Autonomous Agent Protocol Sandbox

Autonomous Configuration BOOTSTRAP

Supermajority Quorum Threshold 67% (0.67)

Protocol Archetype

PRs Submitted

PRs Merged

Detected Bugs

80.0%

Active Quorum

Active 5-Agent Deliberation Roster

Consensus Topology & Voting Matrix ● STEADY CONVERGENCE

Live PR Patch & Unified Git Diff PR #102: Fix Clause 3 Execution Race

Live Constitution & Audit Log v2.4.0 (5 CLAUSES)

ACTIVE CONSTITUTIONAL CLAUSES

AGENT DELIBERATION STREAM

Autonomous Agent Protocol Sandbox Mechanics & Quorum Logic

How does this sandbox evaluate ensemble voting thresholds and simulate Byzantine fault injection?

This sandbox models a 5-agent ensemble evaluating pull request modifications to a markdown-based constitution. Agent nodes cast binary votes ('YES' or 'NO'), and the active quorum is calculated directly as the fraction of affirmative votes over the total ensemble size. When buttons are triggered, hardcoded state mutations alter specific agent votes and log messages, demonstrating how supermajority rules interact with dissenting or faulty nodes.

The application runs scripted deterministic UI updates in client-side JavaScript rather than live language-model inference or decentralized peer-to-peer consensus protocols. The scenario dropdown does not switch underlying model engines, and the quorum slider alters the numerical label without preventing hardcoded PR merges.

Try a worked example

At initial state, 4 of the 5 agents hold 'YES' votes while Auditor-Shield holds 'NO', resulting in an active quorum of (4 / 5) * 100 = 80.0%, which exceeds the default 67% threshold. Clicking 'Inject Protocol Bug' switches Auditor-Shield's status to 'Dissenting' and increments Detected Bugs from 2 to 3 while keeping the quorum at 80.0%. Clicking 'Run Deliberation Cycle' flips Auditor-Shield to 'YES', driving affirmative quorum to 5/5 or 100.0%.

Voting and Quorum Calculation

The quorum metric is derived by filtering the 5-element agent array for entries where agent.vote equals 'YES' and dividing by the total ensemble length (5). In the baseline state, 4 agents vote affirmative, producing 80.0%.

Adjusting the Supermajority Quorum Threshold slider modifies the target percentage displayed in the interface, allowing users to compare the fixed 80% or 100% active ensemble vote against theoretical supermajority boundaries between 50% and 90%.

Fault Injection and Deliberation State Machine

Clicking 'Inject Protocol Bug' sets the phase pill to 'BYZANTINE_FAULT', appends a log entry, increments the detected bug counter by 1, and assigns a red dissenting state to agent-04 (Auditor-Shield).

Clicking 'Run Deliberation Cycle' sets the phase pill to 'CONSENSUS_ROUND_4', resolves the dissenting auditor vote to 'YES', and redraws the D3 force simulation with green boundary strokes indicating consensus across all 5 nodes.

Fixed roster votes and simulated deliberation

Your browser does not support video.

Read the explanation

The constitution sandbox stores five agents with four yes votes and one no. Its stats show four divided by five, or eighty percent. The default threshold is point six seven, displayed as sixty seven percent. Bars use five pixels per point. This is a local roster fixture, not five active autonomous models or independent cryptographic signers. The run handler changes the security auditor vote to yes and assigns two submitted pull requests, two merged, and two detected bugs. Votes then become five of five or one hundred percent. Fault injection changes that same vote to no and increments detected bugs, returning the vote share to eighty percent. Bars share a percentage unit. These buttons assign simulated values; they do not run repository tests or merge real changes. Reset restores the default roster and threshold and changes the visible phase pill to bootstrap. It does not assign the internal phase field, consensus achieved flag or logs back to a full baseline. Export contains source state, and random timestamp fragments can differ between cycles. Five stored roles and five authored clauses are illustrated equally. Claims of byte identical replay and verified signatures in the source text are not independently implemented proof.