A generative AI-discovered TNIK inhibitor for idiopathic pulmonary fibrosis: a randomized phase 2a trial.
Xu Z, Ren F, Wang P, Cao J, Tan C, Ma D, Zhao L, Dai J, Ding Y, Fang H, Li H, Liu H, Luo F, Meng Y, Pan P, Xiang P, Xiao Z, Rao S, Satler C, Liu S, Lv Y, Zhao H, Chen S, Cui H, Korzinkin M, Gennert D, Zhavoronkov A
- DOI
- 10.1038/s41591-025-03743-2
- Record issued
- 2026-08-15
- Engine
- 7.39.0
- Exported
- 2026-09-21
Prepared by Alpha1. This document is confidential: it is intended for the recipient it was shared with and must not be redistributed. The live record at alpha1science.com/verify/2172d464-0c4f-4a0e-92dd-931f9ab4333c is authoritative.
How this rating was calculated
- IntegrityIntegrity concern ×3−1.5★
- StatisticsStatistic did not reproduce−0.5★
- ClaimsEfficacy rests on an unvalidated surrogate endpoint−0.5★
- ClaimsTreatment effect not shown to be clinically meaningful−0.5★
- StatisticsPrinted percentage does not match its own count (capped) ×6−0.25★
- 01Printed percentage does not match its own count
20.4% does not match the reported count 3/17
“3 in 60 mg QD (20.4%)”
HypokalemiaFind in source - 02Efficacy rests on an unvalidated surrogate endpoint
The primary efficacy claim is based on change in forced vital capacity (FVC), a surrogate endpoint for clinical benefit in IPF. The paper does not provide evidence of target engagement at the tested dose (e.g., PK/PD demonstrating TNIK inhibition) nor does it cite validated evidence linking FVC changes to improved clinical outcomes such as survival or quality of life. The FVC change is presented as a secondary endpoint, and the paper acknowledges the need for larger trials to confirm efficacy.
“We observed increased forced vital capacity at the highest dosage with a mean change of +98.4 ml (95% confidence interval 10.9 to 185.9) for patients in the 60 mg rentosertib QD group, compared with −20.3 ml (95% confidence interval −116.1 to 75.6) for the…”
- 03Treatment effect not shown to be clinically meaningful
The reported effect size is a mean FVC increase of +98.4 ml in the 60 mg QD group, which is small relative to the normal FVC range (baseline mean ~2.7 L) and may not exceed the minimal clinically important difference (MCID) for FVC in IPF (2-6% change). The paper mentions that the percentage change met the MCID, but the absolute change is modest and the confidence interval is wide. No anchor to clinical meaningfulness is provided beyond this.
“Patients receiving 60 mg rentosertib QD experienced a mean increase in FVC percentage change of 2.82% by our modeling, meeting the minimal clinically important difference (MCID) reported for FVC in IPF of 2–6% (ref. ).”
- 04Printed percentage does not match its own count
17% does not match the reported count 3/18
“3/18 (17%) in the 60 mg QD group”
Liver injury or dysfunctionFind in source - 05Printed percentage does not match its own count
77% does not match the reported count 55/71
“55 (77%) completed the 12-week placebo-controlled period”
Completed 12-week periodFind in source - 06Printed percentage does not match its own count
88% does not match the reported count 15/17
“15 (88%) in the placebo group”
Completed 12-week periodFind in source
4 further findings of this severity or below — every one is in the sections below, filed under its error type.
This Kaimen Rigor review uses Kaimen Rigor reviewers trained on a curated corpus of high-fidelity and retracted papers, with expert supervision and curation. It can still make mistakes; verify each finding against the source before relying on it.
This is a well-conducted phase 2a randomized controlled trial with strong scientific premise, rigorous design, and transparent reporting. The main weaknesses are the lack of a formal efficacy power calculation, some threshold-based p-value reporting, and minor internal inconsistencies in the safety results.
Both reviewers classified the study as interventional, and I adopted that classification. The evaluation covered the full text, with all eight dimensions scored. Non-applicable sub-criteria (e.g., animal housing, cell line authentication) were excluded. The statistics verification component could only recompute a subset of tests; 7 of 8 were inconsistent, but the coverage note limits interpretation, and no specific errors were identified.
Numerical inconsistencies
3 findings · worst highValues that contradict each other or are impossible for the stated sample: recomputed p-values and test statistics, GRIM/GRIMMER checks on summary numbers, percentages against their own counts, totals against their parts, and estimates against their own confidence intervals.
