The Scratch Pad
Where concepts become SpeakQL before they become code.
What is SpeakQL?
SpeakQL is pseudo-code that communicates intent without revealing architecture. Like T-SQL that speaks before it executes. The concept is public — timestamped, attributed, open for community discussion. The implementation is ours until we are ready to ship it. Follow along. Add your thinking. Watch ideas become real.
PBOM — Prompt Bill of Materials
Every prompt registered, hashed, and verified before execution. A cryptographic provenance layer for AI interactions. PBOM could revolutionize AI security — the concept is here, the implementation is being built.
SpeakQL
PBOM.register(prompt) .hash(SHA256) .verify(signature) .execute() // only if verified .log(provenance)
Zero Trust AI
Capability constraints enforced by design, not policy. In traditional Zero Trust you verify identity. In Zero Trust AI you verify intent, capability, and provenance — at every step, for every action.
SpeakQL
ZeroTrust.agent(agent) .verify(identity) .constrain(capabilities) .require(confirmation, for: "consequential") .audit(every_action)
Recursive Validation — RAG to the 9th
Nine recursions minimum to eliminate hallucination and surface convergent signal. Based on Tesla's 3-6-9 principle. The namesake of RAG9. Bias is signal — catalog it, find the center, show integrity.
SpeakQL
RAG9.query(input) .recurse(minimum: 9) .catalog(bias) .find(center, wings) .converge(signal) .return(truth_with_context)