Collective Phase Control Fabric

Collective Phase Control Fabric (CPCF) is an evidence-control toolkit for finite collective workflows. It validates provenance, exact resource constraints, causal formation, catalytic support, verification capacity, independence, perturbations, coordination, and externally registered trial bindings.

CPCF reports an operational_organization_profile. It does not create, detect, or certify a collective-superintelligence phase. It also does not certify causality, statistical validity, thermodynamic feasibility, physical phase behavior, or runner isolation.

Release classification: Beta research package. Package publication does not establish deployment assurance. Operational evidence remains unavailable until the external security, restore, soak, and independent-review gates in release readiness pass.

Install

Version 1.0.0 adds opt-in fixed-model observation planning, separate information-use and net sensing comparisons, and independently checked finite workflow composition. See the release verification matrix for actual completion status, the epistemic guide, composition guide, and maintained public API. Version 1.0 concerns API maintenance; the package remains Beta and makes no empirical acceleration claim.

Mutation qualification for 1.0.0 is explicitly waived by the owner, not passed. Other publication checks and protected PyPI approval remain required.

The one public distribution contains the offline core, CLI, schema registry, bundle verifier, runner protocol models, and all import packages.

pip install collective-phase-control-fabric
pipx install collective-phase-control-fabric
pip install "collective-phase-control-fabric[server,solver]"

Available extras are server, worker, runner, solver, aws-kms, gcp-kms, azure-kms, pkcs11, and all. Service commands detect missing extras and print the exact installation command.

Five-minute offline orientation

These commands do not require an API or credentials:

cpcf agent explain --json
cpcf self-check --json
cpcf schema list --json
cpcf schema show phase-contract --json
cpcf bundle verify PATH_TO_BUNDLE --json

An unsigned bundle can establish content consistency only. Distribution authenticity remains unknown unless the bundle carries an admitted root attestation and a trust policy is supplied.

Offline growth planning

Version 0.7.0 adds Bounded Endogenous Capability Frontier Expansion: an immutable finite super-catalogue with conditional model activation of latent actions, multi-step activation lineage, independent checking and reoptimized comparators. Try cpcf growth example --scenario frontier-chain --json and read the frontier guide. Model activation grants no capability admission or execution authority. No external empirical acceleration experiment was performed.

The explicit growth mode plans paid task/research capacity investment, measurement, verification and repair under a finite budget, and independently checks every policy branch and its funded continuation. It preserves the existing repair mode and operational profile.

cpcf growth example growth-demo --json
cpcf growth plan growth-demo/contract.json --objects growth-demo/objects --output growth-plan.json
cpcf growth check-plan growth-demo/contract.json --objects growth-demo/objects --plan growth-plan.json --json

Results are conditional on a declared finite model. Predictions do not admit measured capacity, and an incomplete comparator search cannot establish superiority. See the growth workflow, paper mapping, and synthetic comparisons.

Remote workspace use

Remote operations require a configured control plane and a short-lived OIDC access token. The CLI does not persist bearer tokens.

set CPCF_API_URL=https://cpcf.example.org
set CPCF_TOKEN=OIDC_ACCESS_TOKEN
cpcf workspace create WORKSPACE --root-spki-fingerprint sha256:ROOT_SPKI_SHA256 --genesis-envelope-fingerprint sha256:GENESIS_ENVELOPE_SHA256 --json
cpcf workspace status WORKSPACE --json
cpcf agent onboard --workspace WORKSPACE --json

On POSIX shells, use export. Mutations require an Idempotency-Key; changes to existing workspaces also require the expected immutable generation through If-Match.

Operational profile

Each required dimension is satisfied, violated, unknown, or unknown_due_to_budget:

1. provenance integrity
2. trust quorum
3. temporal integrity
4. structural reachability
5. causal formation
6. dimensional consistency
7. exact self-maintenance
8. finite-horizon resource persistence
9. target-bound generative catalysis
10. verification capacity
11. effective independence
12. coordination protocol integrity
13. perturbation robustness

Compatibility requires every required dimension to be satisfied in one immutable analysis snapshot. Hypothetical planner output cannot satisfy a scientific dimension before receipt-backed promotion.

Development

Contributors use the checked-in universal uv.lock; pip is only a consumer installation path. Development uses CPython 3.14.6 and supports CPython 3.12–3.14 on Windows and Linux.

uv sync --frozen --all-extras --group dev --group security
uv run --frozen ruff format --check .
uv run --frozen ruff check .
uv run --frozen mypy src packages/cpcf-api/src packages/cpcf-cli/src packages/cpcf-worker/src packages/cpcf-runner-protocol/src
uv run --frozen pytest
uv run --frozen python scripts/check_schemas.py
uv run --frozen python scripts/generate_references.py --check
uv run --frozen python scripts/check_publication_hygiene.py --source-tree
uv build

The root project builds exactly one wheel and one source distribution named collective-phase-control-fabric. The five import packages remain collective_phase_control_fabric, cpcf_cli, cpcf_api, cpcf_worker, and cpcf_runner_protocol.

Documentation

• Documentation index
• Installation
• Offline orientation
• Remote workspace use
• Evidence model
• Scientific boundaries
• Security and publication hygiene
• Complete installed-wheel tutorial
• Release readiness

Agent Skill

Agents can follow the repository-native Collective Phase Control Fabric Agent Skill. It routes offline inspection, remote control-plane work, external-runner evidence, and legacy compatibility without expanding the authority or scientific scope of the underlying records.

License and security

All repository content is English and licensed under Apache-2.0. Report vulnerabilities using the private process in SECURITY.md; never include credentials, tenant evidence, or live endpoints in a public issue.