The Grand Intellect

A constitutional architecture for collective intelligence, implemented over the AETHER semantic coordination fabric.

The Grand Intellect turns a collection of agents into a governed cognitive institution. It separates the production of possibilities, contact with reality, purposive judgment, institutional memory, execution, review, and disposal so that no agent can manufacture and certify its own success.

Intelligence is the disciplined production of alternatives, their confrontation with reality, their selection under purpose, their preservation across time, their realization as consequential artifacts, and the deliberate removal of what no longer deserves cognitive space.

TROVE–CURATA governed curation fixture chain

TROVE–CURATA is the GCL-contained data-curation programme bootstrapped in INTELLECT. Its first five retained/synthetic fixtures form a governed chain from HTML extraction through PII observation, authorized transformation, duplicate observation, and quality-signal observation. The chain is evidence-bound and deliberately does not authorize production-corpus admission, dataset-quality certification, privacy certification, or training-fitness qualification.

The fixture ladder is deliberately fail-closed: provider outputs remain observations, authority stays GCL-owned, and review defects are preserved and prospectively remediated rather than rewritten.

TC-BOOTSTRAP-CLOSE-001 is the governed T3 boundary candidate for closing this bootstrap ladder and moving substantive implementation to a dedicated GCL repository. The destination remains inactive until the INTELLECT source closure, destination acceptance, and two-sided protected readback all complete. No Fixture 006 work or corpus-admission authority is created by preparing the closure package.

Current release boundary

This repository is an executable foundation, not a completed autonomous research institution.

Implemented now:

• an event-sourced work-package model;
• eight constitutional phase gates;
• five Minder offices, ten Council offices, Executor and Human Steward offices;
• explicit office mandates and phase-specific review obligations;
• replayable reviews, decisions, contact records, memory records, and disposal records;
• an agent registry and Council dispatcher that preserve office separation;
• a deterministic in-memory fabric for tests only;
• an authoritative HTTP adapter for AETHER's append, cut, history, policy, and provenance boundary;
• JSON schemas, work-package templates, a CLI, an example campaign, and CI.

Not yet implemented:

• a production AETHER deployment bundled with this repository;
• model-provider adapters or autonomous agent prompting;
• semantic compilation of every constitutional gate into AETHER rules;
• distributed execution, scheduling, leases, or artifact sidecars owned by INTELLECT;
• a user interface;
• unattended authority over irreversible actions.

The project fails closed when an obligation is absent. It does not infer that an agent reviewed a work package merely because the agent was invoked.

Architecture in one page

Human Steward
     │
     ▼
Grand Intellect application layer
  ├── Work-package commands
  ├── Executable constitution
  ├── Minder cycle
  ├── Council review dispatch
  ├── Artifact and decision contracts
  └── Gate reports
     │
     ▼
AETHER semantic coordination fabric
  ├── Append-only datoms
  ├── Authoritative cuts and replay
  ├── Policy-aware visibility
  ├── Provenance-bearing facts
  ├── Recursive semantic closure
  └── Explainable derived tuples
     │
     ▼
Executors, repositories, experiments, proofs, datasets, and deployments

INTELLECT is a constitutional application over AETHER, not a competing semantic kernel. The Python layer may project AETHER history and provide admission checks, but production truth remains anchored to AETHER.

The three orders

The Minders

The Minders govern the lifecycle of thought:

• Possibility Minder: produces materially distinct alternatives;
• Reality Minder: demands evidence capable of changing judgment;
• Purpose Minder: selects under explicit objectives and trade-offs;
• Continuity Minder: preserves reasons, scope, provenance, and lineage;
• Capacity Minder: retires what no longer earns active attention.

The Council

The Council governs intellectual rigor through bounded offices:

• Axiomatist;
• Cartographer;
• Verifier;
• Adversary;
• Formalist;
• Steward;
• Grammarian;
• Composer;
• Amanuensis;
• Referee.

Each office discharges a distinct obligation. Review is evidence of work performed, not attendance.

The Executors

Executors build proofs, programs, experiments, datasets, reports, and deployments under an explicit specification. They may realize an approved contract but may not silently redefine it.

Work-package lifecycle

Charter
  → Generation
  → Specification
  → Realization
  → Confrontation
  → Judgment
  → Integration
  → Disposal
  → Complete
                  ↘ governed reopening → Generation

Every transition requires both substantive artifacts and approved office reviews. For example, the Confrontation gate requires a contact record whose test could have disconfirmed the claim, an uncertainty statement, and approved Reality Minder, Verifier, and Adversary reviews.

See CONSTITUTION.md and docs/WORK_PACKAGE_LIFECYCLE.md.

