Margins decide whether a good idea is a business. This skill builds the numbers
for a business that has not opened yet: what each sale leaves after its own
costs, how many sales a day cover the fixed costs, what year one looks like
month by month, how deep the cash hole gets before the business pays for
itself, and which assumption would sink it. It then computes a verdict from
those numbers and the demand evidence collected so far. The verdict is never
written by hand.
The agent treats every number as traceable: each cost line carries its source
(quote, public price, or estimate), and the report says which conclusions rest
on guesses.
that has not launched
lender can audit
For recurring-revenue software metrics (LTV, CAC, churn, MRR) use a SaaS metrics
skill instead; this one models per-sale businesses with a physical or service
capacity.
quote, public or estimateThe ramp and the largest cost lines shape this output more than anything else.
Settle the inputs below before producing anything. When one is missing or vague
and would change the result, ask about it; when it wouldn't, use a sensible
default and say which one you used.
computed per unit; an average ticket and a best-seller price give different
businesses.
and therefore break-even; an estimate here moves the verdict more than any
other line.
physically reachable.
not with the startup budget.
Ask about the two or three inputs that matter most for this request, not the
whole list. If the user wants a result straight away, produce it and state your
assumptions at the top, with every assumed line tagged estimate.
# 1. Model the economics (exit 1 if any check fails)
python3 finance/launch-economics/scripts/launch_economics.py \
--input finance/launch-economics/assets/sample_launch_numbers.json
# 2. Save the report as JSON for the verdict
python3 finance/launch-economics/scripts/launch_economics.py \
--input finance/launch-economics/assets/sample_launch_numbers.json \
--format json > report.json
# 3. Compute the verdict from the report and the demand evidence
python3 finance/launch-economics/scripts/viability_verdict.py \
--economics report.json \
--evidence finance/launch-economics/assets/sample_evidence.json
| Tool | Input | Output | Exit codes |
|---|---|---|---|
launch_economics.py | Numbers JSON | Contribution, break-even/day, margin at plan, 12-month table, cash needed, payback month, four what-ifs, findings | 0 clean, 1 fail finding, 2 bad input |
viability_verdict.py | Report JSON + evidence JSON | VIABLE / UNCONFIRMED / NOT YET / INCOMPLETE with every check and next actions | 0 VIABLE or UNCONFIRMED, 1 NOT YET or INCOMPLETE, 2 bad input |
Both tools are stdlib-only, deterministic, and support --format text|json.launch_economics.py also takes --price and --volume-factor for quick
what-ifs without editing the file.
average and record how it was estimated.
each line in the right bucket (see references/pre-launch-unit-economics.md).
a landlord's quote or the pay they plan to offer, are asked for, not invented.
month 1 and say how.
launch_economics.py. Resolve every FAIL before going further; eachWARN needs one sentence in the memo.
python3 finance/launch-economics/scripts/launch_economics.py \
--input numbers.json --format text
cannot afford to get wrong.
python3 finance/launch-economics/scripts/launch_economics.py --input numbers.json --price 52
python3 finance/launch-economics/scripts/launch_economics.py --input numbers.json --volume-factor 0.8
ones the founder can actually pull.
needs more volume than the counter can serve is not an improvement.
--format json.funding_available and every demand test withpeople reached, people converted, and whether the real price was shown.
viability_verdict.py. The verdict is the quality gate: the agent neverwrites VIABLE by hand and never rounds a failing number into a passing one.
action that fixes each. If UNCONFIRMED, the next step is a priced real-world
test aimed at the most sensitive driver.
