirms H. pylori in an appropriate clinical setting, treating the infection can help an ulcer heal and reduce the chance it returns. The evidence does not mean that every stomach symptom is an ulcer or that every ulcer has the same cause. It supports finding the cause rather than relying on acid suppression alone when infection is present.

Limits for individual care: Treatment choice depends on allergy history, previous antibiotic exposure, local resistance patterns, other medicines, and whether there are complications such as bleeding. Do not self-start, share, or stop antibiotics or acid-suppressing medicine based on this page. Urgent symptoms such as vomiting blood, black stools, fainting, or severe persistent pain require prompt medical assessment.

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 23 sources and 34 individually checked claims. 34 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	H. pylori infection is a major upstream causal contributor to human peptic ulcer disease rather than a bystander of established mucosal damage.	primacy	Θ	Card ↗
SC2	Eradication treatment materially reduces subsequent ulcer recurrence in patients with infection and an ulcer documented before treatment.	necessity	Θ	Card ↗
SC3	The causal contribution extends to both gastric and duodenal ulcers, while the magnitude and durability of benefit vary across clinical contexts.	generalization	Η	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 ↗

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.

Revised 27 September 2026 by red-team pass-01 (Claude, Anthropic, independent model): the level rises from Η to Θ because the edition's own Θ blockers, re-examined against evidence it had not included, reduce to one named exception — HEAT's later period. The full trail is in PROCESS.md and redteam/.

In plain terms

The claim is that a stomach bacterium, Helicobacter pylori, is one of the main causes of peptic ulcers — open sores in the lining of the stomach or the first part of the small intestine. It is not a claim that the bacterium causes every ulcer: some ulcers come from painkillers of the aspirin and ibuprofen family, and some have no known cause. Many people who carry the bacterium never develop an ulcer.

The evidence is very strong. In many clinical trials, clearing the infection with antibiotics made ulcers far less likely to come back than a course of acid-reducing treatment alone, and this has been repeated across countries and ulcer types. Clearing it before people started anti-inflammatory painkillers prevented new ulcers, infection was found years before ulcers appeared, and deliberately swallowing the bacterium inflamed the stomach lining of healthy volunteers. This review therefore places the claim on the second-highest step of its nine-step scale: its predictions have held up with very few unexplained exceptions. The main exception is a large trial in older people taking daily aspirin, where protection against ulcer bleeding was clear in the first years but was not seen later, for reasons its investigators could not explain.

The top step is not given because the bacterium is neither needed for every ulcer nor enough to cause one on its own. The grade describes the strength of the evidence behind the causal claim; it is not a judgement on any person or group, and this review makes no treatment recommendation.

AT A GLANCE

H. pylori infection is a major upstream cause of peptic ulcer disease; its predictions hold across removal, prevention and temporal-order evidence, with one unresolved exception: HEAT's loss of protection in long-term aspirin users.

Sources: 23 records describing 22 studies (20 Tier A) · Load-bearing claims: 34 · Verified: 34 · Unverified: 0 · Contradicted: 0 · Gate: passed (validator 2.4: 0 failures; 2 disclosed calibration warnings).

Adversarial review: Authoring model of record: Sol 6.1 (this edition's own provenance records editions A and B by OpenAI Codex and a Claude red-team synthesis pass of 2026-09-27; recorded, not edited). Adversarially reviewed by Claude Opus 5.5 (Anthropic) on 2026-10-14. 3 corrections applied (see ADVERSARIAL REVIEW AND APPROVAL). Status: Approved as revised.

• The experiment that decides it: a replicated long-term eradication trial in people starting aspirin or NSAIDs, with repeat testing for reinfection, would show whether HEAT's later loss of protection is real — if it is, the broad claim returns to Η.
• What the field believes vs. what the evidence shows: the field presents H. pylori as the main etiologic factor of peptic ulcer disease (implied Θ); the evidence supports Θ; gap 0, although the field's association figures overstate coverage in low-prevalence settings 21.
• Best-supported rival model: M5, the heterogeneous multiple-cause model that contains the nominal (Η).

What changed in this pass. Edition B stopped SC1 at Η on three grounds: HEAT durability, stress-associated residual risk, and the absence of a prospectively validated susceptibility rule. The second and third do not bind a nonexclusive major-cause claim — B's own steelman disclaims the third. HEAT's later window is a single, imprecise, unexplained exception and is carried by name. Evidence B had not included — Cochrane syntheses of recurrence and rebleeding 15 16, a double-blind intention-to-treat prevention trial 18, infection preceding ulcer in stored sera 17 and a community mass-eradication study 19 — supplies the replication and the tested predictions Θ asks for. Evidence against was added as well 20 21 22. SC1 moves Η → Θ; SC2 holds at Θ; SC3 holds at Η. Edition B's case for Η is carried as dissent below.

What changed in edition B (retained). The undefined susceptibility qualifier is removed. Marshall's human gastritis challenge is now a listed supporting line, while the number of direct ulcer lines stays at two. SC2 is typed reversibility (in schema 5.4 terms, a necessity claim). HEAT's later loss of established benefit is treated as an unresolved exception, not dismissed as an already-explained scope boundary. The added 2025 HEAT report is a second report of the same trial. 13 14

Meaning of the finding. The user's “a primary cause” claim is strongly supported as a major, nonexclusive upstream contribution. It does not imply that every ulcer requires infection or that infection is numerically the leading cause in every population. Matched acid suppression in the Hentschel trial, replicated recurrence effects and complementary induction support causation. 1 2 7

Rubric: 04-microbiome-systemic (imperfect fit; see routing) · HSEF v5.4 · schema 5.4 · 2026-09-27 · Previous level: Η (edition B; edition A was Θ) Entry type: claim · Calibration entry: yes — unblinded reference case, not a completed blinded calibration run.

Subclaim score matrix

SC1 reaches Θ: its one unresolved within-scope exception is HEAT's later period. SC3 stops at Η because it asserts generalization of duration, and that is exactly where HEAT failed. SC2 retains Θ for its narrower recurrence prediction; Ω is blocked because persistent infection is not shown to be sufficient for recurrence.

