Lesson 6 · The conservative turn
Strict WTP: "you can charge" vs "they paid"
Willingness-to-pay is the single most gameable signal in any corpus of creator content. The final rule: WTP exists only when money actually moved — a payment, a commitment, a budget, real buyer behaviour.
What counts — and what never does
| OBSERVED ✓ | NEVER ✗ |
|---|---|
| "The client paid $4,000 and signed a retainer." | "You can charge $2,000 to $5,000 a month per client." |
| "Businesses are paying $10,000 for AI agents." | "Target: $8,000 to $15,000 per month." (a goal) |
| "O cliente pagou R$2.000 por uma aula." | "A $5/mo VPS gets you started." (tool cost → spend) |
| "Three customers signed annual contracts with a deposit paid upfront." | "monthly recurring revenue… 5% of the end goal." (creator MRR journey) |
| "started generating revenue with it." | "signed a three-year lease" (not a commercial object) |
| "a single HVAC job pays $300-800" (domain pricing, not buyer WTP for your offer) |
WTP signals after v4
The strict compound (numeric money × buyer verb × no advice-marker) cut corpus WTP by 97%. What survived is evidence a founder can put in front of an investor without blushing.
DO_NOW candidates
Not one theme×persona cluster holds observed multi-package WTP. The honest output is zero DO_NOW — anything else would be padding.
WATCH candidates
Real ICPs with real pain and a cheapest falsification test each. WATCH does not mean ignore; it means run the next test before building.
Why this is the product's spine
A queue that inflates WTP teaches a founder to spend on fantasies. A queue that tells the truth — "the corpus shows pain, but nobody was observed paying" — tells them exactly which test to run next. That is the difference between an opportunity list and an evidence machine.