Python bindings for the sentil engine. pip install sentil and import it.

SENTIL has three main capabilities. Deterministic STL monitoring, offline over a recorded trace or streaming one sample at a time. Probabilistic monitoring, which fits a noise model to sensor data and estimates satisfaction probability with confidence bounds. And synthesis, from a specification to a control input to an online controller.

Your first monitor

import sentil

trace = sentil.Trace([0, 1, 2, 3, 4], {"speed": [12, 9, 7, 4, 6]})
phi = sentil.Formula.parse("always (speed > 5)")
print(phi.robustness(trace))   # -1.0

The robustness is -1 because the speed dips to 4 at t = 3, one unit under the bound, so the property fails by exactly one. A non-negative value would mean it holds, and the magnitude is the margin. Formulas compose with Python operators, and a trace behaves like a mapping from a name to a NumPy array:

from sentil import var, always, eventually

phi = always(var("speed") > 5) & eventually(var("gap") > 2)
trace["speed"]        # the values, as an ndarray

Online streaming

An OnlineMonitor folds one timed reading at a time, at O(1) amortized cost per sample and memory that scales with the window, not the length of the trace. The verdict carries resolved, satisfied, and value, so you can watch a live system and stop the moment it breaks.

import math
import sentil

monitor = sentil.OnlineMonitor("always[0, 10] (x > -0.9)")
for t in range(60):
    x = math.sin(t * 0.3)
    verdict = monitor.update(float(t), {"x": x})
    if verdict.resolved and not verdict.satisfied:
        print(f"violated at t={t}, robustness={verdict.value:.3f}")
        break
else:
    print("held over the whole stream")

satisfied only carries a verdict once resolved is true; until then the monitor is still filling the window, so bound the horizon on a future-time operator.

Probabilistic monitoring

A P~p operator asks whether a formula holds with probability at least (or at most) p. Register a noise model for each sensor; SENTIL lifts every reading into an ensemble, evaluates the formula on each, and returns the probability with a Wilson confidence interval.

import sentil
from sentil import Formula, LiftingRegistry, NoiseModel, SmcConfig

trace = sentil.Trace(list(range(20)), {"x": [0.4 + 0.05 * i for i in range(20)]})

lifting = LiftingRegistry()
lifting.register("x", NoiseModel.gaussian(0.0, 0.3))

phi = Formula.parse("P>=0.9 (always (x > 0))")
result = phi.check(trace, lifting, SmcConfig(samples=5000))
print(f"probability {result.probability:.3f}, "
      f"interval [{result.interval.lower:.3f}, {result.interval.upper:.3f}], "
      f"holds {result.holds}")

Specifications

The premade library is on the Python side too: vetted specifications across ten domains (aerospace, automotive, controls, financial, industrial, medical, networking, power, robotics, UAV), each with a description, a citation, default parameters, and a deterministic and a probabilistic form. Reach for one instead of writing your own.

import sentil

phi = (
    sentil.SpecBuilder("automotive/safe_following_distance")
    .with_param("rho", 1.0)  # the follower's reaction time
    .build_formula()
)
# phi monitors gap, v_r, and v_f against the RSS safe-distance bound

List them with sentil.SpecBuilder.available(), or browse them under specifications/.

Benchmarks

These plots put Python and the Rust core against the baseline tools, from the same runs.

Per-sample streaming cost across the bindings, with the Rust core in front. The offline baselines have no online mode, so nothing else can stream a sample at a time.

Offline cost over the trace length, Python and the core against RTAMT, MoonLight, and Banquo.

Peak memory over the length of the stream.

The full set, including the dense-time, statistical model checking, rare-event, and synthesis benchmarks, is in benchmarks/, and all the results are in docs/CLAIMS.md.

Install

Package manager

From PyPI, with pip:

pip install sentil

or with uv:

uv add sentil

The optional extras pull their dependencies: sentil[pandas] for DataFrame ingest, sentil[plotting] for the robustness plots.

Prebuilt release

Every tagged release attaches prebuilt wheels to the GitHub release; download the one for your Python and platform and install it directly.

Linux and macOS

The shell expands the wildcard to the file you downloaded:

pip install ./sentil-0.3.0-*.whl

Windows

Pass the exact filename, since the shell leaves the wildcard unexpanded:

pip install .\sentil-0.3.0-cp311-cp311-win_amd64.whl

Build from source

You need a Rust toolchain and maturin.

git clone https://github.com/sedislab/SENTIL
cd SENTIL/sentil-py
maturin develop --release
python -m pytest tests

Contributing

Build the extension into your environment and run the suite:

maturin develop --release
python -m pytest tests

The pull-request flow is in the repository CONTRIBUTING.md.

Documentation

The documentation site carries the guides, the specification syntax, and the long-form tutorial.

Citation

If SENTIL is useful in your work, please cite the paper:

@misc{quansah2026sentilruntimeverificationtool,
    title={SENTIL: A Runtime Verification Tool for Probabilistic Temporal Logic},
    author={Paapa Kwesi Quansah and Ernest Bonnah},
    year={2026},
    eprint={2605.21676},
    archivePrefix={arXiv},
    primaryClass={cs.LO},
    url={https://arxiv.org/abs/2605.21676}
}

License

SENTIL is by Paapa Kwesi Quansah and Ernest Bonnah at the SEDIS lab, Baylor University. It is dual licensed under either MIT or Apache-2.0, at your option.