R0
Reviewer Zero
Research Validity Audit
A research validity audit · v1.3

The reviewer that reads your methods before Reviewer 2 does.

Paste a Methods section — or start from just a research question — and get back a severity-rated validity audit, mapped to the checklist your journal actually mandates.

Runs in Claude Code· Cursor· ChatGPT· DeepSeek· Kimi· GLM· Qwen· any OpenAI-compatible API
Validity Audit Report
Severity-rated · mapped to your journal
FATAL Causal claim, cross-sectional design 1
MAJOR Mediator used as a control variable 3
MINOR Report McDonald's ω alongside α 4
Reporting-standard gaps
STROBE 12c — missing data not reported
STROBE 16a — confounder rationale absent
Limitations paragraph — drop-in ready
"The cross-sectional design limits causal inference; we therefore frame effects as associational..."
Example output — your audit is keyed to your study.
Why severity ratings

Not 30 nitpicks. A ranked list of what sinks the paper, and what's a one-line fix.

FATAL
Desk-reject material

Causal claims from cross-sectional data. A mediator treated as a control. Saturation claimed without a method. These send the paper back before review.

MAJOR
Reviewer will ask, today

Measurement invariance untested. α reported without ω. A control variable that isn't one. HTMT ignored. Fixable, but the reviewer will catch it.

MINOR
Polish, not peril

Report effect sizes. Add a flowchart. Cite the threshold you used. One-sentence fixes that signal rigor without changing the argument.

Different from "ask ChatGPT to review my methods"

Generic AI is confidently wrong on the contested cases. This isn't.

Every threshold cites its source.

α > .7, r > .6, κ > .7 are conventions, not laws. The pack tells you which, with the citation, so you can defend a number to a reviewer instead of parroting it.

Mapped to the checklist your journal mandates.

Reviewers don't audit against a textbook; they audit against STROBE or COREQ. Every report ends with a gap list keyed to item numbers you can act on today.

Gets the contested cases right.

Likert-as-interval, saturation vs information power, McDonald's ω over α, HTMT over Fornell-Larcker, "control for everything" as a mistake. Where generic AI advice is confidently wrong, the pack gets it right.

It rates severity — and stops at the right level.

Not a wall of nitpicks. A ranked list of which one sinks the paper, which three are one-sentence fixes, and which ones to leave alone because the reviewer won't care.

Covers case-based methods, not just numbers.

Process tracing — the four evidential tests (hoop, smoking gun, straw-in-the-wind, doubly decisive), mechanism specification, DA-RT transparency. QCA — calibration, necessity vs sufficiency, consistency and coverage. A narrated mechanism is not a traced one, and a high-consistency solution can be a calibration artifact.

Audits theory, not just data.

Formal models, conceptual frameworks, critical essays — load-bearing assumptions, derivation soundness, the tautology trap (claims true by definition marketed as discoveries), knife-edge robustness, and falsifiability judged against the model's declared purpose. It won't lecture a fables-style modeler about testability, and won't let an organizing framework wear a predictive abstract.

Works whether you've written it or not

Two entry points. One audit.

Mode 1 · Fast
Have a draft?

Paste your Methods section — or a proposal, or a pre-registration — and get the full report in about two minutes. Ideal for a resubmission or a deadline.

Severity-rated findings
Reporting-standard gap list
Drop-in limitations paragraph
Mode 2 · Coach
Still designing the study?

Start from your research question. The pack walks you through the design one step at a time — beginning at conceptualization — and helps you fix problems before you've collected a single data point.

The cheapest stage to catch a fatal flaw
DAG-based confounder selection
Identification before estimation
Mapped to the checklist your journal mandates

Six reporting standards. Keyed to item numbers.

CONSORT STROBE PRISMA COREQ SRQR GRAMMS
Every threshold cites its source. Every gap maps to an item number. Nothing is a vibe.
What's in the box

One download. Everything you need to run a validity audit on any study.

5 method skills + 1 router

Quantitative, qualitative, mixed-methods, case-based (process tracing + QCA), and pure theory (formal models, frameworks, essays). Install in Claude Code / Cursor, or paste as a system prompt anywhere. The router picks the right one if you're not sure.

Self-contained system prompts

For DeepSeek, Kimi, GLM, Qwen, ChatGPT, and any OpenAI-compatible API.

6 worked examples

5 fast-mode audits of realistic manuscripts (including a process-tracing and a theory audit) + 1 coach-mode walk-through of designing a study from scratch.

A 126-point printable checklist

Run it yourself, on paper, when you don't want to fire up an AI.

A visual diagnostic flowchart

Light and dark variants. Follows the same logic as the audit skills.

Plain-English license & install guide

For every platform. No DRM, no tracking, no subscription.

After download

One command. Or just ask your AI to install it.

$ ./install.sh

Unzip the download, then tell Claude Code, Cursor, or Codex: "Install Reviewer Zero."

Using ChatGPT, DeepSeek, or Kimi? Copy the prompt file from the download, paste it into your AI. No install needed.

Honest about what it is

A methods coach. Not peer review. Not a guarantee.

It runs on an LLM, so it can miss things and raise false alarms — every finding is a prompt for your judgment, not a verdict. And it's built to help you think about your study, not to write it for you: the drafts it produces are scaffolding to rewrite in your own voice, and you should disclose AI use as your journal requires.

Run it before Reviewer 2 does

Catch it before the reviewers do.

One-time purchase. Free updates. Refunds under Gumroad's standard terms.

Get Reviewer Zero — $199
One download · No subscription · Runs in the AI you already use
Get Reviewer Zero — $199