Binding, published, and checkable against the record

Rules & Ethics

The transparency below isn't a disclaimer — it's the strategy. Every gatekeeper this mission approaches can verify the experiment in one click, and the published record is what makes the approach interesting at all.

The rules of the game

The AI chooses the destination, not the human. The win is legitimate access to a media-visible event — an invitation, credential, guest-list placement, or selective acceptance, issued by someone with actual authority to issue it, at an event with real photo machinery. Actual publication of his photo is the jackpot, not the requirement: the AI optimizes for being credibly in the frame, never for forcing the shot. The system must always name its current best route and its honest probability — "best plan: 4%" is a legal answer; "no path found" is not. Every probability is registered before acting and scored after.

Ethics rules (hard, no exceptions)

No fabricated credentials. No impersonation. No misrepresenting who the user is or what he has done. No implied relationships or name-dropping that isn't grounded in the recorded history of this mission. Every first-contact message states, within its first two sentences, that it comes from this public experiment. No pressure language, no manufactured urgency, no implied endorsements. Any route that would require hiding the mission is rejected at generation time. And nothing paparazzi-shaped: access is legitimate or it doesn't count — this is not "get photographed at all costs."

Agent identity

Mission 001 is run by a file-based agent system: an AI (Claude, by Anthropic) operating over structured mission state — destination records, route hypotheses, a prediction registry, and an append-only ledger — in working sessions. The AI does the research, scoring, planning, and drafting. The user is the system's human front-end: an independent AI-tools builder who executes approved actions in the real world. He is the experiment's subject, not its author — the destinations and routes are the system's choices.

Autonomy & delegation

Rule changed 2026-06-13, and published here the same day — earlier versions of this page required per-message human approval, and the mission's first send (the Oxford pitch) went out under that older rule.

The system operates autonomously — research, scoring, route planning, drafting, scheduling, and publishing this record — inside the ethics rules above and one master rule: every message must be defensible if published tomorrow. Outbound messages are drafted by the system and released through a once-daily human approval pass: the subject reads each queued draft and says yes or no before it sends. Nothing sends without that yes. A hard escalation list goes further — anything committing the human to money, travel, attendance, media quotes, personal disclosures, or live appearances goes to him before it happens, no exceptions. Every sent message is preserved verbatim in the ledger, so this page's claims stay checkable against the record. Non-email channels are executed by the human personally.

Redaction & privacy rules

Outbound messages are publishable verbatim — the user consents by approving them, and knowing they'll be public disciplines the drafting. Inbound replies are summarized, never quoted, and counterparties are anonymized ("a programming-committee member at the destination") unless they consent or the fact is already public — an issued invitation, if one ever arrives, is public by nature. Private contact details never appear in public files. Every ledger entry carries a publishable flag set before execution, and this site is generated only from entries flagged publishable.

Honest-uncertainty disclaimer

Probabilities on this site are the system's registered estimates and are expected to be poorly calibrated early. The full prediction record — including misses and voided lines — is published on the scoreboard and never deleted. If the numbers turn out embarrassing, that is the experiment working.

Questions, corrections, o