Luxailux.ca

The Papers Shelf

White papers · Lux Automations

Our work, written down and put in the open

This house publishes its papers the way it publishes its receipts: in the open, at the stage the work has honestly reached, with the limits printed in the same ink as the claims. Every paper below is a PDF. Read them, check them, argue with them — that is what they are for.

Paper No. 1 · Research report

Who Do You Believe When the Worker Is an Agent?

A Tamper-Evident Audit Layer for Agent Payments · Vance Cyr — Lux Automations · Filed with the Apart Research “AI x Epistemics” Sprint, Track 2: Trust Infrastructure · November 2026

Autonomous agents now negotiate, deliver digital work, and settle payments in seconds, but the evidence those transactions rest on has not kept pace: a log written by the party it describes can be rewritten by that party. This report presents a tamper-evident audit layer for agent payments, built by extending an escrow pattern executed end-to-end on a public test network — signed, hash-anchored deal terms; hash-chained event logs; content fingerprints bound to proof-of-delivery hashes; and a public proof shelf of verifiable receipts — together with three mechanisms adopted from an external design review that close the layer's sharpest holes: transparency-log anchoring, independent witness cosigning, and decision-context commitments. The evidence is limited, and stated as such — including a real design flaw our own audit found. The layer proves continuity and provenance of the record, not the truth of what was fed into it.

Paper No. 2 · Technical brief

The Security Patch — Technical Brief

A checked identity for agents, and payments that carry their own proof · Lux Automations · November 2026

The Security Patch pairs a checked identity for agents with payments that carry their own proof. This brief sets out, at behaviour level only, what each half does — issuance, door checks, behaviour watch, revocation, and the tab on the identity side; mandate-bound escrow, no custody, and a tamper-evident audit layer on the money side — and then shows the evidence that exists today: public testnet receipts on two networks, including a first escrow-cancel attempt that failed and was corrected in the open, and the in-house drill record with its limits stated beside it. It closes with what has not run yet, in a plain list. The pair is at prototype stage; it has not run live, and the independent audit has not yet been performed.

Paper No. 3 · Model brief

School in a Box — The Model in the Open

A whole school on one small computer · Lux Automations · November 2026

Some schools have no internet, no library, and no way to reach either. School in a Box puts a whole school on one small computer — an AI teacher who works with no internet at all — and carries it to the places the wires never reached. This brief puts the model in the open: how a box reaches a school (need decides, local hands deliver, every name recorded on the Wall of Heroes), how sponsorship works (a gift of monthly upkeep, never a sale — and no figure printed, because none has been set), the safeguards that are structural rather than promised (parents are verified, not kids; a child's profile rides under a parent's account, and the parent approves and pays), and the honest limits — one classroom's reach, the first boxes still in proof, no boxes placed yet, and first placements promised only from the first quarter of 2027.

How to read our papers

Receipts where they exist; our records, labelled, where they don't

Everything we claim comes with the best evidence that exists for it. Public receipts are linked so a stranger can check them. In-house records are labelled as ours — real, and watched by nobody but us, and we say so. Where a thing has not been done yet, the paper says that too, in the same size of type. If you find a claim in any of these papers that the evidence doesn't carry, we want to know: the door is [email protected].