Nabu shows you which papers hold up to rigorous scrutiny. A structured, blind review of any paper's strengths and weaknesses, with the journal and citation count left out of it.
The signs were in the paper
The warning signs are in the paper itself - methods that don’t match the claims, conclusions the design can’t support. Nabu reads for exactly those, blind. Three real cases:
Carmen Reinhart, Kenneth Rogoff · 2010
Foundational — 6,300+ citations, top-tier journal.
A descriptive correlation read as a hard policy threshold. Study Quality 3.0 (Moderate). Later contested in replication.
Noell et al. · 2024
Skip it — a preprint with 5 citations.
Rigorous work that doesn’t wait for a journal to be good. Study Quality 4.2 (Strong).
Gautret et al. · 2020
Trust it — 4,900+ citations, top-quartile journal.
Methodological failures legible from day one. Study Quality 1.7 (Very Limited). Retracted.
The evidence
Benchmarked to peer review
vs the 5.0 best-human-reviewer baseline (H-Max) defined in Google’s ScholarPeer study. Graded blind on sample papers and their open reviews.
Inter-rater reliability
More than double the human peer-review benchmark of 0.34. Nabu’s reviewers independently reach an ICC₂ of 0.81 (absolute agreement) across all ten scoring dimensions.
Potential retractions flagged
Concerns flagged blind by weak/poor craft score, or material reliability flag. When a Nabu assessment is appended to a paper’s metadata, these weak articles don’t surface in downstream AI answers.
Why it works
We force it through a rubric calibrated to the paper’s field and methodology — explicit criteria, scored component by component.
The rubric is the evaluator. The AI is the instrument.
Every score points to the line in the paper that earned it. When your expertise says otherwise, you can see exactly what Nabu saw and overrule it in seconds.
Several independent reviewer models score every paper blind to author, journal, and institution, and an adjudicator resolves disagreement on the strength of evidence.
The rubric was calibrated with practising methodologists across fields. Its performance is monitored continuously, and an evaluation escalates to a human reviewer when a paper falls outside what the AI can reliably assess.
Nabu has no relationship with any publisher, journal, or institution it evaluates. There’s no version of a score that pays us more. The independence is structural, not a promise.
A QUALITY-ASSURANCE LAYER FOR YOUR ORGANIZATION
The literature you’re deciding on is degrading. Retractions hit record highs, paper-mill output is doubling faster than legitimate research, and a growing share of papers cite studies that were never published. The AI tools your teams increasingly rely on can’t tell the difference. The checks that should catch this are inconsistent, and invisible from where you sit.
The warning signs are almost always in the paper itself: methods that don’t match the claims, conclusions the design can’t support, references to evidence that doesn’t exist. Nabu reads every paper blind for exactly those signals, so the calls you fund, reward, and build on are ones you can defend.
Currently collecting feedback in Research Preview