assets/viability-memo-template.md with the same numbers, word for word.python3 finance/launch-economics/scripts/viability_verdict.py \
--economics report.json --evidence evidence.json --format json
Use this checklist; every box must pass before the memo leaves the founder's
desk:
| Verdict | Meaning | Recommended next step |
|---|---|---|
| VIABLE | Numbers pass and real people paid at the real price | [RECOMMENDED] Commit spend in stages, re-running the model on real month-1 data |
| UNCONFIRMED | Numbers pass; demand is simulated, unpriced or too small a sample | [PROVEN] Run a pre-sale or pop-up at the real price before signing a lease |
| NOT YET | At least one check fails | [PROVEN] Fix the failing number first; do not spend on brand or marketing yet |
| INCOMPLETE | An input is missing | Fill the gap; do not substitute a guess for a missing check |
| Measure | Healthy | Flag when |
|---|---|---|
| Break-even / capacity | Under 60% | Over 100% (fail): the business cannot break even at full stretch |
| Month-1 ramp / plan | 30-60% | 80% or more (warn): needs evidence |
| Payback | Within 12 months | Beyond 12 months (warn) |
| Estimated share of cost base | Under 50% | Over 50% (warn): get quotes |
| Funding / cash needed | 1.2x or more | Under 1.0x (fail), 1.0-1.2x (warn) |
[RECOMMENDED] Fund the trough plus 20%. Pre-launch timelines slip, and every
month of delay is a month of fixed costs without revenue.
[RECOMMENDED] Test a price rise before a cost cut. The whole price increase
lands in contribution, while most cost cuts touch quality the customer notices.
Cut costs first only when one variable line (often a platform commission) is
larger than the contribution itself.
Mistake: Raising exactly the startup budget and treating it as the money needed.
Why it happens: The fit-out and equipment list is concrete; early operating losses feel hypothetical.
Instead: Fund the cash trough the tool reports (startup spend plus early
losses) plus a 20% buffer. The trough is usually 20-50% above the startup total.
Mistake: Copying steady-state volume into month 1 of the ramp.
Why it happens: The plan volume was the number used to decide the idea was good, so it gets reused everywhere.
Instead: Start at 30-60% of plan unless a pre-sale or pilot measured more,
and cite that test. The tool warns at 80% for exactly this reason.
Mistake: Entering card or platform fees as fixed money amounts, then running a price what-if.
Why it happens: The fee was converted once at the original price and never revisited.
Instead: Recompute percentage-based variable costs at each tested price and
re-run. A 30% platform commission rises with every price increase.
Mistake: Treating a simulated buyer panel's buy rate as demand evidence and calling the plan viable.
Why it happens: The panel produced a number, and numbers feel like data.
Instead: Use the panel to find objections and segments, then confirm with a
priced real test. The verdict tool caps panel-only evidence at UNCONFIRMED.
Mistake: Summarising a plan as "profitable" because the steady-state month is positive.
Why it happens: The steady-state month is the most flattering view and the easiest to read.
Instead: Paste the computed verdict. Year one, capacity and funding are
separate checks, and a positive steady state can coexist with all three failing.
references/pre-launch-unit-economics.md — choosing the unit, filing costs,sourcing numbers, setting a ramp, reading outputs, improving a thin margin
references/demand-evidence-ladder.md — what each kind of demand test proves,why simulated buyers cannot confirm demand, picking thresholds
| File | Purpose |
|---|---|
scripts/launch_economics.py | Unit economics, break-even, 12-month cash model, what-ifs, findings |
scripts/viability_verdict.py | Computed verdict from the economics report and demand evidence |
references/pre-launch-unit-economics.md | How to build and read pre-launch numbers |
references/demand-evidence-ladder.md | Evidence strength, test selection, threshold rationale |
assets/sample_launch_numbers.json | Sound example: a bicycle repair counter (exit 0) |
assets/sample_launch_numbers_flawed.json | Deliberately flawed example: delivery kitchen (exit 1) |
assets/sample_economics_report.json | JSON report from the sound example, input for the verdict |
assets/sample_economics_report_flawed.json | JSON report from the flawed example |
assets/sample_evidence.json | Evidence that confirms demand (verdict VIABLE, exit 0) |
assets/sample_evidence_flawed.json | Panel-only and unpriced evidence with thin funding (NOT YET, exit 1) |
assets/viability-memo-template.md | The memo a verdict ships in |
In Scope:
12-month cash model with launch_economics.py
fixed cost)
viability_verdict.py
Out of Scope:
Important Caveats:
tagged estimate are guesses with reasoning, not facts.
a weighted average ticket and say so.
payroll costs and tax before money moves.