Fail here first.
Adversarial AI that attacks your plan before reality does. A Panel of hostile specialists finds the failure modes you're avoiding — and names the one you must fix first.
Never training dataEU-hostedDelete anytime
We ran BadMars on our own business plan.
It hurt. It was also the most useful document we'd ever read about ourselves. The Frog said one person was the whole company — so we ate it: BadMars is a small team now, with specialists hired where the report named the gaps. That's the product, tested on its maker first.
EVERY REPORT: RANKED FINDINGS · VERBATIM ANCHORS · A HANDLE PER FINDING · THE FROG · THE SCORE
Read the full specimen reportEveryone around you wants your plan to succeed. That's the problem.
Your friends encourage. Your advisors hedge. Your chatbot congratulates you on a great question. Reality does none of those things — it just charges you for every flaw you didn't find. BadMars is the place to meet those flaws early, while they're still cheap.
Not opinions. A procedure.
- 01Evidence
Your plan is treated as evidence
Every finding must anchor to a verbatim quote from your own document. If the Panel can't point at it, it can't file it.
"Our average order value is ninety four euros and rising" - 02The Panel
Twelve specialists. Each one paid to find what the others missed.
Staffing is triage-assigned per Mission — your plan decides who shows up. The Ascent bench reads plans that need to survive a market: business plans, launches, pivots, pricing.
- 01The Hostile InvestorAttacks your unit economics and your moat — with your own numbers.
- 02The Skeptical CustomerReads your value claim and asks why anyone would switch.
- 03The Aggressive CompetitorEstimates how fast your idea can be cloned. Usually: a weekend.
- 04The OperatorCounts the humans your plan assumes and the hours it forgot.
- 05The CFOFinds the number your projection quietly depends on.
- 06The RegulatorNames the compliance exposure you filed under "later."
- 07The JournalistWrites tomorrow's headline about today's shortcut.
- 08The TaxmanFollows your cross-border setup to its expensive conclusion.
The Transfer bench — The Hiring Manager, The Recruiter, The Accountant and colleagues — reads career moves the way the other side will. New specialists are drafted, briefed, and cross-examined before they ever take a seat; Colony settlers meet them first. Meet the full Panel
- 03Cross-examination
The Panel attacks its own findings
A second pass tries to kill every finding. What survives is ranked hardest first — severity times probability, coded against a failure taxonomy built from real plans, not vibes, not a checklist from a blog. Weak findings don't make the report.
- 04The Score
Computed by code. Never by mood.
The model files the findings. The formula sets the number:
SCORE =
round( 5 + 95 · e−(burden / seats) / 12 )The model never sets this number. Findings carry weight · weight becomes burden · the curve does the rest
One finding is the Frog. Eat it, and come back.
Old productivity advice, usually pinned on Mark Twain: if you have to eat a frog, eat it first thing in the morning — nothing worse will happen to you all day. Ours is stricter. Every report names exactly one Frog: the hardest problem, the one the rest of your plan quietly depends on, the one you've been avoiding. Fix it, return with evidence — not promises — and the Panel re-convenes to move your score.
The green in that panel is the only green on this entire page. Green is earned.
Research · The failure map
The 53 ways a business plan fails.
Every failure mode the Panel is trained to find, coded into 15 families — from unit economics to the one-way door.
- Execution5
- Unit economics5
- Market4
- Cash & runway4
- Value & positioning4
- Distribution4
- Regulatory4
- Talent & hiring4
- Pricing3
- Moat & defensibility3
- Competition3
- Tax & structure3
- Personal runway3
- Reputation2
- Irreversibility2
The ladder is the trial.
For individuals: your plan, attacked. For teams — agencies, studios, strategy rooms — every plan on the table, with shared seats, client separation, and comparison views on Fleet and Armada.
One Mission a month. One specialist, redacted bench, three Debrief exchanges. Enough to hurt.
Send your plan to MarsThe full Panel on up to ~80 pages. Complete report, PDF, one free re-run, 48-hour full-refund guarantee.
See the full ladderTen Missions a month, unlimited Debrief across all Programs, Frog history, Score trajectory.
See the full ladderTeams start at Fleet €99/mo · 20 missions · 3 seats — the full ladder
Before you ask.
The questions every founder asks before pasting their plan into a website run by a hostile Panel. Fair questions. Straight answers.
What happens to my plan after I send it?
It gets attacked, and that's all. Your plan is never training data — not for us, not for our model providers, who are contractually barred from it. Your report stays private unless you share it — and a shared link shows the Score, the Verdict, and two findings, never the plan itself. Infrastructure is EU-hosted, GDPR applies in full, and you can request deletion of your payloads, reports, and account at any time.
How is this different from asking a chatbot?
A general chatbot is built to be helpful, so it finds polite problems and congratulates you for asking. The Panel runs versioned specialist briefings against a coded failure taxonomy, must anchor every claim to your own words, cross-examines its own findings, and computes the score with a formula — not a mood.
What should I send?
Anything that has to survive contact with reality: a business plan, a launch, a pivot, a pricing change, a career move. Paste the text — a Flyby reads the essentials, a Landing digests up to roughly eighty pages.
Can AI criticism actually be trusted?
Trust the procedure, not the vibes. Every finding points at a verbatim quote from your document, survives a cross-examination pass, and arrives with a handle — the concrete next move. Where a finding is weak, it dies before you see it. And we say plainly what we are: an AI system, disclosed on every surface.
What if I disagree with a finding?
Good — argue. Every report opens a Debrief: press a finding, bring your counter-evidence, and the Panel defends its position, re-ranks hypothetically, and pressure-tests your reasoning. The filed report never changes to make you feel better — but a finding you can defeat in Debrief teaches you as much as one you can't.
What if the report teaches me nothing?
First Landing carries a 48-hour full-refund guarantee. If the Panel taught you nothing, take the money back — we'll note the plan was flawless and update your score to 100. (We won't.)
Does BadMars work in ChatGPT or Claude?
Yes. BadMars is in both directories. Ask the assistant to review your plan with BadMars, and the score, the verdict and the Frog come back into the conversation. The full report stays on badmars.com. How to connect: AI assistants.
The Panel also takes meetings elsewhere.
Launch a Mission from a Claude or ChatGPT conversation, or carry the Panel on your phone. Same account, same Missions, same report on badmars.com. How connecting works →
Hostile to the plan · loyal to the person
Reality is already reviewing your plan.
It just hasn't filed yet.
Send it to Mars first. Get the findings while they're cheap, eat the Frog, and walk into the real review already knowing the answers.
Never training dataEU-hostedDelete anytime
As featured on