§ Research
The argument, the evidence, and the open door.
This is where Summit's thinking lives in public: the doctrine that motivates the work, the publications that document it, the benchmark that measures it, and the explainers that teach it. If you work on AI assurance — at an agency, a lab, a competitor — start here.
§ 01
Start with the argument
The position
Doctrine→
Why admissibility — not capability — is the deployment standard for machine-generated decisions. The long-form statement of what Summit believes and why.
The record
Publications & Briefs→
Every published artifact in one place: the four standards, policy submissions, and the capability statement. More briefs forthcoming.
The measurement
Benchmark Leaderboard→
AGD-Bench results: Verified Governed Delivery Scores, GA thresholds, and the public state of governed agentic delivery.
The primer
Decision Provenance 101→
Receipts, admissible evidence, replay, policy-as-code, and the questions to ask any vendor — in plain language, no selling.
§ 02
The four standards
All open, all versioned, all CC BY 4.0. The research program and the standards program are the same program.
v1.0
Decision Receipt Spec→
Structure, deterministic evidence IDs, validation rules, trust model.
v1.0 · CAL-1 → CAL-7
Cognitive Security Framework→
Seven functions, seven assurance levels, a maturity model, procurement language.
v0.1
AGD-Bench Methodology→
VGDS scoring, eight task categories, GA thresholds for governed delivery.
v1.0
Decision Failure Atlas→
Documented incidents, each mapped to the control that prevents it.
Standards work is open work.
The specifications, methodology, and submissions are maintained in the open. Read them, implement them, file issues against them.