Constitutional document hierarchy

1. CONSTITUTION.md is the compact constitutional law.
2. Effective instruments in AMENDMENTS/ supplement it.
3. The Grand Intellect: Constitutional Commentary and Operating Doctrine explains the law but cannot change it.
4. governance/constitutional_authority_schedule.json records the effective authority binding and its exact Human Steward, Adversary, Referee, receipt, commit, and timestamp evidence.

GI-AMEND-0001 is effective. Its schedule records that GCL-GHOS was a candidate at amendment activation and delegates current standards admission and programme-adoption status to the subordinate gcl-standards current-status projection. This preserves AETHER's production semantic authority, reserves mathematical certification to MATHCERT, and treats gcl-standards as a subordinate registry and publication repository.

Quick start

Python 3.11 or later is required. The core has no runtime dependencies outside the standard library.

python -m venv .venv
. .venv/bin/activate
python -m pip install -e .
python -m unittest discover -s tests -v
python examples/union_closed_bootstrap.py
intellect init work/WP-001 WP-001 \
  --title "Test a claim" \
  --purpose "Confront a mechanism with discriminating evidence" \
  --scope "One candidate implementation and benchmark" \
  --criterion "The benchmark can reject the mechanism"
intellect gate work/WP-001/events.jsonl WP-001
intellect metrics work/WP-001/events.jsonl WP-001

The in-memory fabric is a test double. A production construction must require an authoritative AETHER fabric:

from grand_intellect import GrandIntellect
from grand_intellect.aether import AetherHttpFabric

fabric = AetherHttpFabric(
    "https://aether.example",
    bearer_token="...",
    namespace="grand-intellect",
    schema_ref={"name": "intellect_v0", "version": 0},
)

system = GrandIntellect(fabric, require_authoritative_fabric=True)

Minimal governed cycle

from grand_intellect import GrandIntellect, InMemoryFabric, Office, ReviewStatus

system = GrandIntellect(InMemoryFabric())
system.charter(
    "WP-001",
    title="Test a claim",
    purpose="Determine whether the proposed mechanism survives controlled evidence.",
    scope="One implementation and one discriminating benchmark.",
    acceptance_criteria=("The benchmark can reject the mechanism",),
)

for office in system.constitution.required_offices(system.state("WP-001").phase):
    system.submit_review(
        "WP-001",
        office=office,
        status=ReviewStatus.APPROVED,
        obligations=tuple(sorted(system.constitution.required_obligations(system.state("WP-001").phase, office))),
        evidence_refs=("CHARTER.md",),
    )

system.advance("WP-001")

AETHER integration

AETHER's current public boundary provides the exact facilities INTELLECT needs: append-only datoms, deterministic replay, policy-aware visibility, provenance, recursive derivation, and explanation. INTELLECT uses the stable HTTP boundary rather than importing or duplicating Rust kernel semantics.

The versioned event mapping is defined in aether/intellect_v0.aether. The adapter:

1. checks required server capabilities;
2. encodes each INTELLECT event as one AETHER entity;
3. appends tagged scalar datoms with provenance and policy envelopes;
4. receives an authoritative cut;
5. reconstructs work-package history through bounded pagination;
6. projects that history into the constitutional view.

See aether/README.md and docs/adr/0001-aether-is-authoritative.md.

Repository map

src/grand_intellect/     executable constitutional runtime
schemas/                  machine-readable event and artifact contracts
aether/                   AETHER schema and boundary documentation
examples/                 governed campaign exemplars
tests/                    lifecycle, agent, CLI, and adapter tests
docs/                     architecture, pedagogy, threats, roadmap, templates
.github/workflows/        required validation

Development contract

Full-ten-office advisory proceedings use the fail-closed Council matter contract documented in docs/COUNCIL_MATTERS.md. Council compilation validates deliberative evidence; it cannot exercise Human Steward authority or advance a governed transition.

Before proposing a change:

python -m compileall -q src tests examples
python -m unittest discover -s tests -v
python examples/union_closed_bootstrap.py
intellect init work/WP-001 WP-001 \
  --title "Test a claim" \
  --purpose "Confront a mechanism with discriminating evidence" \
  --scope "One candidate implementation and benchmark" \
  --criterion "The benchmark can reject the mechanism"
intellect gate work/WP-001/events.jsonl WP-001
intellect metrics work/WP-001/events.jsonl WP-001

A change that modifies office powers, phase gates, deletion authority, or the AETHER truth boundary is constitutional. It requires an ADR, explicit threat analysis, and Human Steward review.

Read next

1. CONSTITUTION.md
2. docs/CONSTITUTIONAL_COMMENTARY_AND_OPERATING_DOCTRINE.md
3. AMENDMENTS/0001-commentary-and-gcl-ghos.md
4. SPEC.md
5. docs/ARCHITECTURE.md
6. docs/PEDAGOGY.md
7. docs/ROADMAP.md
8. docs/STATUS.md