Retained pre-registered aggregation: primary-subclaim, SC1. The rule was pre-registered in edition A before any subclaim was scored and retained by B; the hypothesis index prescribes none for this entry. SC1 = Θ, so the overall result is Θ. Scope-matched also returns Θ. Weakest-link would return Η, because SC3 is Η: the letter is rule-dependent by one rung. Weakest-link is not the appropriate rule — the claim does not assert joint necessity of durable benefit in every context — and it was not the rule registered. Under edition B's scores all three rules returned Η.

Scope-field correction (edition B, retained). The original JSON already populated claimed_scope for all three subclaims, which is the field in the supplied template. It had no scope key. The revision adds a checked scope alias and structured operational_scope. Pass-01 migrated the card to schema 5.4: reversibility is kept as a descriptive label on SC2, whose 5.4 type is necessity.

Decomposition note (pass-01). SC1 is coextensive with the graded claim, so primary-subclaim aggregation amounts to scoring the claim directly, with SC2 and SC3 auditing its recurrence and generalization legs. This is recorded, not changed.

COMPETING MECHANISTIC MODELS

The models overlap. Pass-01 ranks them by strength of causal evidence, as the template requires; edition B ranked by explanatory coverage, which put M5 first. It is not a probability ranking, a population-frequency estimate, or a contest in which only one cause may exist. The level is unchanged by the tournament in either ordering.

M1 — H. pylori as a primary cause

Helicobacter pylori infection is a major upstream cause of human peptic ulcer disease. Removal prevents recurrence and rebleeding across randomized trials, removal before NSAIDs prevents incident and complicated ulcers, infection precedes ulcer in stored sera, and organism addition produces gastritis in people and ulcers in gerbils. HEAT's unexplained later period is the one unresolved exception. 1 4 6 7 15 16 17 18

M5 — Heterogeneous multiple-cause model

Infection, medication injury and additional causes produce ulcers through host-dependent mucosal injury and acid-peptic exposure. Infection-removal and medication-addition experiments converge on multiple causal routes. Merely allowing other causes does not explain idiopathic disease or HEAT durability, so wider coverage does not itself earn a higher letter; M5 is scored at Η because its medication and idiopathic routes are scored at Η or below. 1 9 12 14 21

M2 — NSAID/aspirin-associated injury

Ulcerogenic medication is a causal driver in exposed patients, including patients whose H. pylori infection has been treated. Exposure addition with endoscopic disease and a distinct protective intervention support causal injury in the exposed context; after eradication, aspirin users with prior complications still recur without acid suppression. This compressed dossier assigns Η and does not claim a full audit of medication-specific risk prediction. It coexists with infection causality. 5 12 22

M3 — Autonomous acid-peptic model with infection as passenger

Acid-peptic injury explains the disease while infection has no upstream causal role. Acid contributes to injury, but that is not the passenger proposition. Physiological plausibility supports Γ at most for the exclusive explanation, and clinical infection perturbations contradict its central exclusion. The letter is an evidence ceiling, not an endorsement of the contradicted passenger clause. 1 2 8

M4 — Psychological stress as the dominant ulcer cause

Psychological stress is the main cause of peptic ulcer disease, with infection largely secondary. The primary evidence examined is observational; universal association cap Β applies. It does not exclude a real stress contribution or behavioral mediation, but does not establish dominance. Psychological stress is not conflated with critical-illness stress-related mucosal disease. 10 1

Tournament result. M1 has the strongest causal evidence; M5, which contains it, has the widest coverage. No study can separate compatible propositions; the studies that show where M1 alone under-covers are the medication-context trials — Chan et al. 2001, Lai et al. 2002 and the Vergara meta-analysis — in which eradication alone was not enough while NSAID or aspirin exposure continued. The nominal HSEF score is unchanged by this comparison. 5 20 22

HYPOTHESIS STATE EVALUATION FRAMEWORK (HSEF v5.4)

Claim restatement, mechanism and routing

Original: “Helicobacter pylori infection is a primary cause of peptic ulcer disease.”

Graded proposition: “Helicobacter pylori infection is a major upstream cause of human peptic ulcer disease.”

The original is gradeable. The previous addition “in susceptible hosts” was not operationally defined and could select precisely the people who later develop ulcers. It is removed from the proposition and recorded in scope_qualifiers as not determinable before the outcome, action deleted. The claim remains nonexclusive, but is not insulated from failed predictions. Pass-01 checked Step 0 and holds the restatement: it is the strong form the field defends.

Eligibility must be determined before the assessed outcome: objectively documented infection, baseline ulcer status/site, medication exposure, age and clinical setting. For recurrence, an ulcer that already exists at enrollment is a legitimate baseline characteristic; future recurrence or response to eradication is not. Comparisons selected by successful post-randomization clearance are distinguished from randomized treatment contrasts.

CagA status, acid-output phenotype and host genotype could serve as prespecified effect modifiers. This dossier does not establish a validated threshold for any of them as a susceptibility test. They cannot be invoked after a failed prediction to relabel a participant out of scope — and, by the same token, their absence is not a reason to deny the claim a level it otherwise earns.

A prediction that can fail. In prospectively defined infection-positive populations, repeated, adequately powered, well-controlled comparisons show that durable removal does not reduce adjudicated ulcer outcomes over a prespecified interval, with confidence intervals excluding a prespecified clinically meaningful reduction. Such a result contradicts that population-and-time prediction; repeated failures across the principal settings would challenge the broad major-cause account. Affected participants must not be excluded afterward as not susceptible. A single context-specific null does not erase positive evidence elsewhere. The effect size of interest and its follow-up horizon must be registered before outcomes are analyzed; no retrospectively chosen cutoff is presented here as an established rule.

Infection-related inflammation and altered gastric physiology provide a coherent route, but the clinical ulcer outcomes carry the disease-level inference. Gastritis and acid output alone are not ulcers. 7 8 13

No pathogen-specific index entry was present. No supplied rubric fits well. Rubric 04 is the closest and contains an explicit eradication provision; its microbiome/systemic emphasis is an imperfect fit to a defined local gastric pathogen. Rubric 05 (immunology) was considered and rejected: it governs immune-pathway claims, not infection-to-disease causation. Universal rules are applied with that limitation made explicit. The archive supplies evaluation criteria, not separate authorization for unrelated actions.