- Summary statistic impossible for the stated N (GRIM/GRIMMER)Recomputed
- Printed percentage does not match its own countRecomputed
- Internal contradictions in the reported numbersAssessed
Recomputed 1 test: 1 consistent, 0 inconsistent; 1 via agent-written checks. 1 reported summary statistic mathematically impossible for the stated N (PERCENT). 6 printed percentages that do not match their own count.
- PERCENT20.4% does not match the reported count 3/17
“3 in 60 mg QD (20.4%)”
HypokalemiaFind in source - PERCENT17% does not match the reported count 3/18
“3/18 (17%) in the 60 mg QD group”
Liver injury or dysfunctionFind in source - PERCENT77% does not match the reported count 55/71
“55 (77%) completed the 12-week placebo-controlled period”
Completed 12-week periodFind in source - PERCENT88% does not match the reported count 15/17
“15 (88%) in the placebo group”
Completed 12-week periodFind in source - PERCENT89% does not match the reported count 16/18
“16 (89%) in the 30 mg rentosertib QD group”
Completed 12-week periodFind in source - PERCENT67% does not match the reported count 12/18
“12 (67%) in the 30 mg rentosertib BID group”
Completed 12-week periodFind in source - PERCENT67% does not match the reported count 12/18
“12 (67%) in the 60 mg rentosertib QD group”
Completed 12-week periodFind in source
- CONSISTENTreported p = .050 · recomputed p = .050Reviewers 1, 2Check p-value for LCQ comparison between 60 mg QD and placebo.
“our modeling indicated a significant increase of least-squares mean change of the LCQ scores of patients receiving 60 mg rentosertib compared with those receiving placebo (two-sided P = 0.0495)”
Taken as given: The p-value is from a two-sided test.; The test statistic is approximately normal (z = 1.96).Method: Approximated p-value from z-score of 1.96 using normal distribution.How we recomputed it: pZ(1.96)
- lowinternal contradictionIn the primary safety endpoint section, the percentage for 3 out of 18 in the 60 mg QD group for hypokalemia is reported as 20.4%, but 3/18 is 16.7%.
“3 in 60 mg QD (20.4%)”
ResultsFind in source - lowinternal contradictionIn the same section, ALT increase is listed twice for the 60 mg QD group with different counts (6 and 3), which is inconsistent.
“6 in 60 mg QD (33.3%) and 3 in 60 mg QD (16.7%)”
ResultsFind in source
Overstated conclusions
3 findings · worst highConclusions that reach past what the paper's own results support — including a significance claim that no longer holds when the statistic is recomputed, and efficacy resting on an unvalidated surrogate endpoint.
- Efficacy rests on an unvalidated surrogate endpointAssessed
- Treatment effect not shown to be clinically meaningfulAssessed
- Conclusions only partially backed by the presented evidenceAssessed
5 major claims checked against the paper's own evidence: 3 only partially supported (evidence backs part of the claim; gaps or caveats remain); the rest adequately supported.
- partialReviewers 1, 2TNIK inhibition modulates IPF pathophysiological pathways.The proteomic data show downregulation of fibrosis-associated proteins, but this is exploratory and not directly linked to clinical outcomes.Evidence: Proteomic analysis shows downregulation of COL1A1, FAP, FN1, MMP10 and enrichment of ECM organization pathway.
“Downregulation of this IPF-associated protein profile supports the hypothesis that inhibition of TNIK modulates IPF pathophysiological pathways”
DiscussionFind in source - partialReviewers 1, 2Rentosertib at 60 mg QD improves FVC compared to placebo.The observed FVC improvement (+98.4 ml) is suggestive but not statistically significant (CI includes zero), and the study was not powered for efficacy.Evidence: FVC change results: +98.4 ml (95% CI 10.9 to 185.9) vs -20.3 ml (95% CI -116.1 to 75.6).
“We observed increased forced vital capacity at the highest dosage with a mean change of +98.4 ml (95% confidence interval 10.9 to 185.9) for patients in the 60 mg rentosertib QD group, compared with −20.3 ml (95% confidence interval −116.1 to 75.6) for the placebo group.”
AbstractFind in source - partialReviewer 2The AI-discovered drug demonstrates potential efficacy in IPF.The efficacy signal is preliminary and based on secondary endpoints; the study was primarily a safety trial.Evidence: FVC improvement and biomarker changes.
“Treatment with 60 mg rentosertib QD over 12 weeks was associated with a trend toward an increase in FVC in patients with IPF.”
