0097 — The chi-squared tail joins the published four, by the rule 0095 already set

Status: accepted · Date: 2026-09-10

Context

Decision 0095 published four numerical members and wrote the rule it published them under:

Four members become public, and nothing else. […] Nothing has asked for them, and 0081's asymmetry holds in both directions: publishing later is always available, unpublishing never is.

#569 asks. A log-rank test compares two survival curves and reports a chi-squared p-value on one degree of freedom. Lodestar.Survival is a separate package and cannot reach Gamma.RegularizedQ, which 0095 listed by name among the members that stay internal.

The issue put three options, and required that they not be answered twice, differently, from the way #566 answered the same question for the Student and Fisher tails.

Decision

Distributions.ChiSquaredSf becomes public. Lodestar.Stats goes 0.2.0 → 0.3.0, and Lodestar.Survival takes an edge onto it, exactly as Lodestar.Stats.Regression takes one for the three tails before it.

This record does not amend 0095 so much as apply it. 0095's sentence is "the numerical layer publishes what one caller needs, and no more"; a second caller has now named a fifth member, and the answer is the one 0095 wrote down in advance. RegularizedIncomplete, LogGamma, RegularizedP, RegularizedQ, Erfc, Normal.Sf, Normal.Quantile, the finite-sample Kolmogorov distribution, PartialPivotLu, JacobiSvd and the DenseBlock helpers stay internal, unchanged.

Options refused

Re-derive a chi-squared tail inside Lodestar.Survival. Refused on the ground 0081 already stated against exactly this, in the note that erfc is Q(1/2, x²): a second far-tail approximation in one repository has to agree with the first, and nothing makes it. The published member and the tail ChiSquare.GoodnessOfFit and ChiSquare.Contingency already use are now the same function, and a test asserts that a statistic routed either way gives the same p-value.

Publish the whole numerical layer. Refused, as 0081 and 0095 both refused it.

Move the layer into Lodestar.Abstractions. Refused, as 0095 refused it and generalised the refusal: Abstractions carries the types packages exchange, and a function is not a type two packages exchange.

Consequences

• One member, one reference page, one row in the tails index, one use in the packaging sample, and six cases added to tests/oracles/stats_distributions.json — reaching 7.7e-26, compared relatively, because an absolute 1e-9 there accepts an implementation returning zero.
• x is not validated. The three members before it refuse a non-positive degrees of freedom and nothing else; this one also leaves x unguarded, and returns one below the support rather than surfacing the internal helper's own ParamName from a public method.
• Lodestar.Survival cannot land in the same pull request. src/ reaches a sibling through a PackageReference on a published floor (decision 0012), so 0.3.0 must be on the feed before a package may floor on it — CONTRIBUTING's Working across two packages states it, and #566 followed it: its four members landed in their own commit and were tagged before Lodestar.Stats.Regression raised the floor. This lot is split the same way, for the same reason.