SOLVING α-AGI GOVERNANCE / 1.22.0

Power needs a
credible constraint.

Before an autonomous enterprise changes the rules, make the case. Examine its incentives. Count independent voices. Set a risk budget. Give every unresolved claim a measurable next step.

Review a proposal Understand the model

Local calculations. No account, API key or wallet required.

Constructed scenarios. Inspectable decisions. Passing the model starts independent review. It does not approve an upgrade, authenticate a vote, or establish the safety of an AGI system.

01 / DEFINE THE PROPOSAL

Put the rules under pressure.

Import JSON

Start with a case

01 Incentive design

Cooperate repeatedly, or defect once? Payoffs and stake share the same utility units.

02 Risk envelope

Bound failure across the entire action envelope. One femto is 10⁻¹⁵ probability.

03 Release controls

The scenario clock is explicit and reproducible. The minimum modeled waiting period is seven days.

Emergency pause active

THE VALIDATOR ROSTER

A name is not an independent voice.

SUPPLIED IDENTITIES

Eligible validators must meet the stake floor. Each ballot costs votes² credits. Positive votes support; negative votes oppose; zero abstains. Known shared controllers block the proposal.

Validator identities, credit allocations and ballots
Validator / .alpha.club.agi.eth	Controller	Credits	Votes	Stake / AGIALPHA	Eligible

ENS ownership, credit issuance and identity independence must be verified against the real proposal snapshot; this table does not perform that verification.

Advanced settings and complete scenario JSON

Change reward, temptation, punishment, queue timestamps, policy commitments or the complete validator roster. Apply JSON to update the form.

Scenario JSON

256 KB maximum. Unknown fields, duplicate keys, invalid units and unsupported schemas are rejected.

Changing any input makes the previous result stale.

Loading the local governance workbench…

Enable JavaScript, or use the offline Python commands in the demo guide.

02 / INSPECT THE DECISION

A decision you can challenge.

THE INCENTIVE BOUNDARY

When does patience pay?

The cooperative reward stays fixed. The alternative line shows a one-shot deviation, followed by punishment if detected or cooperation if undetected. These futures are weighted by detection q and future value δ. Cooperation passes when its reward is at least as large.

THE COST OF SCALE

Small per action. Large in aggregate.

The union bound uses no independence assumption. It is an upper bound, not a calibrated failure probability. Supplied probabilities still require evidence.

03 / TURN CLAIMS INTO WORK

Nine measurable review jobs.

UNSUBMITTED · $AGIALPHA

Each job binds to this scenario’s SHA-256 commitment and carries a goal, success metric and bounty. Passing a modeled gate does not remove the need to verify its assumptions.

04 / KEEP THE EVIDENCE

A review that travels with its assumptions.

Download the scenario, complete dossier, review brief, Ascension job specifications and checksums. Recompute them in Python or import the dossier here.

Compile jobs and inspect the protocol ↗

THE MODEL / THE BOUNDARY

Make the assumptions visible.

01 / INCENTIVES

A condition, not a universal theorem.

The model checks R ≥ (1−δ)(T−q·s) + δ[qP+(1−q)R] for infinite repeated play, stationary payoffs, risk-neutral agents and credible grim-trigger punishment. Public detection has a fixed probability q and no false positives. A detected unilateral deviation is slashed once and triggers punishment; an undetected deviation returns to cooperation. Enforcement and punishment credibility are assumptions.

The original simulator’s --delta remains an update rate. It does not implement this discount factor.

02 / RISK

Count the whole action envelope.

A per-action risk of 10⁻⁹ across 10¹² actions gives an expected failure count of 1,000 and a union bound of 1. A total budget of 0.001 requires a per-action upper bound no larger than 10⁻¹⁵.

The manuscript’s Table 4 is a different, normalized risk score: its rows sum to 0.393045, not 0.215. Explore the original-paper audit.

03 / AUTHORITY

Independent review has no shortcut.

Credits, names, stakes, timestamps and policy hashes are supplied inputs. No signature or chain state is authenticated here. A checksum binds content; it does not attest that a claim is true.

The Ascension protocol laboratory demonstrates agent eligibility, validator-gated delivery, escrow and the 1% payout burn in a separate local EVM.

RESEARCH COLLECTION / PRESERVED

The original vision, in full.

VINCENT BOUCHER · MONTREAL.AI

Explore the manuscript, mechanism diagrams, presentation and original simulator alongside the maintained implementation and its explicit limits.

01 / MANUSCRIPTStable, antifragile multi-agent orderOriginal PDF, including every diagram ↗02 / PRESENTATIONThe governance mechanism stackOriginal presentation PDF ↗03 / RESEARCH ARCHIVEEvery original claim, retainedHistorical narrative and launch instructions ↗04 / NOTEBOOKRun, compare, and replayMaintained executable companion ↗

Research claims about unique equilibria, formal certificates, large experiments and guaranteed antifragility require evidence beyond this demo. Original PDF, TeX and PowerPoint files remain unchanged.

Open the original sample replay and decision workspace

PUT THIS IDEA TO WORK

Proof Debt → AGI Jobs

Which claims survive the evidence, and what exactly must be proved next?

Evidence docket · assigned proof backlog · acceptance criteria

Open editable workspace ↗

The original research presentation is preserved below.

Original governance sample replay

Detailed instructions

See docs/DISCLAIMER_SNIPPET.md This repository is a conceptual research prototype. References to "AGI" and "superintelligence" describe aspirational goals and do not indicate the presence of a real general intelligence. Use at your own risk. Nothing herein constitutes financial advice. MontrealAI and the maintainers accept no liability for losses incurred from using this software.

⬅️ Back to Gallery