DiscussionFind in source - supportedReviewers 1, 2Rentosertib is safe and well tolerated in patients with IPF.The primary safety endpoint (TEAE rates) was similar across groups, and treatment-related SAEs were low.Evidence: Primary safety endpoint results: TEAE rates 72.2%, 83.3%, 83.3%, 70.6% across groups; treatment-related SAEs 5.6%, 11.1%, 11.1%, 0%.
“These results suggest that targeting TNIK with rentosertib is safe and well tolerated”
AbstractFind in source - supportedReviewer 1The AI-discovered drug rentosertib is the first AI-discovered drug to reach phase 2a trials.The paper states this is the first phase 2a trial of an AI-discovered drug, and the context supports this claim.Evidence: Introduction and abstract state this is the first phase 2a trial.
“Here we present the results of the first phase 2a multicenter, double-blind, randomized, placebo-controlled trial testing the safety and efficacy of rentosertib”
AbstractFind in source
Premise concern: surrogate not validated for clinical benefit; effect size not shown to be clinically meaningful.
- INADEQUATESurrogate endpointThe primary efficacy claim is based on change in forced vital capacity (FVC), a surrogate endpoint for clinical benefit in IPF. The paper does not provide evidence of target engagement at the tested dose (e.g., PK/PD demonstrating TNIK inhibition) nor does it cite validated evidence linking FVC changes to improved clinical outcomes such as survival or quality of life. The FVC change is presented as a secondary endpoint, and the paper acknowledges the need for larger trials to confirm efficacy.
“We observed increased forced vital capacity at the highest dosage with a mean change of +98.4 ml (95% confidence interval 10.9 to 185.9) for patients in the 60 mg rentosertib QD group, compared with −20.3 ml (95% confidence interval −116.1 to 75.6) for the placebo group.”
- INADEQUATEEffect sizeThe reported effect size is a mean FVC increase of +98.4 ml in the 60 mg QD group, which is small relative to the normal FVC range (baseline mean ~2.7 L) and may not exceed the minimal clinically important difference (MCID) for FVC in IPF (2-6% change). The paper mentions that the percentage change met the MCID, but the absolute change is modest and the confidence interval is wide. No anchor to clinical meaningfulness is provided beyond this.
“Patients receiving 60 mg rentosertib QD experienced a mean increase in FVC percentage change of 2.82% by our modeling, meeting the minimal clinically important difference (MCID) reported for FVC in IPF of 2–6% (ref. ).”
Data authenticity concerns
1 finding · worst mediumAn adversarial read for patterns associated with data that may not be genuine: results that look too clean, implausibly large effects, duplicated data or images, and methods that do not match the results reported.
- Other integrity concernAssessed
3 integrity concerns flagged (0 high).
- mediumotherTrial NCT05938920 was first submitted to ClinicalTrials.gov on 2023-06-28, after the registered study start date of 2023-06-19. Retrospective registration means the protocol and outcomes were not on the public record before the study ran, which is what prospective registration exists to establish.
NCT05938920
reviewer’s wording
Reporting gaps
None foundRequired detail the manuscript never states — study design, biological variables, ethics approval and consent, key resources, statistical reporting, data and code availability, and overall transparency.
Checked — nothing surfaced.
The introduction extensively cites prior research on AI in drug discovery, TNIK's role in IPF, and the limitations of current therapies. It explicitly acknowledges the high failure rate of AI-discovered drugs in phase 2 trials and the lack of phase 3 successes, framing the study as addressing this gap. The rationale for targeting TNIK and testing rentosertib is logically derived from prior evidence, and the study aims to address the limitation of prior work by conducting a randomized controlled trial.
“Here, we present the results of a phase 2a multicenter, double-blind, randomized, placebo-controlled trial to assess the safety, tolerability, PK and impact on FVC of rentosertib”
“Here, we present the results of a phase 2a multicenter, double-blind, randomized, placebo-controlled trial to assess the safety, tolerability, PK and impact on FVC of rentosertib at a range of doses for up to 12 weeks in patients with IPF.”
Randomization was performed via interactive response technology in a 1:1:1:1 ratio, and blinding was comprehensive, covering subjects, investigators, and study staff. Inclusion/exclusion criteria are detailed. The power analysis is based on safety feasibility (90% probability of observing at least one AE with 15 subjects per arm), which is appropriate for a phase 2a safety trial, though not a formal efficacy power calculation. Outlier handling is addressed through the pre-specified analysis population and multiple imputation for missing data.
