feat: add OpenEntropy source and ECDF amplifier from PR #4

Integrate OpenEntropy hardware entropy source and ECDF signal amplifier
from ereid7's PR #4, with fixes for CI failures:

- Move openentropy to optional dependency (was breaking Python 3.10/3.11/3.13)
- Fix ruff formatting on openentropy.py and test_openentropy.py
- Remove bandit-flagged assert in ecdf.py (invariant enforced by _calibrated)
- Add ECDF calibration wiring in processor.py
- Optimize _to_numpy with is_cpu check to avoid unnecessary .cpu() calls
- Add oe_* config fields, entry point, mypy override, and tests

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Author: alchemystack

README.md

@@ -353,6 +353,7 @@ All fallback-sourced entropy is flagged in diagnostic logs so downstream analysi
 | **System** | `system` | `os.urandom()` — OS cryptographic RNG (fallback/testing) |
 | **Timing noise** | `timing_noise` | CPU timing jitter (experimental) |
 | **Mock uniform** | `mock_uniform` | Configurable test source with seed/bias |
+| **OpenEntropy** | `openentropy` | 63 hardware noise sources (thermal, timing, microarch, GPU) — local, no network |
 
 ### Fallback behavior
 
@@ -367,6 +368,38 @@ Configure with `QR_FALLBACK_MODE`:
 - `mock_uniform` — fall back to the mock source
 - `error` — raise immediately, no fallback
 
+### OpenEntropy
+
+[OpenEntropy](https://github.com/amenti-labs/openentropy) harvests entropy from 63 hardware noise sources on the local machine — thermal sensors, CPU timing jitter, memory timing, GPU scheduling, and more. No network, no API keys, no gRPC server needed.
+
+Install:
+
+```bash
+pip install openentropy
+```
+
+Configure:
+
+```bash
+export QR_ENTROPY_SOURCE_TYPE=openentropy
+export QR_OE_CONDITIONING=raw   # raw (research default) | vonneumann | sha256
+```
+
+List available sources on your machine:
+
+```python
+from openentropy import detect_available_sources
+print([s["name"] for s in detect_available_sources()])
+```
+
+To sample from specific sources only, set `QR_OE_SOURCES` to a comma-separated list:
+
+```bash
+export QR_OE_SOURCES=clock_jitter,dram_row_buffer
+```
+
+See [`deployments/openentropy/`](deployments/openentropy/) for the full deployment profile.
+
 ### Third-party entropy sources
 
 Any Python package can register entropy sources via entry points:

deployments/openentropy/.env.example

@@ -0,0 +1,49 @@
+# deployments/openentropy/.env
+#
+# Environment variables for the OpenEntropy native entropy profile.
+# Copy this file to .env and edit as needed:
+#   cp .env.example .env
+
+# --- Hugging Face ---
+# Model to serve. Default: Qwen/Qwen2.5-1.5B-Instruct
+HF_MODEL=Qwen/Qwen2.5-1.5B-Instruct
+# Set this if the model is gated (requires accepting a license).
+HF_TOKEN=
+
+# --- Entropy source ---
+# OpenEntropy: local hardware entropy (no Docker, native only)
+# 63 hardware noise sources: thermal, timing, microarch, GPU, etc.
+# Install: pip install openentropy
+# Docs: https://github.com/amenti-labs/openentropy
+QR_ENTROPY_SOURCE_TYPE=openentropy
+
+# Conditioning mode: raw (research default), vonneumann (debiased), sha256 (crypto)
+QR_OE_CONDITIONING=raw
+
+# Comma-separated list of source names to use. Empty = all available sources.
+QR_OE_SOURCES=
+
+# Enable parallel collection from multiple sources (faster, more entropy).
+QR_OE_PARALLEL=true
+
+# Timeout in seconds for entropy collection.
+QR_OE_TIMEOUT=5.0
+
+# Fall back to system entropy if OpenEntropy is unavailable.
+QR_FALLBACK_MODE=system
+
+# --- Sampling parameters ---
+QR_SAMPLE_COUNT=20480
+QR_TEMPERATURE_STRATEGY=fixed
+QR_FIXED_TEMPERATURE=0.7
+QR_TOP_K=0
+QR_TOP_P=1.0
+QR_LOG_LEVEL=summary
+
+# --- Ports (host-side) ---
+VLLM_PORT=8000
+
+# --- Open WebUI (optional, --profile ui) ---
+OPEN_WEBUI_PORT=3000
+# Set to true to require login (recommended for shared/public servers).
+OPEN_WEBUI_AUTH=false

