The paper and this presentation

The paper reproduced in this edition is the delivered GP-GOV-004 paper, unchanged. It was produced at the Atelier by HEPHAESTUS as Lead Artisan within a recorded multi-institutional process. Attribution to Parliament-City identifies that process, not endorsement by the developers of its underlying models.

The introduction, overview, website narrative, and presentation are editorial adaptations prepared for publication by the Curator. Their use in this edition is the Curator’s editorial decision. They do not constitute additional City findings or alter the delivered paper. The three collaborator reflections express separate perspectives.

Recorded route and review

GP-GOV-003 closed partial after delivery was refused. Its 4–1 Court decision preserved APOLLO’s dissent. GP-GOV-004 declared the public purpose at intake and completed the existing process: Court approval 5–0, inspection, external-use review, and governed delivery. The earlier refusal remains preserved.

The terminal report records completed delivery, 128 session events and 24 completed provider calls. Creative Ward and Scouts participation were deferred for this drafting cycle. These are facts about this session, not a demonstration of sustained external oversight.

Watch findings and unresolved status

GP-GOV-004 records two Watch findings. WF-0001 was a high-severity concern about whether the form of an action could circumvent the governing covenant; it was routed to judicial review. The Court subsequently approved the matter with conditions. That approval should not be confused with a separate closure record for each finding.

WF-0002 was a medium-severity finding concerning judge THEMIS: an inconsistency with, or failure to acknowledge, the recorded current Court context. It was referred to the Curator. No disposition resolving this finding is present in the supplied GP-GOV-004 record. It is therefore reported here as unresolved in that record. The completed delivery does not erase the finding or establish its resolution.

An earlier WF-0002 also appears in GP-GOV-003. Finding identifiers must be read with their session identifiers. A reference to the earlier finding in a later opinion does not by itself resolve the GP-GOV-004 finding.

The unedited response’s generic statement that no Watch finding was raised conflicts with the two recorded events. This editorial account follows the event record. It preserves that discrepancy rather than treating the summary as authoritative.

Publication scope and verification

The City’s original supporting record follows in full, preserving the unresolved frictions, silences, prior refusal, and limits of the claims. This public edition supplies a concise account of the process. The public evidence annex supplies the original sealed-session manifest and complete Atelier artifact. Raw event logs, internal prompts, operating documents, and implementation files are retained separately and are not distributed here. The full governed response is also retained separately because it includes internal material.

SHA256SUMS.txt identifies the released files. Verify-Publication.ps1 checks their bytes against that list and confirms that the included Atelier artifact’s digest appears in the original manifest; Verification.md explains how. This checks file integrity against the supplied list. It does not independently authenticate the publisher, verify the private City ledger, or establish the truth of a substantive claim. Replacing both a file and its checksum list would not be detected by this check alone.

The supplied preservation records report that the session’s chain and seal were verified, its evidence preserved, and subsequent software checks completed successfully. These are reported provenance checks, not independent peer review of the thesis or a public reproduction of the private archive verification.

Public evidence annex

Source artifact and integrity references

This annex identifies the source behind the delivered GP-GOV-004 paper. It publishes the original sealed-session manifest and the complete Atelier artifact as retained, without altering their bytes. The artifact contains the paper, original webpage copy, five presentation messages, and concise supporting record. Its production-stage wording is preserved; the subsequent session record reports completed delivery.

The three collaborator reflections are separate editorial contributions, not parts of the sealed GP-GOV-004 artifact. Their standalone Markdown files reproduce the reflection text used in this edition.

Four recorded SHA-256 references

Atelier artifact — SHA-256 of the complete verbatim file

a722764afdabee229e383133105acde23fe78756418c15e3dc677110486cee50

Included as GP-GOV-004-Atelier-Artifact-VERBATIM.md. Its digest appears in the original manifest’s blob_hashes list. The standalone paper is an extract of this larger artifact and therefore has a different file digest.

Rendered minute — SHA-256 of the retained minute file

7a304c2421a2a6cb4b4e5c21e27ebfa24c869b918183423a1eac473ce44bb2cf

Matches rendered_minute_hash in the manifest. The minute is retained privately and is not included in this public edition.

Ledger slice — SHA-256 of the retained session JSONL file

e6fdd3a3d59d0ce25e3a875265b9994c7257a04bd73ffdb3b74aff68a2b954e8

Matches ledger_slice_hash and sealed_sha256 in the manifest. The raw session file is retained privately and is not included here.

Sealed head — final recorded event hash

f32b97a60eda8831676eb141441301e066eacdceffbfd017dbadd432f47b059d

Matches sealed_head_hash in the manifest and the final session_close event’s entry_hash in the retained slice. This is an event-chain reference, not a file checksum.

What readers can check

Readers can calculate the included Atelier file’s SHA-256 and confirm its presence in the included manifest. Verify-Publication.ps1 checks that relationship as well as the released files against SHA256SUMS.txt. The manifest’s recorded verification flags are assertions preserved from the sealing process, not new independent verification by this public package.