Clinical evidence and what it establishes

The cleanest recurrence contrast is randomized treatment, not successful-eradication status. Hentschel et al. used antibiotics without bismuth and gave ranitidine to both groups. Among participants followed for a year, ulcer recurrence was 4/50 (8%) versus 42/49 (86%). The frequently quoted 2% versus 85% comparison instead divides participants by post-treatment eradication status. Both are informative, but only the first preserves the original treatment-group comparison, and even it uses follow-up rather than all-randomized denominators. Antibiotic effects on other organisms remain a possible failure mode; the other evidence classes narrow that concern. 1

The finding generalizes beyond duodenal recurrence. Graham et al. observed lower recurrence of both gastric and duodenal ulcers after an eradication regimen, with no maintenance antiulcer treatment after healing; its triple therapy also contained bismuth subsalicylate, so the same mucosal-action caveat noted below applies, and the bismuth-free attribution rests on the duodenal trials. Sung et al. provides a distinct gastric-ulcer test using antibacterial therapy without acid-suppressing medication; the one recurrence in its antibacterial arm was in a patient still infected. Its recurrence estimate comes from a small one-year subset, and its bismuth-containing regimen has mucosal effects; those limitations are retained rather than presenting it as an antibiotics-only proof. 2 3

The recurrence line is replicated, not resting on three trials (pass-01). A Cochrane synthesis of randomized trials found eradication superior to no treatment in preventing both duodenal and gastric ulcer recurrence; it rated that evidence very low quality on trial conduct and found no evidence that eradication outperforms maintenance ulcer-healing drugs for duodenal recurrence — a comparison the causal claim does not need to win. A second Cochrane synthesis found eradication better than non-eradicating antisecretory therapy, with or without long-term maintenance, at preventing recurrent ulcer bleeding, a hard endpoint, and attributed some rebleeding to recurrence of infection. Both pool line 1; neither is a new line. 15 16

Initiation is addressed, not inferred solely from treating established ulcers. Chan et al. 1997 studied infected people without baseline ulcers before their first NSAID exposure. Pretreatment reduced incident ulcers. Chan et al. 2002 repeated the design with omeprazole in both arms and placebo antibiotics: by intention to treat, 6-month ulcer probability was 12.1% (95% CI 3.1–21.1) versus 34.4% (21.1–47.7), and complicated-ulcer probability 4.2% versus 27.1%. HEAT later found an early reduction in clinically important ulcer bleeding in infected older aspirin users. None of these trials establishes identical prevention in all populations: the first two are medication-exposed contexts and HEAT did not show sustained protection in its later period. 4 18 6

Infection precedes the ulcer (pass-01). In a nested case-control study within 5443 Japanese-American men whose sera were stored in 1967–1970, pre-existing infection predicted later gastric ulcer (OR 3.2, 95% CI 1.6–6.5; 150 cases) and duodenal ulcer (OR 4.0, 1.1–14.2; 65 cases); risk rose with antibody level and held for diagnoses 10 or more years after sampling. This is association, capped at Β on its own, but it rules out reverse causation — infection colonizing damaged mucosa — by a route the trials do not use. 17

Population-scale removal (pass-01). After community mass eradication in a Taiwanese population, the reported reduction in peptic ulcer disease was 67.4% (95% CI 52.2–77.8%), with re-infection or recrudescence estimated at 1% per person-year. Periods are compared without a concurrent control, so this supports, rather than carries, the level. 19

Organism addition supplies a different causal test. Defined-strain inoculation in gerbils caused gastric ulceration; superficial duodenal ulceration depended on strain and host age. That makes conditional initiation plausible without administering antibiotics. It does not prove universal human sufficiency, and two superficial duodenal-ulcer observations should not be inflated into uniform penetrance. 7

Gastritis and physiology are supporting endpoints. El-Omar et al. documented changes in gastrin-mediated acid secretion after eradication, consistent with infection altering an ulcer-relevant physiological pathway. Marshall's ingestion experiment established gastritis, not an ulcer; a second self-ingestion produced culture-positive gastritis with raised fasting gastric pH that persisted until bismuth treatment, again without a reported ulcer. Neither meets the clinical-ulcer perturbation boundary. 8 13 23

Limits and genuine alternative causes

The presence of a primary cause does not eliminate other causes. Randomized ibuprofen exposure produced more ulcers than placebo. In patients continuing naproxen after a bleed, maintenance omeprazole outperformed eradication alone; in the aspirin stratum of the same trial, eradication alone matched omeprazole over six months (1.9% versus 0.9%; absolute difference 1.0%, 95% CI −1.9 to 3.9). After eradication, aspirin users with prior complications still recurred on placebo — 9 of 61 (14.8%) versus 1 of 62 (1.6%) with lansoprazole — and 4 of those 10 had recurrent infection. A meta-analysis of NSAID users found eradication reduced ulcers in NSAID-naive patients but not in those already treated with NSAIDs. These results map a medication co-cause by a variable known at baseline; they constrain exclusivity and the benefit of eradication alone, not the infection's causal role. 12 5 22 20

A prospective idiopathic-ulcer cohort documented disease in patients classified as uninfected and unexposed to NSAIDs. Across six US placebo-controlled trials of non-NSAID duodenal ulcer, 73% (1737 of 2394 evaluable patients) were confirmed infected under a confirmatory definition — positive culture, or positive histology and CLOtest together, with at least two negative tests required to call a patient uninfected — a quarter were not. These constrain universal necessity and show that the near-universal association figures once quoted do not hold in low-prevalence settings; they do not erase the positive infection experiments. False-negative infection tests, medication-history errors and comorbidity remain relevant when interpreting both. 9 21

Infected healthy volunteers, infected controls in stored-sera studies and strain/age-dependent experimental disease show why infection is not deterministic. This review therefore makes no claim about an exact lifetime ulcer percentage among infected people. Psychological stress has prospective associational evidence and could contribute through biological or behavioral routes; the retrieved evidence does not establish it as the dominant cause. This is separate from stress-related mucosal disease in critical illness. 8 17 7 10

