Test strange patterns, not metaphysics.
A tiny, zero-dependency Python library and CLI for spotting unusual statistical structure in time-series data.
This project does not prove whether the world is real or simulated. It only helps detect suspicious regularity, quantization, and temporal structure in observed data.
The original idea started as a thought experiment:
- Can we look for signs of excessive regularity?
- Can we detect quantized or grid-like behavior in data?
- Can we score unusual sequential patterns without overclaiming?
这个库的灵感来自一个"如果世界是模拟的，我们能不能用数据检测出来"的思想实验。 当然，它不能证明任何关于世界本质的东西。但它可以帮你快速检测数据里不寻常的统计结构。
In 2024, scientists completed the full connectome of a fruit fly brain — all 139,255 neurons and 54.5 million synaptic connections — and successfully simulated its neural activity in a computer (Nature, 2024).
This means:
- Biological brains can be digitized. If a fruit fly brain can be fully mapped and run as software, the gap between "biology" and "simulation" is not a wall — it's a scale problem.
- Simulating larger brains is a matter of compute, not principle. A mouse brain has ~70 million neurons. A human brain has ~86 billion. The architecture is similar; we just need more power.
- If a simulation can produce real behavior, how would the simulated entity know? The simulated fruit fly responds to stimuli just like a real one. From the inside, there may be no difference.
This doesn't prove we live in a simulation. But it removes the strongest objection: "you can't simulate a brain." We already can.
What RealityLab does: if a simulation has finite precision — quantized time steps, grid-snapped values, or deterministic pseudo-randomness — those artifacts might leak into observable data. This library looks for exactly those patterns.
2024 年科学家完成了果蝇大脑全部 139,255 个神经元、5450 万个突触连接的完整图谱，并在计算机中成功模拟了其神经活动。
这意味着：生物大脑可以被数字化运行。模拟更大的大脑只是算力问题，不是原理问题。 如果模拟能够产生和真实一样的行为，被模拟的对象从内部可能根本无法区分。
RealityLab 的逻辑是：如果模拟存在有限精度（量化时间步、网格化数值、伪随机数），这些痕迹可能会泄露到可观测数据中。这个库就是用来寻找这些痕迹的。
- Analyze recorded sensor data / 检测传感器数据异常
- Inspect latency or timing traces / 分析网络延迟序列
- Compare synthetic vs natural-looking sequences / 区分人工和自然序列
- Build weird but fun demos for blogs, talks, or social posts / 做有趣的数据实验
| Feature | Description |
|---|---|
| Shannon entropy | Measures information density in the sequence |
| Autocorrelation | Detects temporal dependencies between adjacent values |
| Discretization score | Finds quantization / grid-like step patterns |
| Run-length deviation | Checks if high/low runs are unnaturally regular |
| CSV loader | Load any CSV with a numeric column |
| CLI | One-command analysis from terminal |
| Zero dependencies | Pure Python standard library |
pip install .from reality_detector import RealityAnomalyDetector
data = [1, 4, 2, 7, 5, 9, 3, 6]
report = RealityAnomalyDetector(data).analyze()
print(report.to_dict())# Text report
reality-detector examples/demo.csv --column value
# JSON output
python -m reality_detector examples/demo.csv --column value --jsonpython examples/demo_interactive.pyRuns 5 experiments comparing natural vs artificial data: random numbers, temperature, heartbeat, neuron spikes, and stock returns.
运行 5 组对比实验（真随机 vs 假随机、自然温度 vs 量化温度、心跳 vs 时钟、神经放电、股市收益），直观展示检测器的区分能力。
Example output:
Reality Anomaly Detector Report
===============================
Samples: 12
Entropy: 3.5850
Autocorrelation (lag=1): -0.1940
Discretization score: 0.2727
Run-length deviation: 1.0000
Anomaly score: 0/4
Interpretation: No obvious anomaly detected.
Any CSV file with at least one numeric column works.
Example:
step,value
1,10
2,12
3,9
4,14If --column is not provided, the CLI picks the first numeric-looking column.
| Score | Meaning |
|---|---|
| 0 | No obvious anomaly / 未发现异常 |
| 1 | Mild anomaly, likely normal variation / 轻微异常，很可能是正常波动 |
| 2 | Notable anomaly, collect more data / 值得注意，建议增加样本 |
| 3-4 | Strong anomaly signal, but still not proof of simulation / 异常较强，但仍不能证明是模拟 |
| File | Description |
|---|---|
examples/demo.csv |
Basic numeric series |
examples/temperature.csv |
Natural vs quantized temperature |
examples/neuron_spikes.csv |
Natural vs simulated neuron firing |
examples/demo_interactive.py |
Full 5-experiment comparison script |
RealityLab/
├── src/reality_detector/ # Core library
│ ├── core.py # Detector & report
│ ├── io.py # CSV loader
│ └── cli.py # CLI entry point
├── tests/ # Test suite
├── examples/ # Demo data & interactive script
├── scripts/ # Image generation helpers
└── .github/workflows/ # CI + PyPI publish
- JSON and NDJSON input support
- Visualization helpers (matplotlib / plotly)
- FFT-based periodicity checks
- Chi-squared and Kolmogorov-Smirnov tests
- Benchmark dataset bundle
- PyPI release
See CONTRIBUTING.md for guidelines.
Made with curiosity. Star this repo if you find the idea intriguing.
如果觉得有趣，欢迎点一颗星。