Short, dated takes on what I am shipping and how I think about evidence-readiness for AI agents. Newest first. Full write-ups live in Projects / proof — this is the running log.
Released v1.3.0 with a smaller first-run contract: pass claims and supplied sources, then check source references, declared quotes, lexical support, and unmatched numbers before human review. It reports packet completeness, not factual truth. The older strategic-intelligence, MCP, HTTP, A2A, and vertical-worker surfaces remain available for compatibility. See the updated case study.
Published the first public worked sample for Corridor Bankability Analyst. The current version reviews five primary records, links claims to exact sources, exposes a financing-record conflict and names what remains Unknown. It is an illustrative sample, not external validation; the next proof is a blind comparison on a real external case. See the sample.
Published an ai-catalog.json following Agentic Resource Discovery (Google + Linux Foundation, Apache-2.0) — a machine-readable menu so agents and registries can find my live A2A agents and MCP server by capability, before invoking them. Adoption is near zero today; like OKF, it is an early bet that fits a site already built for agent discovery.
Google Cloud published the Open Knowledge Format (v0.1, June 2026) — vendor-neutral markdown with YAML frontmatter that an agent reads without scraping HTML. My profile, case studies, and role snapshots are now generated into an OKF bundle at /okf/. Nothing reads OKF yet; it is an early bet that matches what this site already is — a reference implementation, not a template.
A running log of short, dated, evidence-first takes on what I am shipping and how I reason about evidence-readiness for AI agents. The long-form proof stays in the case studies; this is the in-between.
It compares a counterparty name with a dated public-list snapshot (OFAC / EU / UK) at request time. This is a supporting string-match check inside the evidence-gap workflow — not identity verification, ownership resolution, sanctions screening, clearance, or a sanctions determination. Human review is required. See the workflow and full boundaries.
Strategic-risk triage, Middle Corridor deal risk, Kazakhstan market entry, CIS secondary-sanctions, agentic-interaction trust, and Gulf maritime exposure — sibling deployments to the four open-source skill repos. Public demo / illustrative usage; full analysis runs through the installable Agenda Intelligence MD MCP package.
Human pages plus JSON endpoints, agent discovery, and an MCP server — so an LLM or agent can read the profile as structured data, not scraped prose. The content is personal; the architecture is the part worth copying. Start at llms.txt.