Function error_budget 

pub fn error_budget(
    base: &ResolvedSolveRequestV1,
    sources: &[(InputAxis, f64)],
    ranges_m: &[f64],
) -> Result<ErrorBudgetReportV1, KernelError>

Propagate each declared per-input uncertainty in sources to impact covariance at every range in ranges_m, via central differences through the real solver (central_difference), and rank the sources by their share of impact variance.

sources is (axis, sigma) pairs: sigma is the caller’s one-sigma uncertainty for that axis, in the axis’s own physical unit (crate::perturbation::axis_meta(axis).kind). Every sigma must be finite and non-negative and every axis must appear at most once – both are validated up front, before any solve; see “Sources and ranges are validated up front” and # Errors below. (Declaring the same axis twice would double-count its variance and make its own leave-one-out counterfactual ambiguous – which of the two entries would “removing this source” mean? – so it is rejected rather than given an arbitrary answer.)

Honesty

ErrorBudgetReportV1::method and ErrorBudgetReportV1::assumptions state, in the payload itself (not only in prose documentation), that: sources are treated as INDEPENDENT (no correlation between them is modelled); propagation is FIRST-ORDER/local-linear about the nominal solution, using the axis’s own small default differencing step regardless of the declared sigma (not exact for large or non-Gaussian input uncertainty – see “Why the differencing step ignores the declared sigma” above); the 95% ellipse assumes an approximately Gaussian impact distribution; a source’s derivative may be one-sided rather than central; and an unavailable source is not the same fact as a zero-contribution one. See the_report_declares_independence_and_linearity in this module’s tests.

Unavailable sources

See this module’s top-level “Unavailable sources” doc section.

Errors

• KernelError::Observation immediately, before any solve, if any range_m in ranges_m is not finite or falls outside [0, base.shot.max_range_m] – see “Sources and ranges are validated up front” above.
• KernelError::NonFinite immediately if any declared sigma is not finite or is negative.
• KernelError::DuplicateAxis immediately if the same axis appears more than once in sources.
• Otherwise, propagates any KernelError from central_difference that is not one of the four structural refusals recorded in ErrorBudgetReportV1::unavailable_sources instead – a genuine solver or trajectory failure, not a normal “this input cannot be perturbed here” fact.

A thin wrapper over error_budget_with_target passing None – every row’s p_hit and every source’s p_hit_gain_if_perfect come back None (never a fabricated number), and method/assumptions say nothing about hit probability. Call error_budget_with_target directly to also get those.