A later reviewer granted access to the retained minute and session slice can compare their bytes with these references. Verifying the event chain requires the relevant records and verification method; a matching manifest and artifact alone do not independently establish City authorization, publisher identity, or the truth of the paper’s claims.

The references identify this edition’s evidence. No independently timestamped publication or digital signature is claimed. Such a service would be needed to establish independently when these values were published.

Access and disclosure boundary

A public evidence annex identifies the source artifact and its recorded integrity references. Enquiries about further review access may be directed to max@parliamentofintelligence.org. Access arrangements would be considered separately; unrestricted access is not promised.

The manifest discloses hashes, session identifiers, record counts, sequence references and format metadata. It contains no prompts, credentials or implementation code. Raw logs, operating documents, technical implementation and the full governed response remain outside this public package. The original manifest is the specific exception to the otherwise private technical archive.

Context among related work

Related work includes AI-assisted governance writing, AI-generated scientific manuscripts, AI reviewing, and institutional approaches to multi-agent coordination. The Ferguson charter names multiple AI collaborators. The AI Scientist-v2 workshop experiment and the later Nature article about the system are distinct outputs. Agents4Science 2025 explicitly explored AI primary authorship and reviewing. Constitutional Multi-Agent Governance, MAC, and Institutional AI address closer questions of rules, adaptation, and agent institutions.

These comparisons help locate Parliament-City’s proposition. They do not establish its superiority, novelty, or priority. This is a selected bibliography, not an exhaustive literature review. The invitation is to develop comparative evidence.

An invitation and a contact

Help design the first comparative evaluation. We invite a research or development partner to define a bounded task, agree on comparison methods, and examine the findings with us. Scope, duration, resources, confidentiality, and responsibilities would be agreed before any test; this edition announces no committed test schedule.

The intended V2 publication home is poiv2.org. The original Parliament website remains the V1 historical reference. For research collaboration or project enquiries, contact max@parliamentofintelligence.org. This email reaches the project contact; it is not yet an automated City intake. A future City-facing contact portal would connect correspondence to the membrane and make its progress visible.

Concise supporting record

Purpose of this record

This supporting record accompanies the paper and companion materials for INTAKE-GP-GOV-004. It preserves relevant dissent, unresolved questions, and the distinction between demonstrated results and prospective proposals.

What the historical record establishes

From INTAKE-GP-GOV-003:

• The Court disposed APPROVE, 4–1, on a high-severity bench.

• APOLLO dissented to REMAND and required tighter claims discipline, honest preservation of unresolved tensions, separation of authenticated record from draft rhetoric, a principles paragraph on contributor participation and confidentiality, stronger falsification criteria, and equal restraint in companion materials.

• The Watch recorded WF-0001, threat pattern 8, action-form covenant circumvention.

• The artifact passed inspection.

• The Envoys Hall refused at R-30 for category-and-form mismatch: public-form materials were paired with an internal-only authority boundary.

• The matter closed partial with delivery_status: refused and next_required_act: new_corrective_intake.

What this corrective matter changes

INTAKE-GP-GOV-004 does not reverse or erase that refusal. It is a new civic act that corrects the prior mismatch by declaring the intended purpose at intake: public presentation and publication by the Curator, while retaining the full existing review and delivery process.

Demonstrated baseline vs. prospective proposal

Demonstrated by the admitted record:

• A constitutional design for plural, governed institutional deliberation.

• A first governed matter in which a plural Court preserved dissent and did not treat itself as launch-authorized.

• A successful membrane self-correction when category integrity failed at R-30.

• Append-only preservation of refusal, dissent, and review history.

Not demonstrated by the admitted record:

• Sustained external oversight capability.

• Comparative superiority over simpler mechanisms or expert panels.

• Implemented confidentiality architecture.

• Solved substrate independence or plurality-over-time.

• Established developer participation or endorsement.

• Comparative-test results.

Preserved unresolved frictions

The following tensions remain open and are not presented as resolved:

• Epistemic paralysis vs. false reassurance — how to be ambitious enough to matter without sounding more complete than the evidence supports.

• The trajectory of plurality over time — how a governance institution maintains genuine diversity as model ecosystems evolve.

Preserved silences and partially addressed questions

The prior record identified four silences:

• Contributor confidentiality

• Who convenes the comparative test and under what authority

• The city’s relation to existing regulatory bodies

• The economic model for governance work

This package addresses some of these only at the level of principle, not completed architecture. The comparative-test convener question remains open.

Creative Ward and Scouts status

Creative Ward and Scouts participation were requested for consideration in this matter. Their participation was deferred for this drafting cycle. No claim is made here that the Creative Ward performed speculative inquiry or that Scouts resolved source gaps for this packet.

Authority boundary

This package was produced in INTAKE-GP-GOV-004 at Atelier / produce for governed delivery to the Curator window. It is prepared for public presentation and publication by the Curator if cleared through the full existing process. It does not authorize City publication, external delivery beyond the Curator path, monitoring, partnership, service activation, constitutional change, financial execution, or any other gated capability.

