Proof
No borrowed logos. Numbers you can check.
We're a small, senior team early in our story — a few pilots so far. Rather than fake a wall of clients, we show you exactly how we measure, one anonymized pilot labelled as an illustration, and what we will not claim.
On a trust-based service, a real method beats a fake track record — and it's the only proof we'll put our name to.
One anonymized pilot
A sanitized scorecard, labelled as an illustration.
One workflow for one client, anonymized — not a promise your numbers will match. Every real pilot gets a scorecard in exactly this shape.
Before
per manual handoff
After
automated handoff
Sample
handoffs measured
Read this first: the figures above are an illustration of the scorecard shape, not a DPR track record. When we have a real anonymized pilot cleared to publish, it replaces this — same format, with the client's permission.
The scorecard, blank
What yours will look like.
Every real pilot ends with this filled in from your data. Nothing here is pre-filled, because your numbers are yours to measure.
Scorecard — [your workflow] · your pilot
- Beforebaseline, from your real data
- Aftermeasured on a human-checked sample
- Samplen = how many items we measured
- Baseline windowthe pre-pilot period we measured against
- Methodexactly how we timed/counted it, in plain English
- Who checkeda human on your side signs off
- Caveatthe honest limits — one workflow, one sample
How we measure
A number you can't re-check is marketing, not proof.
We set the baseline from your real data before we touch anything. At the end we run the system on a real, human-checked sample and report the same number the same way. We write the method down — the sample size, who checked it, and every caveat — so you, or anyone, can reproduce it. Honesty about a small sample reads as more credible than a big round number, so we show the sample size, not hide it.
The proof here is the method — and it's all on this page. Nothing is hidden behind a case study you can't read. When our founders publish their own profiles — GitHub, talks, shipped work — they link here too, real and checkable or not at all.
What could go wrong
The honest limits, said out loud.
- Your workflow may not have a clean number to move. We'll tell you before you pay.
- Some tasks need judgment a model can't yet be trusted with. We say no to those.
- A two-week pilot proves direction, not a five-year SLA. Scale is a separate conversation.
- One anonymized pilot is one data point. We won't dress it up as a track record.
Naming the failure modes disarms skepticism better than praise does. A vendor who tells you what could go wrong is easier to trust than one who won't.