01
Vendor-neutral
We don't sell an agent, so we have no reason to grade one up.
Independent referee
Know which agent work actually shipped.
SDF is the independent referee for coding agents. It verifies every agent attempt against your acceptance criteria and shows Cost per Accepted Outcome, across every agent you run.
For platform, security, or VP Engineering leaders running two or more coding agents.
§ 01 — The problem
“Task complete” is a claim, not a result.
Copilot, Cursor, Devin, and Factory each grade their own work. You pay per seat or per token, while the question that matters — did it work? — goes unanswered.
- CopilotGrades its own work
- CursorGrades its own work
- DevinGrades its own work
- FactoryGrades its own work
§ 02 — How it works
How it works
Your agents stay in place. SDF checks the work and keeps the record.
- 01
Declare
Declare the objective and acceptance criteria, linked to the business goal.
- 02
Route
Route the task to any agent you already use. Bring your own subscriptions or keys.
- 03
Verify
SDF runs independent checks the agent can't forge. PASS is tied to the exact code revision.
- 04
Record
Every attempt, cost, and piece of evidence lands in an append-only Decision Graph you can audit.
- ObjectiveAcceptance criteria, linked to the business goal
- AttemptRouted to an agent you already use
- EvidenceChecks the agent cannot forge
- VerdictBound to the exact code revision
- CostRecorded on the same attempt
The record is hardened, evidence-gated verification in a Decision Graph that links objectives to attempts, cost, and evidence.
§ 03 — Three commitments
Vendor-neutral, verifiable, Cost per Accepted Outcome.
02
Independently verifiable
Evidence is re-runnable and revision-bound. Security and finance can check it without trusting us or the agent.
03
Cost per Accepted Outcome
Compare agents on what you actually got: Cost per Accepted Outcome, not seats or tokens.
verified Attempt. A verified Attempt is a recorded run whose evidence was checked independently of the agent that produced it, against criteria that run did not get to choose.
Cost per Accepted Outcome is what you spend on work whose result a named principal actually accepted, not what you spend per seat, per session, or per agent claiming a task is done.
§ 04 — Pricing shape
Priced per workspace and per verified outcome, never per seat.
Dollar amounts are intentionally unpublished until design partners validate them.
§ 05 — Design partners
Become a design partner
We're onboarding 3 to 5 platform teams already running two or more agents. Measure Cost per Accepted Outcome before and after SDF. Apply to be a design partner.