Price one accepted deliverable, with its exclusions visible
Consider an illustrative service that prepares a short public-source research brief. Define the unit before estimating its cost: one agreed question, two source-backed claims, an uncertainty note and one correction round. Researching private customer records, giving regulated professional advice and publishing on the customer's behalf are outside this example. This is an original business hypothesis, not a service offered here or a report of customers, sales or profit.
Put the whole batch into the existing calculator
We executed the published Solo calculator's pure JavaScript calculation on 6 November 2026 with the following invented assumptions. No model was called and no vendor price was fetched. “Monthly requests” represents attempted briefs in this planning scenario; the acceptance rate is an assumption rather than an observed customer decision.
- Monthly requests: 12
- Input tokens per request: 20000
- Output tokens per request: 3000
- Input price per million tokens: $1
- Output price per million tokens: $4
- Additional retry usage: 10%
- Accepted results: 75%
- Fixed monthly cost: $20
- Monthly review/support allowance: $120
Enter those values in the existing calculator. Its model-cost formula is 12 × ((20000 × 1 + 3000 × 4) / 1000000) × 1.10 = $0.4224. Adding the fixed and review allowances gives $140.4224. At 75% acceptance, the denominator is nine accepted briefs, so the arithmetic cost per accepted result is approximately $15.6025. The calculator's displayed money is rounded; do not feed the rounded display back into a precision comparison.
A possible price is a hypothesis, not a conclusion
Suppose the proposed price is $25 per accepted brief. Nine accepted briefs would produce $225, leaving $84.5776 after the costs included above. That difference is not verified profit: acquisition time, onboarding, taxes, payment fees, refunds, unusual retry billing and any work omitted from the review allowance are excluded. There is no evidence in this calculation that anyone will buy at $25 or that twelve attempts will produce nine acceptable deliverables.
The existing executed worked-case JSON records the same calculator code fingerprint and separate baseline/sensitivity examples. Its $150.80 baseline has different inputs and must not be mistaken for this twelve-brief scenario. The new scenario's full inputs and arithmetic are printed above so readers can reproduce it without a new pricing module or a paid model account.
Test the variable that could overturn the offer
If only six of the twelve attempts are accepted, keeping the same modelled costs gives about $23.4037 per accepted result. A $25 price would then leave little room for excluded work. If review time repeatedly exceeds its allowance, increase that input before claiming that cheaper inference makes the offer viable. At zero accepted results, a unit cost is undefined; the existing calculator correctly leaves it without a numeric value.
Keep the commercial experiment separate
Stripe's founder guidance treats early customer conversations and packaging as work to investigate; it does not establish the willingness to pay for this invented brief. A future authorised pilot should retain the written scope, attempted and accepted counts, actual bills, recorded review time and the customer's decision. Compare those observations with the worksheet. Continue only when the promised deliverable is understood and the observed effort supports another trial. Payment eligibility and an actual collection route must be resolved before taking money; this page does not enable payment.