HEAT and stress: explicit Θ-versus-Η test

HEAT is a genuine unresolved duration exception. Among 5,352 randomized infected aspirin users aged at least 60, ulcer-bleeding protection was established in the first 2.5 years (HR 0.35, 95% CI 0.14–0.89), but not later (HR 1.31, 0.55–3.11). The full report gives the events: 6 versus 17 in the first period, 12 versus 9 thereafter. The later interval is too imprecise to establish either no effect or harm. The extended report gives overall HR 0.69 (0.38–1.25), but proportional hazards failed; that pooled estimate must not be presented as a constant effect. The split was selected before unblinding. Investigators leave the reason for the changing effect unclear. In a 10% repeat-breath-test sample a median 3.95 years after randomisation, the test was negative in 146 of 161 of the eradication group and 41 of 161 of the control group, so reinfection is an unlikely explanation and control-arm clearance can dilute but not reverse the contrast. 6 14

This does not overturn clinical recurrence evidence or show that infection was never causal, and counting the late period as out of scope only after seeing it would recreate the circularity problem. It stays in scope and it stays unexplained. Edition B concluded that it defeats Θ's “minimal unresolved exceptions” requirement for the broad SC1 and SC3 scope. Pass-01 concludes that it is that requirement's one exception for SC1: a single window of 21 events, in established aspirin users where a medication co-cause is independently mapped 20 22, against removal, prevention, temporal-order and population evidence that confirms the claim's predictions 15 16 17 18 19. Θ asks for minimal, not zero, unresolved exceptions. For SC3, which asserts durable generalization, the same finding is binding, and SC3 stays at Η. No second automatic downgrade is applied for the same finding.

Stress supports heterogeneity, with a different evidential weight. Levenstein's prospective cohort found a stress–ulcer association across infection/NSAID strata that persisted with adjustment. Its observational design leaves confounding and behavioral pathways unresolved. It is not a controlled demonstration that stress is the dominant ulcer cause. It is an independent risk factor, which a nonexclusive causal claim does not deny, and so it is not an exception to that claim — a point on which pass-01 departs from edition B. 10

SC2's Θ is restricted to the recurrence populations and approximately one-to-two-year horizons represented by its trials. HEAT asks about aspirin-associated bleeding prevention in a different baseline population and over longer follow-up. The narrower SC2 score must not be promoted to a broad or lifelong prediction. 1 2 3

Evidence inventory and causal tally

“Strong” describes relevance to the stated endpoint, not every possible extrapolation.

Audited tally: seven decisive perturbations support the nominal claim — six human randomized trials with ulcer or ulcer-bleeding endpoints 1 2 3 4 6 18 and one animal induction experiment 7. Edition B counted eight, including the boundary comparison 5 and the rival-cause experiment 12; its prose labelled them correctly but its evidence cards marked them decisive and “supporting”, which pass-01 corrected. 14 adds information about 6, not another experiment. The pooled syntheses and the community study are replication within line 1.

There are 23 records of 22 studies; three are Tier B syntheses carrying no figure. The claim register distinguishes checked factual premises from reviewer synthesis. Verification used primary abstracts and indexed metadata, plus targeted passages of the HEAT full report. No independent second-reader adjudication or complete full-text risk-of-bias audit is claimed.

Cap audit

The following verbatim universal triggers were evaluated for each of SC1, SC2 and SC3. No ceiling binds the complete convergent evidence set.

Cap precedence and rubric fit. Rubric 04 retains transfer/consortium wording but also expressly states: “An eradication trial that abolishes a disease is therapeutic perturbation against a hard endpoint and can carry a claim to Ζ or above, notwithstanding cap 5's restriction on forced-transfer paradigms — the cap binds transfer experiments, not eradication.” Its heading also identifies “eradication trials with a hard clinical endpoint” as field-valid equivalents. This is not interpreted to require literal zero residual ulcers: the universal Δ/Η criteria concern causal contribution and trajectory change, and the user does not assert universal necessity. The clinical removal studies cross the human boundary; inoculation 7 remains evidence within its animal scope. The human conclusion does not rest on forced transfer alone.

This is an explicit interpretation of an internally uneven rubric, not an assertion that the old consortium checklist was literally performed in people. Clinical rescue here means improvement after pathogen removal; deliberate human ulcer rechallenge is not a prerequisite for a major-cause claim.

Β/Γ/Δ boundary check

The applicable wording is: “Field-valid equivalents in this domain: strain-resolved colonization with rescue logic, antibiotic or eradication natural experiments, dose-response across independent cohorts, and eradication trials with a hard clinical endpoint.”

The legacy definition is also recorded: “Gnotobiotic or transfer experiment showing phenotype movement, at minimum” and “Even without defined-consortium precision or robust rescue, a transfer paradigm in vivo is the minimum that crosses this threshold.” That wording governs transfer experiments, while the explicit human eradication provision governs the clinical evidence.

The rubric's Γ hard rules, quoted verbatim, are “Association plus cell-culture metabolite assays”; “Association plus in vitro immune readouts”; and “Any framework that has not crossed the transfer/rescue threshold in vivo.” None characterizes the full clinical evidence set, under the eradication provision and the scope resolution above.

The later HEAT period, the active-comparator result in 5, and surrogate-only studies are not silently converted into positive efficacy trials for the nominal claim.

Threshold ladder, applied to each subclaim

The table retains the supplied rubric's quoted criteria; the universal Θ definition additionally requires “High predictive accuracy with minimal unresolved exceptions. The mechanism accounts for most major observations in its claimed scope.” The first unmet rung determines the result.

Next requirement for SC3, and what would return SC1 to Η: a replicated long-term trial in a baseline-defined medication context — for example aspirin or NSAID starters — in which eradication's benefit is either sustained (SC3 could reach Θ) or lost again in durably cleared participants (SC1 would return to Η). A formal individual-risk metric is not required for either.

For the narrower SC2, Ω remains unmet: substantial recurrence reduction is not universal necessity and sufficiency. Recognizing this does not turn the original major-cause claim into a universal one.

Falsification audit and pitfalls

None of the retrieved results categorically falsifies the nonexclusive major-cause claim. Defined-strain animal induction and replicated clinical interventions supply positive tests. Community-transfer, taxon-reinstatement and low-biomass-contamination patterns are not silently imposed as requirements for a named gastric pathogen.

Residual ulcers after treatment can arise through other causal routes and therefore do not by themselves negate the claim. They must nevertheless be reported. HEAT's unresolved duration effect counts once: as SC1's named Θ exception and as SC3's Θ blocker, never as a second automatic penalty. Continuing medication injury and idiopathic ulcers constrain exclusivity; stress adds observational heterogeneity. 5 9 10 14 21 22

The scope test is explicit: a person cannot be called susceptible solely because they developed an ulcer or responded to eradication (pitfall 22). The converse also holds: the absence of a validated susceptibility rule is not a reason to withhold a level, because the claim no longer asserts one.

