A verdict you can't check is just an opinion delivered confidently. So here is where the numbers come from, what they are actually worth, and where this method is weak.
Every number in your report is computed by one deterministic function from the inputs you typed. Nothing is generated by a language model. The comparison bands are curated industry ranges — not your data, and not an audited dataset.
You give a revenue goal and a handful of assumptions. From those, one module decomposes the goal downward: revenue into customers, customers into leads, leads into traffic, traffic into budget. Same inputs always produce the same output — there is no sampling, no temperature, no run-to-run drift.
Then we compare what your goal implies against ranges for your vertical: visit-to-signup, signup-to-paid, monthly churn, and the realistic cost and ramp time of each acquisition channel. The verdict is the gap between the two — not a score we invented.
Curated industry ranges, expressed as a 25th / 50th / 75th percentile band per vertical. They are assembled from public reporting and operator experience — which makes them directional: useful for seeing whether you sit far outside normal, not for justifying a number to three decimals.
We hold ourselves to one rule: every band carries a visible status — sourced when there is a public citation, directional when it is an informed estimate, own-data when it is computed from aggregated usage. Today most bands are directional. We would rather label that than let an estimate pass for a fact.
This is the line we do not cross. Arithmetic lives in one deterministic module; the language layer exists only to explain what that module produced. A model that both computes and narrates can talk itself into a number, and you would have no way to tell. In the current build the wording is templated and deterministic as well — there is no model in the loop yet.
Naming the soft spots is not a disclaimer — it is the reason the rest is worth reading.
No. It is an outside read on whether a number is plausible, given ranges from comparable companies. Treat it as a second opinion before a decision, not as the decision.
Not yet. Per-band provenance labels are being built; until they ship, assume every band is directional. If that is not good enough for your decision, it should not be — and we would rather you know now.
Not today. If we ever do, it will be anonymous aggregation, labelled own-data with the sample size shown — never your figures exposed to anyone else.
Coverage and provenance are separate axes, and we are honest that breadth currently costs us depth of sourcing. The plan is to anchor one vertical with real citations and let the others earn their sources over time, rather than quietly present estimates as facts.
No spreadsheet, no card. The verdict first, the numbers underneath it.
Run the report