Hypothesis Stress Test

A workbench for adversarial self-examination of research hypotheses. 18 curated attacks. Your work stays on your device.

A workbench for adversarial self-examination of research hypotheses. THE PROBLEM Graduate students and researchers rarely subject their own hypotheses to genuine adversarial examination before a reviewer, examiner, or supervisor does it for them. The result is the universal experience of having a central idea attacked at viva or in peer review and realising — too late — that the attack identifies a weakness you had subconsciously avoided thinking about for two years. Pre-mortem methods exist in industry. The methods exist in the literature — Popper, Kuhn, Lakatos, Quine in philosophy of science; Pearl, Rubin, Hernán, Rosenbaum in causal inference; the meta-research literature on what tends to go wrong in published research. The synthesis into a usable research workflow does not exist as a tool. Researchers re-invent it badly or skip it entirely. WHAT THIS TOOL DOES It presents, every time, the same eighteen durable attacks against research hypotheses. For each attack you decide whether it applies, write a defence, rate your own confidence, and acknowledge what remains weak. At the end you have a vulnerability map — an honest accounting of where your hypothesis is well-defended, where it is thin, and where you have not yet looked. Eighteen attacks across four categories: CAUSAL / INFERENTIAL • Alternative explanations • Reverse causation • Confounding • Selection effects • Mediator-vs-moderator confusion • Statistical-conclusion validity METHODOLOGICAL • Construct validity • Ecological validity • External validity / generalisability • P-hacking exposure • Garden-of-forking-paths PHILOSOPHICAL / FOUNDATIONAL • Underdetermination • Theory-ladenness of observation • Falsifiability • Mechanism plausibility EMPIRICAL / ROBUSTNESS • Robustness across operationalisations • Dose-response coherence • Sensitivity to influential cases Each attack opens with a scholarly grounding paragraph citing the source literature, the attack statement itself, three to five sub-prompts forcing specific engagement, a "How to defend well" guidance paragraph, and a worked example from a published critique or famous failure case (Mozart effect, the WHI hormone-replacement reversal, ego depletion, the marshmallow studies, social-priming, and more). WHAT THIS TOOL IS NOT It does not use AI. It does not evaluate the quality of your defences. It does not score, suggest, complete, or rewrite anything you produce. That is the point. The skill the workflow exists to build is your own adversarial self-examination; AI grading the defences would short-circuit the meta-skill. Substantive evaluation of a defence requires a substance-aware mind. The tool measures engagement — never quality. EVERYTHING INCLUDED, FREE No Pro tier, no subscription, no licence key, no account, no trial. All eighteen attacks are unlocked. The adversarial follow-up branching (deterministic, authored — not AI), cross-hypothesis comparison view, defended-positions document, and bibliography exports — all free. PRIVACY POSTURE Your hypothesis content stays on your device, in chrome.storage.local. The shipped code can reach exactly one external host — the GradSummit sign-up Worker — and only when you explicitly opt in to product updates and click Subscribe. In that case, and only that case, exactly the email address you typed (plus a tag identifying this extension) is sent. No hypothesis content, defence text, evidence, residual vulnerabilities, internal notes, or version history are ever transmitted. The opt-in is off by default; you never need to provide an email to use any feature. Internal notes are never included in any export. The vulnerability map, defended-positions document, comparison matrix, and bibliography all read from the public defence fields only — the privacy contract is enforced by code. OUTPUTS • Vulnerability map (Markdown / plain text) — single hypothesis, grouped by your self-rated confidence, with an honest summary footer. • Defended-positions document (Markdown) — long-form compilation across multiple hypotheses, with deduplicated bibliography. • Cross-hypothesis comparison matrix (Markdown) — every catalog attack against every hypothesis, with colour-coded confidence cells. • Bibliography (Markdown) — deduplicated, alphabetised across all defences. • Full JSON backup (Settings) — for device migration or backup. WHO THIS IS FOR Doctoral and master's research students preparing dissertation chapters, conference papers, or pre-defence. Postdocs preparing first-author papers, fellowship applications, K-awards. Faculty preparing major papers and grant proposals. Methodology educators teaching adversarial self-review. Anyone making causal or empirical claims who would rather catch the error before a reviewer does. The catalog is discipline-agnostic — it applies across empirical research broadly, from psychology to epidemiology to political science to ecology to applied physics. AUTHOR Built by Dr. Rafiq Muhammad, PhD. The catalog is original synthesis drawing on the cited primary sources. No AI was used to author the catalog content. Free. On-device. No account. No subscription. No AI.