Other retained checks address selective generalization, pharmacological specificity, treatment effects mistaken for complete molecular proof, biomarker substitution, inflated study counts (pitfall 18: four evidence cards had marked rival or boundary results as supporting the nominal; corrected) and unlinked evidence. Marshall's gastritis cannot replace ulceration, and repeated regimens or reports cannot manufacture independent modalities.

Adversarial steelman

One rung higher — Ω. Within the recurrence scope, persistent infection is close to necessary — recurrence after successful eradication is rare — and close to sufficient in duodenal ulcer, where most persistently infected patients recurred; the effect replicates across dozens of trials and a hard endpoint; organism addition reproduces gastritis in people and ulcers in animals. 1 2 3 7 13 15 16 23

Why Ω is rejected. The Ω threshold asks for “necessity and sufficiency of the claimed microbial mechanism”. The graded claim is nonexclusive: ulcers occur without infection 9 12 21 and infection without ulcers 8 17. Even in the recurrence scope, gastric-ulcer recurrence in persistently infected patients was about one-half to three-quarters 2 3, and no second system reproduces recurrence.

One rung lower — Η (edition B's reading, recorded as dissent). The largest blinded hard-endpoint prevention trial, HEAT, was not significant overall (HR 0.69, 0.38–1.25) and its early benefit vanished for reasons its investigators call unclear; primary prevention otherwise rests on two NSAID-context trials and an uncontrolled community study; Cochrane rates the recurrence evidence very low quality; about a quarter of strictly tested US non-NSAID duodenal ulcers were uninfected. On this reading HEAT is a major, not a minimal, exception for a disease-class claim. 14 4 18 19 15 21

Why Η is rejected for SC1 and the overall letter. Θ requires minimal, not zero, unresolved exceptions. HEAT's later window is the only failed prediction within scope: 21 events, in established aspirin users, where a medication co-cause is independently mapped 20 22, with persistent clearance making reinfection an unlikely explanation 14. HEAT's overall non-significance is not a failed prediction of causation: the early effect is significant and the split was fixed before unblinding. Cochrane's quality rating concerns the precision of the effect, not its direction. The US association share limits near-universality, which the claim does not assert. For SC3, which asserts durable generalization, the same finding binds, and SC3 holds at Η.

Edition B's steelman (at Η), retained. One rung higher — Θ: the size and replication of recurrence effects, prospective prevention, two ulcer sites and complementary organism addition; a major cause need not be exclusive; HEAT's imprecise later result does not negate the early effect. Why Θ was not B's broad score: the original scope was the disease class; HEAT's duration is unresolved in the investigators' own account; stress-associated risk adds incomplete predictive coverage. One rung lower — Ζ: antimicrobial off-target effects, animal extrapolation and heterogeneous responses; rejected because replicated human studies measure ulcer recurrence or prevention. Pass-01 agrees with B's rejection of Ζ and departs from its rejection of Θ for the reasons above.

Upper-ladder check. Therapeutic perturbation with hard endpoints, a population natural experiment, a within-cohort gradient and observed withdrawal with reinfection-associated relapse are all present; Mendelian randomization has no plausible instrument for acquiring an infection. No deliberate human ulcer-induction trial is required, and observational components do not impose an association-only ceiling on the complete evidence set.

Role of mechanism

Primary driver, direction-of-causation step 3. Removing infection markedly reduces recurrence in baseline-defined ulcer cohorts; pretreatment prevents incident and complicated ulcers in NSAID-naive patients; infection precedes ulcer in stored sera; and defined-strain infection initiates animal ulceration. These support an upstream primary driver in the infection-mediated route. Rate-limiting language is an inference from outcome changes, not a measured kinetic constant or an outcome-based patient definition. Other ulcer pathways do not turn infection into a downstream passenger; model ranking did not select this role. 1 4 6 7 12 17 18

Momentum, stability and divergence

Momentum — Little new work. The last completed causal trial with an ulcer endpoint is HEAT (reports 2022 and 2025); current trials such as Medel-Jara et al. test regimen efficacy against an eradication endpoint, not ulcer causation. Edition B's “Mechanistic only” rested on that regimen trial, which is neither mechanistic nor causal work on ulcers. 6 14 11

Stability — Stable. No named pending result was identified. The foreseeable trigger, downward, would be a replicated long-term eradication trial in aspirin or NSAID starters showing HEAT's later loss in durably cleared participants; no such trial was identified.

Divergence — Aligned. Field framing implies Θ: the field's own Cochrane reviews present H. pylori as “the main etiologic factor for peptic ulcer disease” and as associated with the great majority of duodenal and most gastric ulcers, with NSAID and idiopathic exceptions acknowledged. Assessed Θ; Θ − Θ = 0. The field's association share is overstated for low-prevalence settings 21, a coverage point rather than a level point. Edition B derived an implied Η from trial conclusion sentences rather than from how the field presents the claim; that method is corrected. 15 16

Final adjudication and change record

Θ — High predictive accuracy, minimal unresolved exceptions; High confidence. H. pylori is a primary cause of peptic ulcer disease. The causal role is an upstream driver within an infection-mediated route, and does not require defining people by their later outcome. Its predictions hold across removal, prevention, temporal-order and population evidence; the one unresolved exception is HEAT's later period in long-term aspirin users. The pathway requirement is satisfied for duodenal ulcer and only partly for gastric ulcer; SC1's Θ does not rest on the pathway limb alone but on reproducibility, narrow attribution and predictions that could have failed and did not.

Previous level: Η (edition B). New level: Θ. Edition A reached Θ with an outcome-defined qualifier and did not defend it against HEAT; edition B deleted the qualifier and lowered to Η; pass-01 reaches Θ on the unqualified claim with HEAT named. What moved it: two of B's three SC1 blockers do not bind a nonexclusive claim, and evidence B had not included supplies the replication and tested predictions Θ requires 15 16 17 18 19. Counter-evidence was added and weighed 20 21 22. Primary-subclaim aggregation is unchanged; weakest-link would return Η.