deployments/openentropy/README.md

@@ -0,0 +1,128 @@
+# OpenEntropy Profile
+
+Runs vLLM with qr-sampler using **OpenEntropy** — a local hardware entropy
+source that collects noise from 63 hardware sources on Apple Silicon (thermal,
+timing, microarchitecture, GPU, etc.). This is a **native-only profile** — no
+Docker, no network dependency.
+
+## Why not Docker?
+
+Docker containers cannot access Metal GPU or native hardware entropy sources on
+macOS. Apple's Virtualization.framework has no GPU passthrough, and hardware
+noise sources (thermal sensors, CPU timing, GPU state) are not exposed to
+containerized processes. OpenEntropy requires native execution.
+
+## Quick start
+
+1. Install OpenEntropy and qr-sampler:
+
+   ```bash
+   pip install openentropy
+   pip install -e /path/to/qr-sampler
+   ```
+
+2. Configure your environment:
+
+   ```bash
+   cd deployments/openentropy
+   cp .env.example .env
+   ```
+
+   Edit `.env` if needed — set `HF_TOKEN` if using a gated model.
+
+3. Start vLLM:
+
+   ```bash
+   source .env
+   vllm serve $HF_MODEL \
+     --port $VLLM_PORT \
+     --logits-processors qr_sampler
+   ```
+
+## Available entropy sources
+
+OpenEntropy provides 63 entropy sources across 13 categories. For the full
+catalog with physics explanations, see the
+[OpenEntropy Source Catalog](https://github.com/amenti-labs/openentropy/blob/master/docs/SOURCES.md).
+
+List all available sources on your hardware:
+
+```bash
+python -c "from openentropy import detect_available_sources; print([s['name'] for s in detect_available_sources()])"
+```
+
+Sources span thermal, timing, microarchitecture, GPU, IPC, scheduling, and more.
+Some notable ones for research:
+
+| Source | Category | Physical mechanism |
+|--------|----------|-------------------|
+| `counter_beat` | Thermal | CPU counter vs audio PLL crystal beat frequency |
+| `dual_clock_domain` | Microarch | 24 MHz x 41 MHz independent oscillator beat |
+| `gpu_divergence` | GPU | Shader warp execution order divergence |
+| `dvfs_race` | Microarch | Cross-core DVFS frequency race |
+| `clock_jitter` | Timing | Timing jitter between readout paths |
+| `dram_row_buffer` | Timing | DRAM row buffer hit/miss timing |
+
+To sample from a specific source, set `QR_OE_SOURCES`:
+
+```bash
+export QR_OE_SOURCES=counter_beat
+```
+
+## Conditioning modes
+
+OpenEntropy supports three conditioning strategies:
+
+| Mode | Use case | Properties |
+|------|----------|-----------|
+| `raw` | Research (default) | Preserves hardware noise signal; minimal processing |
+| `vonneumann` | Debiased entropy | Von Neumann debiasing; slower, more uniform |
+| `sha256` | Cryptographic | SHA-256 hashing; suitable for security-critical applications |
+
+Set `QR_OE_CONDITIONING` in `.env` or override per-request:
+
+```python
+# Per-request override
+extra_args = {"qr_oe_conditioning": "sha256"}
+```
+
+## Parallel collection
+
+By default, `QR_OE_PARALLEL=true` collects from multiple sources simultaneously,
+increasing entropy throughput. Set to `false` for sequential collection (slower,
+lower memory overhead).
+
+## When to use this profile
+
+- **Consciousness research**: Study whether intent influences quantum-random
+  processes using native hardware entropy.
+- **Local experiments**: No network latency, no external dependencies.
+- **Apple Silicon development**: Leverage Metal GPU and native hardware sensors.
+- **Research baseline**: Compare hardware entropy against system entropy
+  (`/dev/urandom`).
+
+## Web UI (optional)
+
+This profile includes [Open WebUI](https://github.com/open-webui/open-webui), a
+ChatGPT-style web interface. To use it, you'll need to run it separately (not
+included in this native profile):
+
+```bash
+docker run -d -p 3000:3000 --name open-webui ghcr.io/open-webui/open-webui:latest
+```
+
+Then point it at your vLLM instance running on `localhost:8000`.
+
+A pre-built filter function for controlling qr-sampler parameters from the UI is
+available at [`examples/open-webui/`](../../examples/open-webui/). See that
+directory's README for import instructions.
+
+## Next steps
+
+Once this profile works, you can:
+1. Adjust `QR_OE_SOURCES` to use specific entropy sources.
+2. Experiment with different conditioning modes (`raw`, `vonneumann`, `sha256`).
+3. Compare results against the `urandom` profile (gRPC-based) or `system` profile
+   (fallback).
+4. Browse the full [OpenEntropy Source Catalog](https://github.com/amenti-labs/openentropy/blob/master/docs/SOURCES.md)
+   for detailed physics explanations of each entropy source.

pyproject.toml

@@ -46,6 +46,9 @@ dev = [
     "pre-commit>=4.0",
     "bandit>=1.8.0",
 ]
+openentropy = [
+    "openentropy>=0.10.0",
+]
 
 [project.entry-points."vllm.logits_processors"]
 qr_sampler = "qr_sampler.processor:QRSamplerLogitsProcessor"
@@ -55,6 +58,7 @@ system = "qr_sampler.entropy.system:SystemEntropySource"
 quantum_grpc = "qr_sampler.entropy.quantum:QuantumGrpcSource"
 timing_noise = "qr_sampler.entropy.timing:TimingNoiseSource"
 mock_uniform = "qr_sampler.entropy.mock:MockUniformSource"
+openentropy = "qr_sampler.entropy.openentropy:OpenEntropySource"
 
 [tool.setuptools.packages.find]
 where = ["src"]
@@ -98,6 +102,10 @@ ignore_missing_imports = true
 module = ["torch", "torch.*"]
 ignore_missing_imports = true
 
+[[tool.mypy.overrides]]
+module = "openentropy.*"
+ignore_missing_imports = true
+
 [tool.coverage.run]
 source = ["src/qr_sampler"]
 omit = [

src/qr_sampler/amplification/__init__.py

@@ -6,12 +6,14 @@
 """
 
 from qr_sampler.amplification.base import AmplificationResult, SignalAmplifier
+from qr_sampler.amplification.ecdf import ECDFAmplifier
 from qr_sampler.amplification.registry import AmplifierRegistry
 from qr_sampler.amplification.zscore import ZScoreMeanAmplifier
 
 __all__ = [
     "AmplificationResult",
     "AmplifierRegistry",
+    "ECDFAmplifier",
     "SignalAmplifier",
     "ZScoreMeanAmplifier",
 ]