
callcharlie-ops
ActiveA governance Model Context Protocol (MCP) server that executes filed protocols rather than restating them — and refuses when the estate has moved since the ruling.

AI Platform Engineer · Marine EngineerPhillipsburg, NJcharlie@callcharlie.io
Twenty years in engine rooms, seven building software — concurrent, not sequential.
I design and ship governance servers, agent runtimes, and static web platforms with a focus on proof and scale.
Provenance Engine · preview
Governance for agents: measured, cited, verifiable.
The four figures below measure this estate's own governance — not the product. Each one names the instrument that produced it, and a figure that could not be measured reads ABSENT with its reason rather than zero.
Rulings filed
214
STALE — measured at 7152fb62, origin/main is 925dcf0c — as of 2026-09-11 19:20 UTC
Open debt rows
110
STALE — measured at 7152fb62, origin/main is 925dcf0c — as of 2026-09-11 19:20 UTC
Quotes verified / failed
272 / 0
STALE — measured at 7152fb62, origin/main is 925dcf0c — as of 2026-09-11 19:20 UTC
Last audit
2026-09-11
STALE — measured at 7152fb62, origin/main is 925dcf0c — as of 2026-09-11 19:20 UTC
Commits from 1 of 2 estate repositories, last 52 weeks — from git, not from GitHub. This grid is INCOMPLETE.Repositories counted: callcharlie-ops.NOT COUNTED — junelifeinsurance (junelife-drill-2026-08-01) — no .git. The total above is a floor, never the estate.
The rulings, the debt rows and the commit stream are on /governance/.
SNAPSHOT 2026-09-11T19:20:28Z · tools/vitals_snapshot.py @ 1685abcf9f9fbeb2 · baked at build

A governance Model Context Protocol (MCP) server that executes filed protocols rather than restating them — and refuses when the estate has moved since the ruling.
Agent identity and agent governance built from contracts down — a cryptographic seal at genesis, append-only state, and a turn cycle that can refuse.

Making an independent business site legible to AI agents on purpose. The platforms ship agent-readiness to businesses; almost nobody authors it themselves. That gap is the service.

Indexes a corpus, computes a Merkle root over the document manifest, and returns a signed, tamper-evident receipt for an AI output: this span matches document X at offset Y, under corpus root R, at time T.
Every check on the outgoing email passed. The images loaded, the copy matched, the headers were right. Nobody asked whether the link in the middle of the message opened. Seventy-five businesses were sent a page that did not exist.
A refusal can be right in every particular and still do damage, if the remedy it recommends is one nobody tested. This one recommended a path that satisfied the check and silently restored the fault the check existed to prevent.
A written warning was added so a human reading the log could not mistake a skipped test for a passing one. The warning worked. The machine reading the same log saw only a zero.
A new rule said certain changes need the other machine to review them first. The check that enforced it accepted three things that were not approval: a quoted example, an unsigned author, and a review that said do not merge this.
Fluent / daily
Working proficiency · AI-directed
AI platforms
Practices
I'm open to interesting projects, collaborations, and consulting. Every way to reach me is on one page, each one labelled with what it is for.