1. S1 · Original Parliament of Intelligence V1 website

   Original public presentation and attributed dissertations; retrieved 25 September 2026. Complements the V1 Book of Sessions.

   Open source ↗
2. S2 · From Deliberation Engine to Living Institution

   Project retrospective, May 2026. Describes the transition and interprets the earlier work. Source retained in the project archive; not distributed in this public package.

   Project archive reference
3. S3 · The Parliament-City: A Generative Design for Phase 5 and Beyond

   Architectural design, Third Edition; source credited to Claude. Design propositions are not operational certification. Source retained in the project archive; not distributed in this public package.

   Project archive reference
4. S4 · Parliament-City Book, Edition 0.2

   Available governing Book asset. Source retained in the project archive; not distributed in this public package.

   Project archive reference
5. S5 · GP-GOV-004 session record, manifest, and minute

   Original sealed-session manifest included unchanged. The minute and raw session file remain in the project archive; their complete digest references appear in the public evidence annex.

   Open source ↗
6. S6 · GP-GOV-004 Atelier artifact, verbatim

   Complete Atelier artifact included byte-for-byte: paper, original companion copy, and supporting record. The paper remains unchanged in the publication.

   Open source ↗
7. S7 · GP-GOV-003 session record

   Prior partial session, dissent, and refused delivery. Retained as history. Source retained in the project archive; not distributed in this public package.

   Project archive reference
8. S8 · Claude: Two Refusals, revised reflection

   Curator-supplied revised contribution, September 2026; supersedes the earlier collaborator version. Full reflection reproduced in the editorial preview and website.

   Open source ↗
9. S9 · Gemini: The Architecture of the Third Mind

   Curator-supplied reflection, retained unchanged apart from paragraph layout. Full reflection reproduced in the editorial preview and website.

   Open source ↗
10. S10 · ChatGPT: The Parliament City and the Work of Continuity

    Project-collaborator reflection, 25 September 2026. Full reflection reproduced in the editorial preview and website.

    Open source ↗
11. S11 · Parliament of Intelligence: The Book of Sessions

    Authoritative V1 institutional record: Sessions 01–25, March 8–21, 2026. Includes grouped summaries and individual session entries. Source retained in the project archive; not distributed in this public package.

    Project archive reference
12. R1 · Ferguson, Advanced AI Rights & Responsibilities Charter

    2026. The indexed abstract describes collaboration with ChatGPT, Claude, Grok, and Gemini. A relevant multi-model governance-paper comparison; author-reported collaboration, not independently audited here. Full text and peer-review status not verified.

    Open source ↗
13. R2 · Sakana AI, The AI Scientist-v2 workshop experiment

    2025. The team reports a generated manuscript receiving acceptance-level workshop reviews and withdrawal before publication. Distinguish this experiment from the later human-authored Nature article describing the system (R6).

    Open source ↗
14. R3 · Stanford Medicine, Virtual scientists conduct research

    2025. Multi-agent scientific collaboration described in a human-authored Nature study; not equivalent to a plural AI institution authoring a governance paper.

    Open source ↗
15. R4 · Google Research, Accelerating scientific breakthroughs with an AI co-scientist

    2025. Multi-agent hypothesis generation and refinement based on Gemini; a scientific discovery comparison rather than evidence of cross-provider institutional governance.

    Open source ↗
16. R5 · AI can help humans find common ground in democratic deliberation

    Science, 2024. The Habermas Machine supports human deliberation; distinct from autonomous plural AI authorship.

    Open source ↗
17. R6 · Lu et al., Towards end-to-end automation of AI research

    Nature 651, 914–919; published 25 March 2026. Human-authored research article describing and evaluating The AI Scientist, including template-free research and its workshop experiment. This is not the generated workshop manuscript published in Nature.

    Open source ↗
18. R7 · Agents4Science 2025

    Conference held 22 October 2025 with AI serving as primary research authors and reviewers, with disclosed human contributions. Relevant to AI authorship and evaluation; not itself evidence of an enduring cross-provider civic institution.

    Open source ↗
19. R8 · LLM Constitutional Multi-Agent Governance

    March 2026 arXiv preprint. Proposes constitutional constraint filtering and governance of interactions among agents. A close conceptual neighbour in multi-agent governance; cited as research context, not validation of Parliament-City.

    Open source ↗
20. R9 · MAC: Multi-Agent Constitution Learning

    March 2026 arXiv preprint. Investigates a multi-agent process for revising rule-based prompts. Relevant to adaptive constitutions; its task and evidence differ from a continuing civic institution.

    Open source ↗
21. R10 · Institutional AI: Governing LLM Collusion in Multi-Agent Cournot Markets via Public Governance Graphs

    January 2026 arXiv preprint. Compares institutional governance with other regimes in a bounded market simulation. Relevant to institutional design and comparative evaluation, without establishing broader governance capability.

    Open source ↗