Earning the Right to Propagate

How a human Village governs its own AI

John G. Stroh · My Digital Sovereignty Ltd. · 12 slides with presenter notes

Earning the Right to Propagate · June 2026 · MDSL

A constitution arrived in June

  • Vincent Boucher — president of QUEBEC.AI and MONTREAL.AI — published GoalOS: The Proof-of-Evolution Constitution (AEP-001, v12.1, June 2026)
  • By his own framing: a doctrine and a falsifiable research program, not a shipped product — its self-references are manuscript drafts
  • It governs how an institution's AI is allowed to improve
  • We govern how human communities hold AI — opposite ends, same primitives. The convergence is the story.
Earning the Right to Propagate · June 2026 · MDSL

Boucher's rule

  • "Do not put intelligence on-chain. Put proof of intelligence on-chain."
  • Keep the machine's reasoning and private data private; make public only the proof that work was done, tested, and allowed to influence what comes next
  • Public doctrine: Aim → Act → Prove → Evolve
  • Each step becomes a signed, versioned object
Earning the Right to Propagate · June 2026 · MDSL

The Proof Gradient — a selection law, not a score

  • "The score is advisory; the gates are mandatory" — a high score never buys past a failed gate
  • Promote only if: proof valid · beats baseline on eval · risk under threshold · rollback drilled · canary scoped · scope authorised · challenge window cleared
  • "No proof, no evolution. No eval, no propagation. No rollback, no release."
  • A proof page "is not a marketing page; it is a claim-bound evidence room"
Earning the Right to Propagate · June 2026 · MDSL

We'd already built it — for people

  • Public-proof / private-data boundary → did:web identities + append-only proof chains; tenant content encrypted
  • Proof-Carrying Artifact → signed sovereign records (provenance hash over origin fields, tamper-evident proof chain, exportable)
  • SelectionCertificate → DirectorEngagement decision record, maker-checker, proposer ≠ approver
  • Evidence Docket → our Decision Records: verify-before-claiming, no defensive-hedge, minutes you can't quietly edit
Earning the Right to Propagate · June 2026 · MDSL

The gap it closed for us

  • Our own AI evolution was empirical but ad-hoc — modest per-community models and the lighter layers around them
  • Humbling record: a retrain tested and not deployed; a run of weight experiments degraded; "don't train aspirationally"
  • We had the evidence habit — not a named law for when a change earns the right to propagate
  • Now adopted: candidate → canary → active (always revocable), through Boucher's gates
Earning the Right to Propagate · June 2026 · MDSL

The one thing we refuse: the chain

  • AEP-001 is blockchain-native: an Ethereum Evolution Ledger, staking, slashing, a reward vault, a token economy
  • We adopt none of it — a public chain and US-anchored crypto infrastructure violate the Village's reason to exist (sovereign EU/NZ, self-hosted, no foreign substrate)
  • His own neutrality clause lets us out: any institution may implement AEP "if it respects the proof interfaces"
  • We respect the interfaces — on sovereign ground. Named human accountability where his design has staking and slashing.
Earning the Right to Propagate · June 2026 · MDSL

The gate the constitution is missing

  • His subject is AI capability: "institutions whose agents may act, but whose improvements must prove themselves before they propagate"
  • A capability gate can promote a superior upgrade that quietly moves authorship away from the people
  • We add ServeAuthorized: does the change keep the community the author? Refused if not — even if it scores higher
  • And consent + cultural fit: touching te ao Māori data governance needs the consent of those it affects, or it does not ship
Earning the Right to Propagate · June 2026 · MDSL

Governance capture: removed, not mitigated

  • His threat model lists "governance capture: token votes or insiders bypass evidence" — mitigated with governance separation
  • But the token-vote substrate is the attack surface
  • We remove it: promotion authority sits with named human guardians under maker-checker — accountable as persons, not as wallets
  • There is no stake to weight and no token to capture
Earning the Right to Propagate · June 2026 · MDSL

Keep AI the servant, not the subject

  • The language of "institutional AI-evolution" has gravity — it pulls toward an AI-ops product with communities attached
  • Hold the opposite line every time the discipline is invoked
  • The proof gradient exists so a community can trust the tool it holds — not so the tool becomes the point
  • If applying it ever makes the AI the subject, it is being misused — stop
Earning the Right to Propagate · June 2026 · MDSL

Postscript: from constitution to product

  • Boucher's product layer — GoalOS Mission OS (June 2026): "set the objective; GoalOS runs until proof is done"
  • Largely substrate-neutral — the reward vault / slashing court / staking are gone; his own product travels without the chain
  • Its "autonomous publication law" (generate → QA → human-reviewed PR → publish) names the discipline the Village already runs; "proof debt" names our hazard
  • Built for autonomous AI work that runs "until DONE" — exactly why we keep the human as the author, not the reviewer
Earning the Right to Propagate · June 2026 · MDSL

A community must remain the author

  • A model can answer. An agent can act. An institution must prove. And a community must remain the author.
  • Take the rigour gratefully; refuse the substrate; keep the gate the constitution leaves out
  • Read the full essay: agenticgovernance.digital/papers/earning-the-right-to-propagate.html
Sources: Vincent Boucher, GoalOS: The Proof-of-Evolution Constitution — AEP-001, v12.1, and GoalOS Mission OS: The Proof OS for Autonomous AI Work, v23, June 2026 (MONTREAL.AI & QUEBEC.AI). Lineage: G. C. Necula, "Proof-Carrying Code," POPL 1997.
Earning the Right to Propagate · June 2026 · MDSL

A briefing companion to the essay. We read a new constitution for governing how institutional AI is allowed to evolve — Vincent Boucher's GoalOS: The Proof-of-Evolution Constitution — and found it describes most of what we had already built, for human communities. This deck: what we built, why the convergence validates it, the one thing we refuse, and the one gate the constitution is missing.

The credibility comes from a serious independent researcher arriving at the same architecture from the machine's side of the problem. He does not overclaim — which mirrors our own claim discipline.

The one-line rule is the spine. The four-step lifecycle is how institutional intent becomes governed evolution. The quote is verbatim.

This gate is the part worth adopting. The key discipline is gates-over-score: capability never overrides the hard conditions.

The convergence table. Be precise: this machinery is shipped, for human governance. That precision is itself the claim discipline both documents insist on.

Be honest that this is a newly-adopted discipline, not a system already running in production. The not-deployed-retrain record is the credibility — we earned the humility.

The conflict is exactly one layer. We take the doctrine and refuse the substrate — and we are entitled to by the document's own words.

This is the seminal turn. Proof-of-evolution is necessary; it is not sufficient. Authorship is not a metric.

A concrete win from making people the subject rather than agents: a whole class of attack simply does not exist in our design.

The permanent guard. This is the non-negotiable that must stay front and centre, every time.

The follow-on paper strengthens the convergence and, by being a product for autonomous AI work, sharpens our one divergence. Generous to the work; we hold the line — AI is the servant, not the subject.

Close on the line that captures the whole argument. The gratitude is real — the convergence validates the shape — and so is the divergence.