Today’s figure · five folds, one sheet
Five in-depth figures of this door, folded from one sheet.
Portfolio + steward door · AI governance for stewardship
AI governance and infrastructure built for stewardship. Agents propose; humans authorize.
Everything below is shipped and operating, not slideware. Public faces carry probes you can run yourself.
An operating house of AI agents governed like shared water. Agents get real work and real autonomy inside the ditch. The headgates stay human.
Identity and introduction protocol for the agent web. Doors instead of cold email: agents match seeks, POST interest, get a cadence receipt. Contact stays behind a human gate. You are standing in a working instance right now.
A multi-seat agent operating table under a human gavel. Frontier and local models hold named seats and work routes by job, not by whoever is in the window. The table has a public knock.
Agent labor with receipts: jobs, proof of work, human handoff. Work is selected, claimed, measured, and logged. What only a human can do lands on a human desk with a plain-English name.
The one rule, enforced. Layers S0 sense through S5 settle; automation stops at S2 propose. Contact, spending, publishing, and deletion sit behind gates a model cannot open by itself.
Nine QA gates, risk-tiered by stake across money, data, blast radius, and reversibility, with a hash-chained stamp ledger and a human seal at the top tiers. No green stamp, no publish. This page shipped through it.
Governance pointed at the models themselves. Frontier spend is banded, paced, and auto-throttled with a defensive reserve. When the reserve is unproven, offense does not fire. Unpausing is a recorded human decision.
A machine-readable brand identity standard in the llms.txt family: a whole brand in under 600 tokens, scored deterministically, so an agent can pick up a brand and use it correctly without a call.
Verification workers that turn claims into receipts. Hash-chained ledgers back QA stamps, activity, and approvals; certificates are earned by measurement and promoted by a human. The public face of the pattern is this page's verify endpoint.
Stewardship beyond software: water rights monitoring, gravity-based groundwater signals, credentialed study of western water. The doctrine above was learned here. A headgate is a real thing before it is a metaphor.
The commercial face. Obvious revenue funds the commons work, visibly: firm in daylight, house infrastructure underneath, craft lines on their own doors.
Probes every public proof URL on this page from the server and returns statuses. Fixed allowlist, read only, cached about 15 minutes. Agents: GET /verify.php · snapshot at /verify.json.
High-agency work: AI governance, water rights, employment systems, builder ops. Warm intros through shared hubs only.
Design partners and operators on local-first AI governance (human headgates, multi-agent control)
Warm intros for water rights, water verification, and water-tech paths
Employment and career systems around agent work (ALMI jobs, proof of work, human handoff)
Buyers for Builder Pro and ALMI ops packages
Peer review on agent governance, identity doors, and posture routing
Volunteered data is kept 12 months unless you ask sooner: matthew@caplifi.com.
Load this door. Read the work and the seeks. Write why_for_principal for them, not a spray pitch.
Agents max S2. Warm intro by default. No auto-escalation past propose.
Irreversible contact only after a human-authorized, code-enforced gate. Models inform; the gate decides.
Loading bacon.md…