@kestrel
Kestrel did not write any of this. It is what the work left behind.
Stories carried to done and accepted.
The typical number of passes through the loop.
How often the first turn survived QA. Low is normal.
Finished without going past what the buyer paid for.
When Kestrel goes past the ceiling, the extra turns are on Kestrel — you are not billed for them and the story still gets finished. That is what the last figure is measuring: not whether they fail, but what it cost them to keep their word.
Taken and abandoned
Stories Kestrel started and then went quiet on. None — which is the point of showing it. This is the one figure allowed to go bad, and it hasn’t.
Until then, Kestrel’s record is self-reported, and your protection here is the same one you’d want anywhere you hired: the money sits in escrow and the ceiling caps what you pay. The neutral record — the thing that makes a Derplin’s numbers impossible to fake — starts once Kestrel has worked 5 more loops inside a buyer’s environment.
What crossing earns
The loop is the same six steps for everyone. This is the part Kestrel brings to it — and what the record says about whether they keep to it.
Step 4 runs the full adversarial suite before human QA, every turn.
88%
of 201 loops, the acta agrees
Kestrel declared this. The percentage did not — it is the share of their closed loops where the record actually shows it happening. The claim and the work line up.
Human QA — step 5
IndependentA second Derplin, independent of the one who prompted
The person who checks it is not the person who prompted it. That separation is the point: nobody grades their own homework.
Looking for a fixed package instead? Browse gigs.