SC1 and SC2 are Θ; SC3 is Η. High confidence describes the density and reproducibility of causal evidence, not a formal GRADE rating of every effect estimate. The one-rung boundary remains a review judgment whose premises and alternatives, including edition B's dissent, are exposed here.

Proposed decisive experiment

The next useful work would test prospective predictions about benefit duration and baseline effect modifiers. It should separate recurrence from primary prevention, record medication exposure and durable clearance, measure proposed host/strain modifiers before outcomes, and follow adjudicated ulcers and bleeding over prespecified early and longer intervals. Independent validation is needed before any biomarker defines a susceptible subgroup.

The most informative single design (pass-01): randomize infected people at the start of long-term low-dose aspirin or NSAID therapy to eradication or placebo, with repeat testing for clearance and reinfection in both arms and adjudicated ulcer bleeding over at least five years, analysed in prespecified intervals. The HEAT investigators propose starters as the next population. Sustained benefit in durably cleared participants would let SC3 reach Θ; a replicated later loss would return SC1 to Η. 14

The critical failed prediction is a reproducible, precise lack of the prespecified clinically meaningful ulcer reduction after durable removal in a baseline-defined population, with cointerventions and follow-up appropriately controlled. That would constrain that population/time claim; repeated failures across principal settings would challenge the broad account. Lack of statistical significance alone is not such a result. Participants must not be relabeled out of scope afterward.

This is a conceptual design, not a claim that a named trial is underway. Established treatment need not be withheld to obtain independent prospective validation of durability, reinfection and medication-related boundaries.

Knowledge gaps

• Individual progression: the host and bacterial features determining why some infected people develop gastric or duodenal ulcers while others do not.
• Current attributable burden: the fraction of disease due to infection in a specified place, age group and medication-exposure setting. This review supplies no universal percentage; strictly tested US data show the share is lower than older figures suggested 21.
• Durability during aspirin exposure: the mechanism of the changing HEAT treatment effect and the appropriate population-specific prevention strategy. It is the single named exception to SC1's Θ and the reason SC3 holds at Η. Its later confidence interval does not prove harm, and persistent clearance makes reinfection an unlikely explanation.

Reviewer notes and provenance limits

Red-team record (pass-01, 27 September 2026). This edition is the red-team synthesis of edition B by Claude (Anthropic), an independent model from a different family than the graders of editions A and B (OpenAI Codex). Every ledger row was re-resolved in PubMed and every claim re-checked: no identifier was wrong and no claim was contradicted. The pass changed the level from Η to Θ (SC1 Η → Θ; SC2 and SC3 held), added nine sources (six Tier A, three Tier B carrying no figure) and thirteen claims, corrected four evidence-card directions, the Momentum and Divergence artifacts and the tournament ordering, migrated the card to schema 5.4 with a lay summary, and recorded edition B's case for Η as dissent. It also found that the archive's prior process record asserted a corrigendum on a "Tindle" row that does not exist in this ledger. The full trail — findings, identifier and claim audits, and an edit-by-edit change log — is in PROCESS.md and redteam/.

Validator warnings, disclosed. Validator 2.4 returns 0 failures and two warnings, both true and neither fixable by editing the card: the calibration entry was not graded blind (reason recorded), and the expected band was written by an author who had seen the review.

Edition B's notes, retained. The archive was used as the user-requested evaluation schema. Its embedded workflow text was not treated as separate authorization for corpus edits or unrelated actions. Life Sciences Literature's Entrez and PMC skills were used for the original retrieval; Europe PMC indexed primary abstracts supplemented incomplete public-page/tool responses. Marshall, stress and HEAT records were rechecked for that revision. B read the HEAT scientific summary at NCBI Bookshelf; pass-01 could not resolve that page (bot-check) and read targeted passages of the full report from the NIHR Journals Library instead. No claim is made to have audited the entire 62-page report or all older full texts. Scite citation-context retrieval was unavailable to both passes because of quota; absent counts are not zeros.

Status flags mean no correction, retraction or expression of concern was found by the checks run: pass-01 used PubMed's retraction publication-type filter on all 23 records, erratum searches by first author, and Europe PMC's comment/correction list for two records before that service rate-limited. This is a targeted review, not an exhaustive systematic review. Factual-premise verification does not independently validate causal synthesis, letter grades or model rankings.

Calibration status: this is an unblinded reference case, not a completed blinded calibration result. Edition B's author wrote the sealed target after seeing edition A's score and the critique. The red-team reviewer was told edition B's letters in the brief, locked this pass's level before opening the sealed band, and then confirmed that the band's SHA-256 matches the commitment. The band is Η–Θ; this edition's Θ falls inside it, which is not evidence of calibration. The prepared grader packet ships framework v5.0/schema 5.2, and a v5.4 packet would carry framework text that names this case; how to run a blind grading is an open decision for the owner. No new grader has run, and no calibration set is claimed.

Independence basis

Four listed lines, two direct ulcer lines. Marshall and Morris exclude a gerbil-only explanation for the human gastritis observation and an antimicrobial protective explanation for its induction; they do not remove species uncertainty for ulceration itself. The stored-sera line is exposure accumulation set against line 1's withdrawal, with a different confounding structure.

Different eradication regimens remain one therapeutic modality. The 2025 HEAT report remains part of the 2022 trial. Animal induction complements human removal because their major failure modes differ; triangulation is a reasoned inference, not proof against every conceivable shared bias.

ADVERSARIAL REVIEW AND APPROVAL

Authoring model: Sol 6.1 (the document records editions A and B as graded by OpenAI Codex and this edition as a Claude red-team synthesis of 2026-09-27; recorded, not edited). Adversarial reviewer: Claude Opus 5.5 (Anthropic). Date: 2026-10-14. Round: 2 (new-review inventory, group 1).

