Signals in the original paper

For a diligence team, the question is whether the rigor problem was visible before the money moved. On each of these, the engine surfaces a concrete, text-level defect — an impossible statistic, verified inconsistent tests, or citations to since-retracted work — in the paper the case was built on.

Assets that later failed, retracted, or stalled — reviewed on the pivotal early paper, with retraction notices removed so the engine judges only the science.

25findings in 8 areas, across 3 of 3 papers

1 of these papers was later retracted. Look for the red RETRACTED band, then read what the engine flagged in the original published text, before any retraction notice.

How the star rating worksShow

Every paper starts at 5★ and loses stars for the concrete problems the review finds — never a vague average, just a running total:

Reporting — 8 dimensionseach dimension that fails (−½★) or is partially met (−¼★)−½ / −¼
Statistics · Integrity · Claimseach serious problem (−1★) · each medium problem−½★
Citationseach retracted or unverifiable reference−½★
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.

Papers are reviewed on their original full text — any retraction notice is removed before analysis — so the findings reflect the science, not the label.

Sort

3 papers · 3 critical · 1 later retracted — caught in the original text

Interventional · BMC Neurosci2/6Action Items · Critical / High

Chronic administration of R-flurbiprofen attenuates learning impairments in transgenic amyloid precursor protein mice

Kukar T, Prescott S, Eriksen JL, ... Koo EH, Golde TE

1 / 5

Significance claim flips when recomputedStatistical analysis inadequately reportedData and code not shared+5 more

The engine rates this 1★ (Critical): its statistics pass verified 5 of 13 reported tests as internally inconsistent (two reversing significance at p<0.05), alongside a mathematically impossible result (31% of 15 mice = 4.65). It also found no randomization or blinding described and no data-availability statement. The behavioral surrogate that carried the asset into a failed Phase 3 rested on this.

A reported p-value flips on recomputation

Conclusions

9 supported · 2 partial of 11

8 Rigor Dimensions

4 of 8 pass · 1 warn · 3 fail

3 Adcurare Rigor Reviewers · high confidenceAI Rigor Reviewers trained on Adcurare corpus · verify findings

Read paperPMC Journal

Interventional · Ann Neurol0/10Action Items · Critical / High

Longitudinal effect of eteplirsen versus historical control on ambulation in Duchenne muscular dystrophy

Mendell JR, Goemans N, Lowes LP, ... Eteplirsen Study Group

1 / 5

Mathematically impossible statisticData and code not sharedStatistical reporting gaps (tests, assumptions, effect sizes)+7 more

The engine rates this 1★ (Critical). It caught a mathematically impossible baseline — a 6-minute-walk mean of 363.2 m across 12 patients cannot arise from integer walk distances (the sum would be 4358.4) — plus the absence of a data-availability statement and missing confidence intervals on the primary effect. The comparison is against an external historical control, not a concurrent one.

Conclusions

11 supported · 1 partial of 12

8 Rigor Dimensions

3 of 8 pass · 4 warn · 1 fail

Integrity scan

4 findings

Verified

3/3 links liveregistered

3 Adcurare Rigor Reviewers · moderate confidenceAI Rigor Reviewers trained on Adcurare corpus · verify findings

Read paperPMC Journal

Retracted· PLoS ONE, 2008 — part of the Duke genomic-signature program; retracted amid misconduct findings after the signatures proved irreproducible and three Duke trials were halted (2010-2011).

Computational · PLoS One0/10Action Items · Critical / High

An integrated approach to the prediction of chemotherapeutic response in patients with breast cancer

Hsu DS, Balakumaran BS, Acharya CR, ... Potti A, Nevins JR

1.75 / 5

Cites retracted workBiological variables not reportedData and code not shared+4 more

The engine rates this 1.75★ (Critical). It shares no code or data for its chemo-response signatures, and its reference list cites three papers that have since been retracted. The reproducibility gap Baggerly and Coombes later exposed — the one that halted three Duke trials and led to this paper's retraction — is visible in the text itself.

Conclusions

5 supported · 6 partial · 1 overstated of 12

8 Rigor Dimensions

3 of 8 pass · 1 warn · 3 fail

Verified

1/1 links live3 retracted refs

3 Adcurare Rigor Reviewers · high confidenceAI Rigor Reviewers trained on Adcurare corpus · verify findings

Read paperPMC Journal

Share this gallery with your network

Post on X Share on LinkedIn Share via Email

How to read this. The engine surfaces specific, verifiable rigor problems. It does notpredict retractions, and a “pass” is not a guarantee of soundness. Papers are analyzed on their published full text with any retraction notice removed, so detection is content-based. Real-world outcome labels (retracted, landmark) are curator-assigned from the public record; the verdict and every surfaced problem are the engine’s own. These are selected examples.