Run your hypothesis through 18 attacks before a reviewer does it for you.
Hypothesis Stress Test is a workbench for the adversarial self-examination most students do only once it's too late. You enter a hypothesis. The catalog presents 18 curated attacks drawn from philosophy of science, causal inference, and meta-research literature. You write the defences, rate your own confidence, and acknowledge what remains weak — and walk out with a vulnerability map you'd otherwise only assemble in front of an examiner.
Your hypothesis statements, defences, evidence and notes stay in your browser. No AI evaluates them; nothing is uploaded.
The attacks that arrive at viva or peer review have been on the literature's shelf for fifty years
You will be asked about alternative explanations, confounding, reverse causation, selection effects, construct validity, falsifiability, mechanism plausibility, robustness — and the question is whether you've engaged with each one before someone else does it for you.
- Supervisor stress-testing is sporadic and gentle. Most supervisors pull punches that real reviewers don't, or share the same blind spots.
- The literature exists; the workflow doesn't. Popper, Kuhn, Pearl, Hernán — the methods are well-developed in philosophy of science and causal inference, but no tool ports them into the rhythm of preparing a paper, grant, or chapter.
- You can't outrun an attack you haven't named. A hypothesis you've never explicitly attacked is one you cannot honestly claim to have defended.
A curated catalog, not an AI critic
How it works
Enter a hypothesis
Required: statement and discipline. Optional PICO-style fields (population, exposure, comparison, outcome, mechanism, scope) if your work is empirical.
Work an attack
Read the scholarly grounding, the attack statement, and the sub-prompts. Write your defence, rate your confidence (1–5), cite evidence, acknowledge what remains weak.
Generate the vulnerability map
Markdown exports of your honest standing on each hypothesis. Pro features included: cross-hypothesis comparison matrix, defended-positions document, bibliography.
What's inside
18 curated attacks
Alternative explanations, reverse causation, confounding, selection effects, construct validity, mediator-vs-moderator confusion, statistical-conclusion validity, ecological validity, external validity, p-hacking exposure, garden-of-forking-paths, underdetermination, theory-ladenness, falsifiability, mechanism plausibility, robustness across operationalisations, dose-response coherence, sensitivity to influential cases.
Honest defence capture
Engagement status (engaged / skipped / not applicable), self-rated confidence 1–5, written defence, evidence citations, and an explicit acknowledged-residual-vulnerability field. Substantive edits archive prior versions automatically.
Deterministic adversarial follow-ups
Save a defence at low confidence, or briefly, and the catalog surfaces sharper follow-up prompts the author anticipated for that situation. Authored, not AI-generated — every follow-up is fixed text triggered by predicates on your saved defence.
Vulnerability map exports
Per-hypothesis Markdown or plain text. Grouped by self-rated confidence; low-confidence attacks flagged as open vulnerabilities; an honest summary footer makes clear the tool measures engagement, not quality.
Cross-hypothesis matrix
Every catalog attack against every hypothesis in your body of work, with confidence-coloured cells. Surfaces which attacks recur as weak across your projects — likely systematic weaknesses, not one-off issues.
Defended-positions document
Long-form Markdown compilation of all engaged defences across all selected hypotheses, with deduplicated bibliography. Suitable for inclusion in a methodology appendix or thesis chapter.
Privacy & Trust
- Your hypotheses never leave your machine. Statements, defences, evidence, residual vulnerabilities, internal notes and version history live in
chrome.storage.localon your device. - No AI runtime, no telemetry, no analytics, no account. The extension does not read any web page and runs no inference on your content.
- One optional network call, only if you ask for it. If you enter an email and click Subscribe for product updates, only that email address is sent to GradSummit's sign-up service — never any hypothesis content, defence text, or notes.
- Internal notes are never exported. The private "internal note" field on every hypothesis and defence is excluded from every export by design.
It is a workbook for adversarial self-examination — not an AI critic, and not a coach. Read the full privacy policy.
Built by Dr. Rafiq Muhammad, PhD — author of the Mastering Research book series.
Pricing
Every feature, for everyone — no subscription, no tiers, no account.
All 18 attacks, the adversarial follow-up tree, cross-hypothesis comparison, the defended-positions document, and every export are unlocked.
Email is entirely optional and opt-in — leave it blank forever, or add it to hear about occasional updates.
Frequently asked questions
Does an AI evaluate my hypothesis or defences?
No — and deliberately so. There is no AI runtime anywhere in this tool. It does not judge your reasoning, suggest defences, generate attacks, or evaluate quality. The catalog of 18 attacks is curated authorial work; the adversarial follow-up branching is deterministic predicates on what you wrote. "Stress test" here means a workbook of structured attacks, not a coach.
Is my hypothesis or defence text uploaded anywhere?
No. Every hypothesis statement, defence, evidence citation, residual vulnerability, internal note, and version history stays in chrome.storage.local on your device. The only thing the extension can send to GradSummit is your email address — and only if you explicitly opt in to product updates and click Subscribe. No content is ever transmitted. See the full privacy policy.
What does "engagement scoring" measure?
Three deterministic counts: engagement coverage (what fraction of catalog attacks you've touched), high-confidence coverage (what fraction you've rated 4 or 5 on), and acknowledged-vulnerability coverage (what fraction have an explicit residual you've named). Every number is computable by hand from what you've entered; nothing is AI-derived, and none of it is a quality grade.
Which 18 attacks are in the catalog?
Five in the Causal / Inferential category (alternative explanations, reverse causation, confounding, selection effects, mediator-vs-moderator confusion, statistical-conclusion validity), five in Methodological (construct validity, ecological validity, external validity, p-hacking exposure, garden-of-forking-paths), four in Philosophical / Foundational (underdetermination, theory-ladenness, falsifiability, mechanism plausibility), and three in Empirical / Robustness (robustness across operationalisations, dose-response coherence, sensitivity to influential cases). Each opens with scholarly grounding and a worked example from the literature.
How much does it cost?
It's free, with no subscription and no paid tiers. Every feature is available to everyone. There's nothing to buy and no account to create.
Can I get my data out?
Yes. Per hypothesis you can export a vulnerability map as Markdown or plain text; across hypotheses you can export a cross-hypothesis comparison matrix, a defended-positions document, and a deduplicated bibliography. A full JSON backup of everything you've written is one click in Settings, as is re-import and clear-all.
Will it remind me by email, push, or calendar?
No. The discipline of opening the workbench is part of the practice. There are no notifications, push reminders, or calendar syncs. If those are essential to your workflow this isn't the right tool.
What if I find a bug?
Email support@gradsummit.com — replies go to a real human (the developer).