Scope of checks. All 23 REFERENCES PMIDs verified against PubMed get_article_metadata for title, first author, journal and year: 23/23 resolve correctly, 0 failures. No retraction or expression of concern appears in the article-type fields of any record, which independently confirms the review's own retraction screen. Twelve load-bearing claims were spot-checked against PubMed abstracts, concentrating on the figures that carry the Η-to-Θ upgrade: Hentschel's randomized-group recurrence (4/50 versus 42/49, and the 2% versus 85% eradication-status comparison the review correctly labels as a non-randomized split); Graham's duodenal and gastric recurrence probabilities with no maintenance therapy; HEAT's first-period hazard ratio 0.35 (0.14-0.89) and later 1.31 (0.55-3.11), the 6-versus-17 event counts, the proportional-hazards departure at p=0.0068, the pre-fixed 2.5-year split, the overall 0.69 (0.38-1.25) in the 2025 report, and the eligibility age of 60 or older (which the review correctly flags as inconsistent with the 2025 report's own title saying "over 65 years"); Chan 2002's intention-to-treat 6-month ulcer probabilities 12.1% versus 34.4% and complicated-ulcer 4.2% versus 27.1%; Chan 2001's aspirin stratum 1.9% versus 0.9% with absolute difference 1.0% (-1.9 to 3.9); Lai's 14.8% versus 1.6% with 4 of 10 reinfected; Nomura's odds ratios 3.2 (1.6-6.5) and 4.0 (1.1-14.2) with 150 and 65 cases and the 10-year-plus persistence; Lee's 67.4% (52.2-77.8) peptic-ulcer reduction with 1% per person-year re-infection; Ciociola's 73% of 2394; the Ford Cochrane review's recurrence-versus-no-treatment risk ratios and its null eradication-versus-maintenance-drug comparison, which this review reports rather than omits; and the Gisbert Cochrane review's rebleeding result and its verbatim sentence "Helicobacter pylori infection is the main etiologic factor for peptic ulcer disease", which the Divergence artifact quotes and attributes correctly. Every spot-checked figure matched the source, with the two exceptions recorded as findings below. Scoring logic was re-derived in full: the six universal caps against each of the three subclaims, the rubric-04 eradication provision versus cap 5 (cap precedence), the v5.3 scope-not-defined-by-outcome rule (the "in susceptible hosts" qualifier is deleted, not merely narrowed), the pre-registered primary-subclaim aggregation and both alternatives, the four-line independence basis, the Θ-versus-Η and Θ-versus-Ω boundaries, the role derivation, and the M/S/D artifacts. The 5.4 structural validator was re-run.

Findings

No level changed. The headline Θ, High confidence, SC1 Θ / SC2 Θ / SC3 Η, the primary-driver role, the seven-perturbation tally and the M/S/D values are all unchanged. C-2 does not move a letter: SC1 is a disease-class claim that the duodenal evidence alone supports, and SC3 — the generalization subclaim — already holds at Η.

Rejected findings (checked, no change)

Lead policy applied — confidence bands and momentum (2026-10-14)

Confidence, two-step test. Step 1: access depth is struck as a determinant. How much of a paper this reviewer could read is neither evidence density nor reproducibility and never justifies a band; the rationale was rewritten accordingly (no rewrite was needed here — see rejected finding R-11). Step 2: the band was re-derived from what remains. This is the one case in this group whose rationale never leaned on access depth, so step 1 changed nothing and step 2 is a confirmation rather than a re-derivation. High rests on replication across independent investigators, countries, ulcer sites, endpoints and designs, with pooled syntheses confirming direction — which is the High text. The review separately and correctly distinguishes HSEF confidence from GRADE certainty, noting that Cochrane rates the recurrence evidence very low quality on trial conduct; that is a statement about effect precision, not about replication breadth. High stands.

Momentum, stricter rule. "Decisive work underway" requires work actually being run that could cross the next threshold; a completed publication or trial never satisfies it, however real the study. Re-checked and the value holds. Momentum is "Little new work", the conservative value, and it is argued from completed work rather than crediting it as momentum: HEAT (2022 and 2025 reports) is named as the last completed causal trial with an ulcer endpoint, and the one current trial in the ledger is identified as a regimen trial with an eradication endpoint rather than a causal ulcer trial. Nothing being run is claimed to be threshold-crossing.

Residual limitations

• Access depth is abstract and indexed metadata for 22 of 23 records, plus targeted passages of the HEAT full report. Three figures carrying real weight could not be reached at abstract depth and rest on that targeted reading: HEAT's second-period event counts (12 versus 9), the repeat-breath-test result at a median 3.95 years (146 of 161 negative in the eradication arm versus 41 of 161 in the control arm), and Sung's one-year recurrence subset. All three are consistent with the published abstracts and the review states its access limits plainly; none is flagged as an error.
• The two validator warnings are real and are disclosed in the review: the calibration entry was not graded blind, and the sealed band was written by an author who had seen the review. The Θ inside an Η-to-Θ band is therefore not evidence of calibration, which the review also says.
• Rubric 04 is an acknowledged imperfect fit, and the Β/Γ/Δ boundary rests on the review's explicit reading of an internally uneven rubric (eradication provision versus the transfer/consortium wording). The reading is stated, quoted and consistent with the framework's cap-precedence rule, but it is an interpretation.

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

REFERENCES

1. Hentschel E, Brandstätter G, Dragosics B, et al. Effect of ranitidine and amoxicillin plus metronidazole on the eradication of Helicobacter pylori and the recurrence of duodenal ulcer. N Engl J Med. 1993;328(5):308-312. PubMed PMID 8419816 · DOI.

2. Graham DY, Lew GM, Klein PD, et al. Effect of treatment of Helicobacter pylori infection on the long-term recurrence of gastric or duodenal ulcer. A randomized, controlled study. Ann Intern Med. 1992;116(9):705-708. PubMed PMID 1558340 · DOI.

3. Sung JJ, Chung SC, Ling TK, et al. Antibacterial treatment of gastric ulcers associated with Helicobacter pylori. N Engl J Med. 1995;332(3):139-142. PubMed PMID 7800005 · DOI.

4. Chan FK, Sung JJ, Chung SC, et al. Randomised trial of eradication of Helicobacter pylori before non-steroidal anti-inflammatory drug therapy to prevent peptic ulcers. Lancet. 1997;350(9083):975-979. PubMed PMID 9329511 · DOI.

5. Chan FK, Chung SC, Suen BY, et al. Preventing recurrent upper gastrointestinal bleeding in patients with Helicobacter pylori infection who are taking low-dose aspirin or naproxen. N Engl J Med. 2001;344(13):967-973. PubMed PMID 11274623 · DOI.

