Van Halen’s Brown M&Ms and Our Skill-Building Checklist
Van Halen’s 1982 tour rider had a clause buried deep in the technical specs: a bowl of M&Ms backstage, brown ones removed.
For decades people read it as rock star excess. It was the opposite. The band was hauling nine semi trucks of staging into arenas built for three. If David Lee Roth walked backstage and saw brown M&Ms in the bowl, he knew the promoter hadn’t read the contract. Which meant the rigging specs hadn’t been read either. Which meant the stage might collapse.
The M&Ms were a diagnostic. Not “did you read the contract, on a scale of 1 to 5?”
Check the bowl.
That’s the design philosophy behind every strategy diagnostic we build at Petrichor. As of July 2026 we’ve shipped 18 interactive workshop diagnostics and 19 free scorecard tools, and every one of them was built against the same internal checklist. This post is that checklist. All of it.
Every diagnostic is a test, not a validation
Each diagnostic is a single skill file that an AI facilitator (we build on Claude) runs one phase at a time. The shape never changes:
→ Six to eight phases. Never fewer, never more. → Opens on context: where the founder actually is, before any scoring happens. → The middle phases inventory, then interrogate, then score. → Closes on a named deliverable and a set of behavioral rules for the facilitator.
And before a single phase gets written, we write one sentence. The inversion line.
“You are not confirming what they believe. You are testing whether those beliefs hold under pressure.”
Every diagnostic has one. If you can’t write the inversion line, you’re not building a diagnostic. You’re building a quiz that agrees with people.
Behavioral questions, never attitudinal
This is the actual craft.
Ask a founder “how well do you understand your customer, 1 to 5?” and every founder alive is a 4.
Ask “how many customers have described your problem back to you, unprompted: 6+, 3 to 5, 1 to 2, or 0?” and the founder who’s failing can’t hide. There is no self-flattering answer to a count.
That one rule carries the whole system: every scoring question asks for receipts, never self-ratings. Counts. Past actions. Things that already happened. The founder doing the work sails through. The founder performing the work feels the floor drop.
The scoring is also selection. The right founder feels exposed and leans in. The wrong founder feels uncomfortable and leaves. If the questions let everyone off easy, the diagnostic is broken, no matter how polished the report looks.
One phase turns their own evidence against them
Every diagnostic includes an interrogation phase. It asks for evidence the founder already holds, then sets it next to their stated belief.
You said your differentiator is speed. You also said your last three deals took nine months to close. Walk me through that.
The contradiction has to come from the founder’s own mouth. The facilitator never asserts it. Assertions invite debate. Their own receipts don’t.
The output is a named artifact, not advice
Every diagnostic ends in a named map: the Belief Risk Map, the Pricing Authority Map, the Blind Spot Taxonomy. Written to a file the founder keeps.
The name matters. “Some feedback on your pricing” evaporates in a week. A Pricing Authority Map is an object. It gets opened in board meetings. Naming the artifact is the same move as naming the framework: whoever names the thing owns the conversation about it.
The facilitator rules, always last
→ One phase at a time. Never advance until the current phase is answered. → Push back on vague answers. Ask for the count, the example, the receipt. → Direct, unsoftened, never cheerleading. → The founder gets real value even if they never buy anything. Holding the insight hostage is the fastest way to lose someone forever.
Before anything ships
→ Dog-food it once on a real company. Not a hypothetical. A real one. → Confirm it’s distinct from every diagnostic already in the line. Overlap means you built a duplicate, and a duplicate means the framework wasn’t sharp enough to deserve its own name.
The checklist isn’t the moat
Anyone can copy the shape. Phases, scoring, a named deliverable. Go ahead.
What can’t be copied is knowing which questions have no self-flattering answer in a specific market. That knowledge comes from reps, not from a checklist.
side note: this is also why the brown M&M clause kept working long after Roth explained it publicly. Knowing tripwires exist doesn’t help you if you still don’t read the rigging specs.
A diagnostic built to flatter tells the founder they’re ready. A diagnostic built to test tells them where the stage collapses.
We’d rather you know.