Audits the numbers before you wire
You are about to wire on a retention chart nobody checked
The experiment was stopped on the fourteenth look. Forty-one percent of the retention cohort has not been around long enough to churn. Both are recoverable from the deck the founder already sent, in forty-eight hours.
The problem
Every fund has someone who reads the code and nobody who reads the statistics
Technical diligence checks whether the product works. Commercial diligence checks the market. The claims in between — this experiment lifted activation 15%, retention is 92% — are taken at face value, because checking them requires a statistician and the partner has a week. Founders are not usually lying. They are usually peeking, and nobody told them that inflates the false-positive rate to 51%.
The insight
Four failures account for almost every overstated metric, and all four are checkable from the summary
Stopping an experiment when it looks good, counting customers who have not had time to churn as retained, reading a seasonal trend as a treatment effect, and claiming precision a sample cannot support. None of these require the raw event log — they require the numbers already in the deck plus the questions nobody asks. That is what makes a forty-eight hour turnaround possible.
Alpha inflation from repeated looks, achieved power against the claimed effect, censoring-corrected retention on the observed cohort only, and severity grading that separates a fatal claim from a qualified one.
How it works
Four steps, no data science team
The claims, the sample sizes, and how many times they looked. That last one is the question nobody asks.
Four failures, each either present or not, with the corrected number where one exists.
Findings by severity, in language you can forward to the partnership.
Usually a conversation rather than a confrontation. Most of them did not know.
Who it is for
The partner who has to defend the memo
Seed and Series A funds writing cheques on metrics, and founders who want their numbers checked before a partner does it for them.
Pricing
- –One claim
- –Written finding
- –48 hours
- –All claims checked
- –Severity-graded memo
- –Corrected figures
- –48-hour turnaround
- –Unlimited deals
- –Portfolio monitoring
- –Founder education sessions
- –Priority turnaround
Competition
What exists, and what it does not do
| Who | What they do | The gap |
|---|---|---|
| Technical diligence firms | Review code quality, architecture and team. | They read the repository. Nobody in that engagement checks whether the growth claims survive contact with statistics. |
| Data rooms and metrics tooling | ChartMogul, Metabase exports and a standard KPI pack. | They present the number the founder computed. Presentation is not verification, and the censoring error is upstream of the chart. |
| The associate with a spreadsheet | What actually happens today. | Recomputes the same number the same wrong way, because the error is methodological rather than arithmetic. |
| Trusting the founder | The default. | Usually fine, occasionally expensive, and the failure is invisible for eighteen months. |
Episodic and relationship-driven: funds buy on trust and referral, and there is no way to shorten that. The work is also adversarial in a way that makes repeat business awkward — the fund is happy, the founder is not, and the founder ecosystem is small. It generates cash and an unusually good network rather than a company, and the honest framing is that its main product is the twenty conversations with funds, not the fee.
Market
Priced per deal against the size of the cheque, which makes $4,000 a rounding error
A few hundred active seed and Series A funds. Fifteen retainers is $1.1M ARR, and the ceiling is set by how many relationships one person can hold.