QSOT-Harness

QSOT stands for Quantum State Over Time. QSOT-Harness is a bounded, epistemically-labeled CLI harness that runs quantum-channel models of curved-spacetime backgrounds and emits machine-readable verification artifacts (result.json, report.md). It verifies model-output consistency under explicitly stated assumptions — not external physical validity. The pipeline is a strict, reproducible "Print-and-Exit" CLI runner.

Honest reconstruction. This is the genuine rebuild of QSOT. Its predecessor ("QSOT Compiler V1") was published as an intentional high-formality slop artifact — executable and formal-looking, but scientifically hollow. QSOT-Harness removes that slop: every reported signal declares its evidence class, every external/optional backend is surfaced in audit_context, the KD signal is reported flat-relative (not as a contextuality proof), non-Markovianity self-tests are kept distinct from model trajectories, and free calibration parameters are listed with what they govern. The claim boundary is machine-readable — see docs/theory/claim_boundary.md.

What this is and is not

• Is: a reproducible verification harness for a phenomenological channel-map model; a rigorously bounded, self-labeling experiment runner whose every output declares its epistemic class.
• Is not: a physics proof, a Theory of Everything, or external experimental validation. A passing check means model-internal consistency under the stated assumptions — nothing more.

---

Features

• CLI-First Architecture: Run experiments directly with configuration files:

  python run_experiment.py --config configs/experiment.yaml

• 8-Phase Verification Pipeline: Executes 50 sequential checks (5 temporal-state axiom checks + 45 model verification checks) across temporal-state axioms, flat/curved baselines, boosts, Kirkwood-Dirac optimization, a Transfer-Tensor-inspired (TTM) non-Markovianity proxy, Rust sidecars, and scientific audit reviews.
• Strict Verdict Resolution: Propagates execution status (PASS, FAIL, SKIPPED, DEGRADED_PASS) to determine the overall experimental outcome.
• Optional & local integrations: optional backends enrich a run when installed; the pipeline degrades cleanly without them and records which backend was used in result.json (audit_context).
  • Scientific compliance audit (spar-framework, optional [audit] extra): produces a bounded, environment-dependent review signal over gate outputs and report text, including a normalized claim-drift score. When the framework is absent, a local deterministic mock runs and the result is marked backend_mode: "mock". This is a review signal, not external physical validation.
  • Accessibility scoring (local): derives an accessibility score from gate status with non-Markovianity, KD-negativity, and risk penalties.
  • Differentiable governance engine (flame_torch, externally provisioned, optional): a covenant check on flat-baseline purity; skipped and reported as SKIPPED when not installed.
• Rust Sidecar Vector Search: Subprocess execution of the turbovec-powered rust_verify package comparing exact dot products against quantized similarity search.

---

Directory Structure

QSOT-Harness/
├── .github/workflows/       # CI/CD pipelines
│   └── ci.yml
├── configs/                 # YAML configuration presets
│   └── experiment.yaml
├── reports/                 # Raw and rendered outputs (result.json, report.md)
├── src/qsot_v2/
│   ├── cli/                 # Argparse and run script entrypoints
│   ├── core/                # Configuration, Results schema, and Phase Runner
│   ├── physics/             # Evolution, Channels, Accessibility, Optimizer, and Scientific Audit
│   ├── reports/             # Jinja2 template and markdown rendering
│   └── rust_sidecar/        # Cargo sidecar implementing turbovec search
└── tests/                   # Complete unit and integration test suite

---

Quick Start

1. Installation

Activate the Python virtual environment and install dependencies:

# Windows
.\venv\Scripts\Activate.ps1
pip install -e .[dev,torch]

2. Build the Rust Sidecar

Build the sidecar binary using the virtualenv context:

cd src/qsot_v2/rust_sidecar
cargo build --release
cd ../../..

3. Run Experiments

Execute the full verification suite against the background presets:

python run_experiment.py --config configs/experiment.yaml

The outputs will be generated inside the reports/ folder:

• reports/result.json: Comprehensive machine-readable execution details and NumPy-sanitized metrics.
• reports/report.md: Beautiful, human-readable summary of checks and physics interpretations.

---

8-Phase Physics Pipeline

0. Phase 0: Temporal-State Axioms: Evaluates temporal quantum evolution representation validity under explicit mathematical constraints (Linearity, CPTP, Trace preservation, Conditionability consistency, and Density matrix validity).
1. Phase 1: Flat Baselines: Establishes rest and boosted flat spacetime limits ($P=1.0, S=0$).
2. Phase 2: Curvature Noise: Drives state purity decay and entropy growth across de Sitter, Schwarzschild, AdS5, and Eguchi-Hanson metrics.
3. Phase 3: Relativistic Boosts: Checks that, within the implemented model, high-velocity observers experience increased effective noise.
4. Phase 4: Kirkwood-Dirac (flat-relative): Runs a bounded basis-search optimization and reports the de-Sitter-minus-flat KD delta as a relative signal under the implemented model — not a contextuality proof. Optimizer convergence is labeled explicitly (a non-converged run is surfaced as DEGRADED_PASS).
5. Phase 5: TTM-inspired memory proxy & Accessibility: Runs a TTM-inspired trace-distance proxy (a bounded drift diagnostic, not a full transfer-tensor reconstruction) and determines accessibility.
6. Phase 6: Rust turbovec: Tests vector search quantization limits under the turbovec sidecar.
7. Phase 7: Scientific Audit: Compares each background's gate outputs and residuals against the project's own predefined expected-behavior table, producing a bounded, environment-dependent review signal — not external ground truth.

---

Running Tests

Run the complete suite of tests and check coverage:

python -m pytest --cov=qsot_v2 --cov-report=term-missing

Coverage is gated at 90% (--cov-fail-under=90 in pyproject.toml); the current suite measures ~94%.