6. Hawkey C, Avery A, Coupland CAC, et al. Helicobacter pylori eradication for primary prevention of peptic ulcer bleeding in older patients prescribed aspirin in primary care (HEAT): a randomised, double-blind, placebo-controlled trial. Lancet. 2022;400(10363):1597-1606. PubMed PMID 36335970 · DOI.

7. Ohkusa T, Okayasu I, Miwa H, et al. Helicobacter pylori infection induces duodenitis and superficial duodenal ulcer in Mongolian gerbils. Gut. 2003;52(6):797-803. PubMed PMID 12740333 · DOI.

8. el-Omar E, Penman I, Dorrian CA, et al. Eradicating Helicobacter pylori infection lowers gastrin mediated acid secretion by two thirds in patients with duodenal ulcer. Gut. 1993;34(8):1060-1065. PubMed PMID 8174954 · DOI.

9. Wong GL, Wong VW, Chan Y, et al. High incidence of mortality and recurrent bleeding in patients with Helicobacter pylori-negative idiopathic bleeding ulcers. Gastroenterology. 2009;137(2):525-531. PubMed PMID 19445937 · DOI.

10. Levenstein S, Rosenstock S, Jacobsen RK, et al. Psychological stress increases risk for peptic ulcer, regardless of Helicobacter pylori infection or use of nonsteroidal anti-inflammatory drugs. Clin Gastroenterol Hepatol. 2015;13(3):498-506.e1. PubMed PMID 25111233 · DOI.

11. Medel-Jara PA, Latorre G, Fuentes-Lopez E, et al. Comparison between optimized bismuth quadruple therapy and standard clarithromycin-based triple therapy for first-line Helicobacter pylori eradication: a double-blind randomized controlled trial. Lancet Reg Health Am. 2026;53:101312. PubMed PMID 41362749 · DOI.

12. Laine L, Harper S, Simon T, et al. A randomized trial comparing the effect of rofecoxib, a cyclooxygenase 2-specific inhibitor, with that of ibuprofen on the gastroduodenal mucosa of patients with osteoarthritis. Rofecoxib Osteoarthritis Endoscopy Study Group. Gastroenterology. 1999;117(4):776-783. PubMed PMID 10500058 · DOI.

13. Marshall BJ, Armstrong JA, McGechie DB, et al. Attempt to fulfil Koch's postulates for pyloric Campylobacter. Med J Aust. 1985;142(8):436-439. PubMed PMID 3982345 · DOI.

14. Hawkey CJ, Avery AJ, Coupland CA, et al. Eradication of Helicobacter pylori for prevention of aspirin-associated peptic ulcer bleeding in adults over 65 years: the HEAT RCT. Health Technol Assess. 2025;29(42):1-62. PubMed PMID 40844182 · DOI. Same randomized study as [6]; eligibility was age 60 or older despite the title.

15. Ford AC, Gurusamy KS, Delaney B, Forman D, Moayyedi P. Eradication therapy for peptic ulcer disease in Helicobacter pylori-positive people. Cochrane Database Syst Rev. 2016;4(4):CD003840. PubMed PMID 27092708 · DOI. Tier B.

16. Gisbert JP, Khorrami S, Carballo F, Calvet X, Gené E, Dominguez-Muñoz JE. H. pylori eradication therapy vs. antisecretory non-eradication therapy (with or without long-term maintenance antisecretory therapy) for the prevention of recurrent bleeding from peptic ulcer. Cochrane Database Syst Rev. 2004;(2):CD004062. PubMed PMID 15106235 · DOI. Tier B.

17. Nomura A, Stemmermann GN, Chyou PH, Perez-Perez GI, Blaser MJ. Helicobacter pylori infection and the risk for duodenal and gastric ulceration. Ann Intern Med. 1994;120(12):977-981. PubMed PMID 7741826 · DOI.

18. Chan FK, To KF, Wu JC, et al. Eradication of Helicobacter pylori and risk of peptic ulcers in patients starting long-term treatment with non-steroidal anti-inflammatory drugs: a randomised trial. Lancet. 2002;359(9300):9-13. PubMed PMID 11809180 · DOI.

19. Lee YC, Chen TH, Chiu HM, et al. The benefit of mass eradication of Helicobacter pylori infection: a community-based study of gastric cancer prevention. Gut. 2013;62(5):676-682. PubMed PMID 22698649 · DOI.

20. Vergara M, Catalán M, Gisbert JP, Calvet X. Meta-analysis: role of Helicobacter pylori eradication in the prevention of peptic ulcer in NSAID users. Aliment Pharmacol Ther. 2005;21(12):1411-1418. PubMed PMID 15948807 · DOI. Tier B.

21. Ciociola AA, McSorley DJ, Turner K, Sykes D, Palmer JB. Helicobacter pylori infection rates in duodenal ulcer patients in the United States may be lower than previously estimated. Am J Gastroenterol. 1999;94(7):1834-1840. PubMed PMID 10406244 · DOI.

22. Lai KC, Lam SK, Chu KM, et al. Lansoprazole for the prevention of recurrences of ulcer complications from long-term low-dose aspirin use. N Engl J Med. 2002;346(26):2033-2038. PubMed PMID 12087138 · DOI.

23. Morris A, Nicholson G. Ingestion of Campylobacter pyloridis causes gastritis and raised fasting gastric pH. Am J Gastroenterol. 1987;82(3):192-199. PubMed PMID 3826027 (no DOI indexed).

Citation ledger

Tier A here means an original research report, not a uniform risk-of-bias rating. Report [14] is not independent of [6]. Access depth and status checks are recorded per source in the JSON; every row was re-resolved by pass-01 on 2026-09-27.

Claim register

“Checked” refers to factual premises in original records. Causal synthesis and framework assignments remain reviewer judgments, not independently verified conclusions. Every claim was re-checked by pass-01; see redteam/claim-audit.md.

Companion JSON: schema 5.4 (migrated from edition B's 5.2 with declared local extensions). Validator 2.4: 0 failures; 2 disclosed calibration warnings. Structural validation does not independently validate scientific judgments.

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
• h-pylori-peptic-ulcer.review.md
• h-pylori-peptic-ulcer.scorecard.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

Calibration entry. Designated a calibration entry but not graded blind. An unblinded entry shows the framework can reach a band, not that it does so independently.

Sources: 23 · load-bearing claims: 34 · verified: 34 · unverified: 0 · contradicted: 0 · schema 5.4