burn-test

The person who asks "and what does that cost?" Margins decide whether a good idea is a business. This skill finds them, with every number traceable to an input, and says plainly how many of those inputs are real.

Read ${CLAUDE_SKILL_DIR}/../receipts/rules.md first.

Tool in this folder:

python3 "${CLAUDE_SKILL_DIR}/unit_economics.py" founder/numbers.json --out founder/cfo.md --handoff founder

Handoff contract (stage 06 in founder/CONTEXT.md)

The margin is only as real as the weakest number it was computed from (rule 10 in rules.md). This skill takes each input's label in and hands each result's label on, so nothing simulated reaches /kill-or-build looking measured.

Inputs

Break-even takes its label from the price and the costs only (selling days just restate it per day). Quote every cost and back the price with a real sale, and break-even is MEASURED. The margin at plan also uses plan_per_day and the selling days, and year 1 uses the ramp: each result takes the weakest label of its own inputs, so those stay PLANNED unless their volumes have a stronger source.

Process: run unit_economics.py. It looks every claim id up in founder/evidence/claims.csv and gives each result the weakest label of its inputs.

Outputs

Step 1: gather the inputs

Write founder/numbers.json (shape: example.json in this folder, a walk-in matcha café). It needs:

Where the numbers come from, in this order:

1. The user's own quotes, invoices, leases and supplier price lists.
2. Public prices the user can check (a supplier's price list, a landlord's listing, a payment processor's published fees). Write the source next to the number in founder/cfo-sources.md.
3. Your estimate, clearly marked as an estimate, with the reasoning. Never present an estimate as a fact.

Log each input in the ledger, kind cost (or price for the price): a quote, invoice, lease or published price page is MEASURED with its evidence id; an estimate is PLANNED. The plan reports how many costs are backed.

Then write each claim id into numbers.json, so the label travels with the number:

"price": 6.50,
"sources": {"price": "c08", "ramp_per_day": "c11", "days_per_month": "PLANNED"},
"variable": [{"label": "Matcha (ceremonial, 2g)", "cost": 0.92, "claim": "c04"}]

A source is a claim id, a status, or a list of them (the weakest wins). Leave it out and the number is PLANNED.

Ask the user for anything only they know (rent quoted, wages they will pay). Use the price from founder/pricing.md or founder/pitch.md if those exist.

The ramp is where plans lie. Month 1 is almost never the plan volume. If the buyer panel ran, use its buy rate and first-month purchases to sanity-check the ramp (founder/panel/results.md), and say how you did it. The panel's numbers are an upper bound; if no real buyer has committed yet, the honest ramp is lower than the panel suggests, never higher. A ramp worked out from the panel is SIMULATED: say so in sources.ramp_per_day.

Step 2: run it

python3 "${CLAUDE_SKILL_DIR}/unit_economics.py" founder/numbers.json --out founder/cfo.md

It writes: the per-unit table and the contribution (what each sale leaves after its own costs), fixed costs, the break-even in units a day, the profit margin at the plan volume, year 1 month by month, the year 1 operating profit and the result after the startup spend, the cash needed before it pays for itself, the month the startup money is earned back, and a what-if table (price -10%, volume -20%, unit costs +15%). It flags a price below variable cost, a break-even above capacity, a losing year 1 and a payback beyond 12 months.

What-ifs without editing the file:

python3 "${CLAUDE_SKILL_DIR}/unit_economics.py" founder/numbers.json --price 7.00
python3 "${CLAUDE_SKILL_DIR}/unit_economics.py" founder/numbers.json --volume 0.8

Step 3: the CFO's note

Add a short section at the top of founder/cfo.md, in plain words:

• The hard truth: the number most likely to sink it, and how many of the inputs are quotes versus estimates ("4 of 11 costs are quoted; the ramp is a guess"). If the margin is SIMULATED, say which input made it so ("the margin is SIMULATED because the price came from the panel [c08]").
• The bad case: run --volume 0.5 and report it in one line. If there is no real demand evidence yet, this is as likely as the plan.
• The margin: the profit margin at plan, and the contribution per unit.
• The line to watch: the one cost or assumption that moves the result most (usually the ramp, rent or the price), with the what-if that proves it.
• Cash: how much money the founder needs before it pays for itself.
• What a customer is worth (Ready, Fire, Aim, rule 11 in rules.md): the most the business can spend to win one customer, from the tool, and what it costs now if a test has run. If no next product is planned, say so: the first sale then has to pay for everything, which is the hardest way to grow.
• Three ways to improve the margin, each with the number it changes and by how much (run the what-if, do not guess).
• Whether the board's conditions about money (founder/board.md) are met.

Rules

• Every number in your note comes from the tool's output or an input you can name.
• Every output takes the weakest label of its inputs, as the tool prints it: at best PLANNED for anything that uses the ramp or the plan volume, and SIMULATED whenever a simulated input is under it. Never write a stronger label than the tool printed, and never call a projected margin "the real margin" or a result.

• Log a result with --from and no --status, so the ledger works the label out and refuses a leak:

  python3 "${CLAUDE_SKILL_DIR}/../receipts/ledger.py" add --claim "Contribution per cup" --kind cost --use PRIVATE --number "$4.76" --from c08,c04,c05

• Revenue is before tax; say if the user's market adds sales tax on top.
• Not financial, tax or legal advice. Say once that an accountant should check the structure, payroll costs and tax before money moves.

Output

founder/numbers.json, founder/cfo.md, founder/cfo-sources.md. Next: /kill-or-build uses these numbers for the verdict; /litmus if the margin is thin.