This profile is not a portfolio of promises; it is a log of evidence.

Automated changes can each pass the same policy check and still violate it when merged cleanly. This project makes agent proposals reviewable through typed checks, linked evidence, and recorded supervisor exceptions that preserve prior verdicts.

• Run the walkthrough: null-hype.tidelands.dev (interactive budget & authority lessons)
• 60–120s Demo & Narrative: Walkthrough guide
• Claim → Evidence Trail: claims and runnable test matrix
• Source Code & Checks: null-hype/agent-plugins
• UI Components & Storybook: Storybook preview

• Pre-Merge Semantic Invariants: Detecting multi-agent capability and authorization drift before PR acceptance.
• Supervisor Protocols & Scoped Exceptions: Recording supervisor overrides without discarding or mutating the original violation record.
• Evidence Reconciliation: Static and runtime verification across declared agent intent, supervisor grants, and observed tool execution.

Status: Research prototype evaluated on synthetic fixtures and CI-exported runs. Not a production control.

• Public Collaboration: Open an issue to discuss open-source setups, reproduction, or test cases (public; do not include private credentials).
• Private Enquiries (Consulting, Research Collaboration, Advisory, Funding): Email public.rant@pm.me for confidential discussions regarding production agent architectures. (Please describe workflow context; do not send secrets or API keys in initial outreach).