How it works
1
Send your spreadsheet
Columns of factors, plus one yes/no outcome column (converted / churned / approved…). 50+ rows ideally.
2
The forge searches
It hunts for the simplest formula that separates your two outcomes — then tests it on data it never saw.
3
You get a formula + verdict
A one-page report: the scoring formula, how accurate it is on held-out data, which factors matter, and whether it's real.
Proven on data we didn't create
We ran the forge on two famous public medical datasets — labels set by clinicians, not us — and it matched standard machine learning, as formulas a person can read. Every number is from held-out data the formula never saw.
0.93Heart-disease prediction (test AUC), from 13 clinical measurements — on a par with published models.
0.99Malignant-tumour detection (test AUC) — as three cell measurements multiplied together.
Why it's different
Most tools give you a black box that always sounds confident. This one gives you a formula you can read — and admits when it can't help.
✗ Typical AI tools
- A black box you can't inspect or embed
- Numbers from data it already saw (looks better than it is)
- Always sounds confident — even on noise
- Needs a data-science team to run
✓ Equation Forge
- A readable formula — runs in a spreadsheet cell
- Accuracy measured on held-out data only
- Tells you plainly when there's no real signal
- You just send a spreadsheet
£49
One spreadsheet · one formula · one honest report · 48-hour turnaround
Get your analysis
What you need: a spreadsheet (CSV or Excel) with a header row, numeric factor columns, and one column that's a yes/no outcome. 50+ rows of each outcome is ideal. What you get back: your scoring formula, its accuracy on held-out data, the factors that matter most, and an honest verdict — including a clear "no reliable signal here" when that's the truth. AUC measures ranking, not calibrated probability; pick a decision cut-off with your own judgement. Not financial, medical, or legal advice.