Simulation_Invariant

A bounded, inspectable AI environment that externalizes memory, exposes unseen state, and reconstructs situational continuity without granting the model authority.

Installation

From a cloned repository, install the package and its console command in editable mode:

python -m pip install -e .

The simulation-invariant command and python -m sim use the same dispatcher. The bounded-candidate executor retains its separately declared Linux/POSIX namespace requirements.

Quick start

Run these commands from the repository root.

Observe the current environment without modifying it:

python -m sim monitor . --exclude .git --exclude .pytest_cache

Present a bounded range of one file and receive a hash-bound presentation receipt:

python -m sim inspect . README.md --max-bytes 4096

Run the complete test suite:

python -m pytest -q

Verify the exact portable candidate fixture published by this repository:

python -c "import json; from pathlib import Path; from sim.bounded_candidate.evidence_bundle import verify_candidate_evidence_bundle as verify; print(verify(json.loads(Path('interop/portable-candidate-evidence-bundle-v1.json').read_text(encoding='utf-8'))))"

The monitor and inspection commands are read-only and grant no write or execution authority.

What VALIDATED_CANDIDATE means

The bounded-candidate pipeline is a Linux/POSIX Python API whose full executor requires mount, network, and PID namespaces. VALIDATED_CANDIDATE means the disposable candidate matched the approved tree-state hash, the declared tests passed under the executor's verified properties, and the source remained bound to its observed manifest during that validation cycle.

It does not mean the candidate was accepted, deployed, promoted, or written back to the source. A validated candidate retains:

write_authority=NONE
promotion_authority=NONE

Capability boundaries

Simulation_Invariant deliberately separates portable observation and evidence verification from the Linux/POSIX containment required by the full candidate executor. Read the capability boundary before interpreting a receipt, status, skip, or validated candidate.

Contributing

Contributions are welcome when they preserve the project's evidence and authority boundaries. Read CONTRIBUTING.md before opening a pull request.