“Adults with IPF were randomly assigned in a 1:1:1:1 ratio via interactive response technology”
“Subjects, investigators, site study staff, reviewers and everyone involved in study conduct or analysis were blinded with regard to the randomized treatment assignments until after data freeze.”
“Given an approximate sample size of 15 subjects per treatment arm, there exists a 90% probability of observing at least one AE if the true population rate is approximately 15%, which was sufficient to assess the feasibility of safety parameters, although a sample size calculation based on statistical power considerations was not performed.”
“Adults with IPF were randomly assigned in a 1:1:1:1 ratio via interactive response technology to receive oral rentosertib (at a dose of 30 mg (QD), 30 mg (BID) or 60 mg (QD)) or placebo (QD) for 12 weeks”
“Subjects, investigators, site study staff, reviewers and everyone involved in study conduct or analysis were blinded with regard to the randomized treatment assignments until after data freeze.”
“Given an approximate sample size of 15 subjects per treatment arm, there exists a 90% probability of observing at least one AE if the true population rate is approximately 15%, which is sufficient to assess the feasibility of safety parameters.”
Sex, age, weight, BMI, and race are reported in Table 1. The study enrolled both sexes, so sex_justified is not applicable. Age, weight, and health status are reported. Demographics are reported, including race (all Asian) and smoking history. Species/strain and housing conditions are not applicable for a human trial.
“Male sex – no. (%) | 18 (100.0) | 15 (83.3) | 16 (88.9) | 15 (88.2) | 64 (90.1)”
“Male sex – no. (%) | 18 (100.0) | 15 (83.3) | 16 (88.9) | 15 (88.2) | 64 (90.1)”
The paper states that the trial was conducted following the Declaration of Helsinki and ICH-GCP guidelines, and that the institutional review board or ethics committee at participating centers approved protocols. Written informed consent was obtained from all patients. This satisfies the requirements for human research.
“The institutional review board or ethics committee at participating centers approved protocols and adhered to local laws before initiation of the clinical trial.”
“All patients in this study provided written informed consent.”
“The trial was conducted following the principles outlined in the Declaration of Helsinki and the International Council for Harmonization guidelines for Good Clinical Practice.”
“The institutional review board or ethics committee at participating centers approved protocols and adhered to local laws before initiation of the clinical trial. All patients in this study provided written informed consent.”
“The trial was conducted following the principles outlined in the Declaration of Helsinki and the International Council for Harmonization guidelines for Good Clinical Practice.”
Rentosertib is identified as the investigational drug, with doses and regimen specified. The sponsor (Insilico Medicine) is named. Software tools for analysis (lme4, clusterProfiler, lm) are identified. Antibodies, cell lines, mycoplasma, and organisms are not applicable as this is a clinical trial without wet-lab assays.
“receive oral rentosertib (at a dose of 30 mg (QD), 30 mg (BID) or 60 mg (QD)) or placebo (QD) for 12 weeks”
“The lme4 package was used for the generalized linear mixed model analysis.”
“The study treatment rentosertib was provided by Insilico Medicine or a designated contract research organization”
“The lme4 package was used for the generalized linear mixed model analysis.”
Statistical tests are named (ANCOVA, mixed model for repeated measures, Spearman correlation, generalized linear mixed model). Assumptions are handled by design (e.g., multiple imputation for missing data). Effect sizes with 95% CIs are reported for FVC changes. Exact p-values are given for some analyses (e.g., LCQ P=0.0495). Software is identified. Data presentation includes per-group n and confidence intervals. Mathematical plausibility checks were not performed due to continuous data and small N, but no obvious errors were noted.
“An analysis of covariance (ANCOVA) model was performed for endpoints including FVC.”
“patients receiving 60 mg rentosertib QD showed improved FVC, with and +98.4 ml (95% CI 10.9 to 185.9)”
“our modeling indicated a significant increase of least-squares mean change of the LCQ scores of patients receiving 60 mg rentosertib compared with those receiving placebo (two-sided P = 0.0495)”
“patients receiving 60 mg rentosertib QD showed improved FVC, with and +98.4 ml (95% CI 10.9 to 185.9)”
“20, 82 and 192 proteins significantly changing with treatment at weeks 2, 4 and 12, respectively (Benjamini–Hochberg-adjusted P value <0.1)”
The data availability statement provides a concrete route: deidentified Olink proteomics and FVC data deposited at OMIX with accession OMIX008341. Study protocol and SAP available upon request. Custom code is available on GitHub. This meets the criteria for adequate data and code sharing.
