The configuration structs that steer a monitor and its statistical methods, their fields and defaults, and how the command-line tool layers settings from files, environment, and flags.

A monitor runs with sensible defaults out of the box, so most checks need no configuration at all. When a run needs tuning, the settings live in a small set of structs, each with a default that suits a first run. The tables below are the field-by-field reference; the concept pages explain when to change a value.

MonitorConfig

MonitorConfig holds how a Monitor checks its formula: the time mode for offline robustness, and, with the statistical feature, the Monte Carlo and rare-event settings. It is built with new() and refined with chained builders.

Builder	Sets	Default
`time(mode)`	the `TimeMode`: `Discrete` or `Dense`	`Discrete`
`smc(config)`	the `SmcConfig` used by `Monitor::check`	`SmcConfig::default()`
`rare(config)`	the `RareEventConfig` used by `Monitor::check_rare`	`RareEventConfig::default()`

use sentil::{Monitor, MonitorConfig, TimeMode};
use sentil::stats::SmcConfig;

let config = MonitorConfig::new()
    .time(TimeMode::Dense)
    .smc(SmcConfig { samples: 50_000, ..SmcConfig::default() });
let monitor = Monitor::new("P>=0.95(G (speed > 5))", config)?;

SPRT and Bayesian testing take their config per call rather than through MonitorConfig, since their hypotheses have no sensible default.

SmcConfig

The fixed-sample Monte Carlo settings, a plain struct with public fields.

Field	Meaning	Default
`samples`	how many noisy realizations to draw	`10_000`
`confidence`	the confidence level for the reported interval	`0.95`
`seed`	the base seed, so a run reproduces exactly	`42`
`interval_method`	which interval to report	`IntervalMethod::Wilson`

SprtConfig

Wald's sequential test, built with SprtConfig::new(p0, p1, alpha, beta, max_samples) and an optional with_seed.

Parameter	Meaning	Constraint
`p0`, `p1`	the indifference region bounds	`0 < p0 < p1 < 1`
`alpha`	the false-positive rate bound	in `(0, 1)`
`beta`	the false-negative rate bound	in `(0, 1)`
`max_samples`	the budget at which an undecided run is `Inconclusive`	positive
`seed`	the base seed	defaults to `42`

BayesConfig

The Bayesian sequential test, built with BayesConfig::new(threshold, bayes_factor, max_samples) and an optional with_seed.

Parameter	Meaning	Constraint
`threshold`	the probability the test decides against	in `(0, 1)`
`bayes_factor`	the evidence cutoff before deciding	greater than `1`
`max_samples`	the budget at which an undecided run is `Inconclusive`	positive
`seed`	the base seed	defaults to `42`

The prior is the uniform Beta $(1, 1)$ ; a larger bayes_factor demands stronger evidence and spends more samples.

RareEventConfig

The adaptive multilevel splitting settings, a plain struct with public fields.

Field	Meaning	Default
`particles`	the particle population; more tightens the estimate at linear cost	`4096`
`margin`	the violation margin; the rare event is robustness dropping to `-margin` or below	`0.0`
`seed`	the seed, for a reproducible run	`42`

SmoothConfig

The smooth-robustness settings synthesis differentiates through: a SoftKind, log-sum-exp or arithmetic-geometric mean, and its temperature. The fields and the semantics they select are on smooth semantics.

Command-line layered config

The CLI reads defaults a team can set once and share. The precedence, highest to lowest, is a command-line flag, then a SENTIL_* environment variable, then ./sentil.toml, then the per-user config under the XDG directory, then /etc/sentil/config.toml, then the built-in default. A file value stands in only where neither a flag nor an environment variable was given, and a later file merges over an earlier one. An explicit --config PATH replaces the whole file chain.

sentil.toml

output = "ndjson"
color = "never"

[alias]
overshoot = "check --spec controls/overshoot"

A config file may set the default output format, the color policy, and named command aliases, which expand when their name is not a built-in verb. Run sentil config to see which files are consulted, which exist, and the values in effect.

For what these settings drive, see statistical methods and the CLI page.

Configuration