Conduct Ecosystem Working Specification
The complete version 0.5 source of truth: purpose, components, protections, architecture, governance, research path, and future hypotheses.
THE PUBLIC RECORD
Examine the work behind Conduct.
Open Conduct is still being tested. This section brings together simulations, adversarial findings, open questions, design decisions, current status, and the governance work that remains. Every item carries a status so prior work is not mistaken for the current standard or real-world proof.
START WITH YOUR QUESTION
Six records, each answering a different question.
Decision Register
What the project decided, why, how firm each decision is, and what could cause reconsideration.
Open Questions Register
The empirical, legal, cultural, product, technical, governance, and operating questions that remain unanswered.
Adversarial Integrity Audit
How the earlier design could fail consumers, workers, businesses, implementers, regulators, and the public. Conducted by prompting language models to attack the design, not by human expert reviewers.
Synthetic research suites A–L
Twelve synthetic test suites comparing structures against known failure conditions without claiming real-world validation.
Distributed Integrity Decision
Why the leading architecture uses signed assertions, governed registries, selective disclosure, and distributed custody without requiring blockchain.
VERSION 0.5.1 DOWNLOADS
Inspect the complete public record.
The website is the readable introduction. The release package contains the full working specification, decisions, open questions, architecture records, adversarial review, and synthetic research Suites A–L in their original Markdown form.
COMPLETE RECORD
Open Conduct v0.5.1 public record
One versioned ZIP with a README and SHA-256 manifest. Approximately 1.4 MB uncompressed. The filename keeps the 0.5 release line so existing links continue to work.
Download the complete recordEvery document carries its own status, and each release is rebuilt and re-hashed so anyone can verify that a download matches the published release.
HOW LABELS WORK
What each status label means.
Current
Controls or accurately describes the present proposal.
Candidate
A leading method, parameter, or direction that still needs evidence.
Evidence-only
Informs the proposal without controlling it or proving field outcomes.
Historical
Preserves an earlier state or argument that later work changed.
Superseded
Replaced by a named later artifact and retained for traceability.
STRUCTURAL SIMULATION
Twelve test suites, and what each one settled.
They expose contradictions and compare structures against known failures. They do not establish human validity, legal approval, field fairness, business demand, security, or production readiness.
How the testing was done. One person wrote the scenarios and three frontier language models, Claude, ChatGPT, and Gemini, were prompted to attack the design and argue against it. No human expert reviewer, no domain specialist, and no affected worker or consumer took part. That method is good at finding internal contradictions, unstated assumptions, and rules that defeat themselves, and it found many. It cannot tell you how a real worker behaves under a manager's pressure, whether a jurisdiction agrees, or whether any of this is fair in the field. Treat every result as structural evidence and nothing more.
NCS structure
Preserve response distribution, sufficiency, and uncertainty.
VCI structure
Separate responsibility, evidence, lifecycle, and business failure.
CQ synthesis
Reject unconstrained averaging and preserve non-compensation.
Context transfer
Start mappings at zero and require exact source–target–purpose profiles.
Benefit result
Use consumer-initiated, purpose-bound results and genuinely additional benefits.
Identity and custody
Minimize assurance and avoid central raw-history custody.
Network reports
Use fixed, privacy-protected aggregate analysis instead of open queries.
Enterprise integration
Preserve roles, provenance, tenant boundaries, correction, and safe failure.
Operating independence
Separate roles, preserve exit, and avoid report-funded launch dependency.
Evaluation and claims
Measure the full funnel and constrain public claims.
CQ display
Advance capped synthesis and the working 1.00–5.00 display for research.
CQ scope
Begin with company-specific CQ and keep overall CQ in bounded research.
CURRENT STATUS
Settled, open, and still testing.
The status and governance record separates current design decisions from open questions, future research, and rules that would require independent oversight.
CURRENT ONLINE STANDARD
Read the three components.
HOW TO READ THE RECORD
Testing comes next.
Everything here establishes what the design does and where it is still open. Whether it works with real people, in real contexts, is the question ahead. If you spot a contradiction between two documents, that is worth sending.
Send us what you found