“Complete deidentified Olink proteomics data, and FVC data, have been deposited at the OMIX database under accession codes at OMIX008341 ( https://ngdc.cncb.ac.cn/omix/release/OMIX008341 ).”
“Custom code used to analyze Olink proteomics data is available via GitHub at https://github.com/HUICUI1992/Code-for-NM .”
“Complete deidentified Olink proteomics data, and FVC data, have been deposited at the OMIX database under accession codes at OMIX008341 ( https://ngdc.cncb.ac.cn/omix/release/OMIX008341 ).”
“Custom code used to analyze Olink proteomics data is available via GitHub at https://github.com/HUICUI1992/Code-for-NM .”
The trial is registered at ClinicalTrials.gov (NCT05938920). Methods are detailed enough for replication. Limitations are explicitly discussed (small cohort, homogeneity, short follow-up). Conclusions are proportional, noting the need for larger trials. Funding and competing interests are declared. Reporting guideline adherence is implied by Nature Portfolio reporting summary.
“ClinicalTrials.gov registration number: NCT05938920”
“The limitations of this study include the small cohort size of each arm, the geographical and demographic homogeneity of the participants (all were residents of China of similar race) and a short period of follow-up”
“F.R., S.R., C.S., S.L., Y.L., H.Z., S.C., H.C., M.K., D.G. and A.Z. are employees of Insilico Medicine. Insilico Medicine was the study sponsor.”
“ClinicalTrials.gov registration number: NCT05938920”
“The limitations of this study include the small cohort size of each arm, the geographical and demographic homogeneity of the participants (all were residents of China of similar race) and a short period of follow-up”
Registered (2 IDs: ClinicalTrials.gov). No reporting guideline cited.
Broken references and links
None foundReferences checked against Crossref, OpenAlex and Retraction Watch for retractions and resolvability, plus declared data and code links probed for whether they resolve to content matching the paper.
Checked — nothing surfaced.
Checked 70 references by DOI: 68 verified — 2 no DOI (shown, not verified).
- NO DOIMachine learning-aided generative molecular designNo DOI in the reference — shown for manual review; not independently verifiable (not a fabrication signal).
- NO DOIChondroadherin binds to type II collagenNo DOI in the reference — shown for manual review; not independently verifiable (not a fabrication signal).
3 data/code links checked; 3 live.
- datahttps://ngdc.cncb.ac.cn/omix/release/OMIX008341LIVEHTTP 200Resolves, but the content could not be matched to the paper.
- datahttps://ngdc.cncb.ac.cn/omix/LIVEHTTP 200Resolves, but the content could not be matched to the paper.
- codeGitHubLIVEHTTP 200https://github.com/HUICUI1992/Code-for-NMResolves to GitHub (code repository).
Copyediting
5 minorWording, consistency and formatting errors that need correcting before submission.
No major wording or formatting errors. 5 minor suggestions below.
5 copyedit issues flagged: mostly consistency, typo, clarity.
- MINORtypoAbstract“with and +98.4 ml”→ with +98.4 mlExtra 'and' in the sentence.
- MINORconsistencyResults, Primary safety endpoint“3 in 60 mg QD (20.4%)”→ 3 in 60 mg QD (16.7%)Percentage for 3/18 is 16.7%, not 20.4%.
- MINORconsistencyResults, Primary safety endpoint“6 in 60 mg QD (33.3%) and 3 in 60 mg QD (16.7%)”→ 6 in 60 mg QD (33.3%)Duplicate listing of 60 mg QD for ALT increase; likely a typo.
- MINORclarityMethods, Statistical analysis“The formula for modeling the treatment effect is ...”→ The formula for modeling the treatment effect is: NPX ~ treatment + visit + treatment:visit + (1|PatientID).The formula is presented in a complex LaTeX format; consider simplifying for readability.
- MINORconsistencyResults, Primary safety endpoint“6 in 60 mg QD (33.3%) and 3 in 60 mg QD (16.7%)”→ 6 in 60 mg QD (33.3%) and 3 in 60 mg QD (16.7%)Duplicate listing of 60 mg QD for ALT increase; likely a typo.
The published work is generally robust, but an informed reader should weigh the retrospective registration (submitted after study start), the lack of a formal efficacy power calculation, and the internal inconsistencies in the safety results (e.g., 20.4% vs 16.7% for 3/18). These issues warrant a correction or independent re-analysis to confirm the reported findings.
- 1.HIGHrigorCorrect the internal inconsistency in the primary safety endpoint: the percentage for 3 out of 18 in the 60 mg QD group for hypokalemia is reported as 20.4% but should be 16.7%.An impossible percentage (3/18 = 16.7%, not 20.4%) is a demonstrable error that undermines the credibility of the safety data.
- 2.HIGHrigorResolve the duplicate listing of ALT increase for the 60 mg QD group, which appears twice with different counts (6 and 3).The duplicate and inconsistent counts for the same adverse event in the same dose group is an internal contradiction that must be clarified.
- 3.HIGHreportingAdd a statement in the Methods or Reporting Summary explicitly noting adherence to the CONSORT reporting guideline, or justify its absence.Explicit reporting guideline adherence is expected for randomized trials and improves transparency.
- 4.HIGHstatisticsProvide exact p-values for all primary and secondary analyses instead of threshold-based reporting (e.g., 'P < 0.1').Threshold-based p-values are imprecise and hinder readers' ability to assess the strength of evidence.
- 5.HIGHstatisticsInclude a formal sample size calculation based on the primary efficacy endpoint (FVC change) in the Methods.The current power analysis is based on safety feasibility only, which is insufficient for interpreting efficacy results.
- 6.HIGHethicsClarify the timing of trial registration relative to study start; NCT05938920 was first submitted on 2023-06-28, after the study start date of 2023-06-19.Retrospective registration undermines the pre-specification of outcomes and should be acknowledged or corrected.
- 7.MEDIUMcopyeditFix the typo in the Abstract: 'with and +98.4 ml' should be 'with +98.4 ml'.The extra 'and' is a grammatical error that detracts from the manuscript's professionalism.
- 8.MEDIUMreportingSpecify the names of the institutional review boards or ethics committees that approved the study, with protocol numbers.Providing specific IRB names and protocol numbers strengthens the ethics statement.
- 9.MEDIUMdata codeClarify the data availability statement for clinical data by specifying the access mechanism and conditions for requesting the study protocol and SAP.The current statement says 'available upon request' without details, which is vague for clinical data sharing.
- 10.MEDIUMstatisticsReport the exact version numbers of statistical software packages used (e.g., R version, lme4 version) in the Methods.Version numbers are important for reproducibility.
- 11.MEDIUMreportingDiscuss the potential impact of the high withdrawal rate (22.5%) on the interpretation of results.A high withdrawal rate can bias results, and its impact should be explicitly addressed.
- 12.LOWreportingClarify the handling of missing data for endpoints other than FVC (e.g., DLCO, 6MWD).The multiple imputation is described for FVC, but other endpoints may have different missing data handling.
- 13.LOWstatisticsProvide a more detailed description of the multiple imputation model used for missing FVC data.Details on the imputation model (e.g., variables included, number of imputations) would improve transparency.
- 14.LOWdata codeConsider providing individual patient data in a repository with controlled access, as is standard for clinical trials.Sharing IPD would enhance reproducibility and secondary analyses.
The star rating is the report’s one-glance summary. Every paper starts at 5★ and loses stars for the concrete problems the review finds — so a rating is never a vague average, it’s a running total you can read line by line under “How this rating was calculated.”
- Reporting — 8 dimensionseach dimension that fully fails−½★
- each dimension partially met−¼★
- Statistics · Integrity · Claimseach serious problem−1★
- each medium problem−½★
- Citationseach retracted or unverifiable reference−¼★
- Copyeditonly when the manuscript needs a full edit−½★
The rating never drops below 1★, and a demonstrable critical failure (an impossible statistic, a proven ethics violation) caps it at 1★ on its own — so the stars can never look healthy when the verdict is CRITICAL.
The rating draws on a panel of agents. Three independent Kaimen Rigor reviewers grade the eight dimensions below across several independent passes (the shown verdict is their majority vote — steadier than any single run), isolate the paper’s major claims and check its own evidence backs them, and flag integrity concerns. Alongside them, a citation agent resolves every reference against Crossref, OpenAlex, and Retraction Watch; a statistics agent recomputes reported tests; and rule-based checks verify that declared data/code links actually resolve. Full text is required — an abstract-only submission is not analyzed.
Graded against NIH, MDAR, ARRIVE 2.0, CONSORT, EQUATOR, and RRID guidelines. A dimension that doesn’t apply to the study type is skipped, never penalized.
This Kaimen Rigor review is model-assisted and is not a substitute for formal expert review. It complements human evaluation by surfacing potential methodological concerns — verify each finding against the source.