Gil.Review
0.2.1
dotnet add package Gil.Review --version 0.2.1
NuGet\Install-Package Gil.Review -Version 0.2.1
<PackageReference Include="Gil.Review" Version="0.2.1" />
<PackageVersion Include="Gil.Review" Version="0.2.1" />
<PackageReference Include="Gil.Review" />
paket add Gil.Review --version 0.2.1
#r "nuget: Gil.Review, 0.2.1"
#:package Gil.Review@0.2.1
#addin nuget:?package=Gil.Review&version=0.2.1
#tool nuget:?package=Gil.Review&version=0.2.1
Gil.Review
Review statistics for checking automated decisions by hand, for .NET — a binding to the Rust crate
iyulab-gil-review (repository):
- which records to review — a seeded pilot draw over strata you form (
ReviewSampler.DrawPilot), then the Neyman allocation of the round after it (ReviewSampler.AllocateNeyman, optionally with a minimum for every stratum); - what the reviewed scores say about the whole population — a stratified, weighted estimate with a normal and a
bootstrap interval (
ReviewEstimator.Estimate), and a band for items the review could not settle (ReviewEstimator.EstimateBounds); - how far two reviewers agree — Cohen's κ, Gwet's AC1 and the agreement specific to each outcome
(
ReviewAgreement.Compute).
Every draw uses one generator, so the same seed gives the same sample on every platform. Record ids are ordered by their
UTF-8 bytes. Nothing is clamped or silently repaired: an assumption that does not hold is reported in Caveats, and an
input that would estimate another population throws ArgumentException.
Scores of 0 or 1 from a system that is right most of the time — a stratum whose reviewed items are all right is
then the usual case, not an exception, and the estimate reports EstimateCaveats.NoVariationInStratum. With four strata of
30 reviewed items each, it does so for about 85% of estimates at a precision of 0.97 in every stratum, and for almost
all of them at 0.99 (one stratum of n items is constant with probability pⁿ + (1 − p)ⁿ). Such a stratum's variance
takes the bound for the variation the sample has not shown, and its bootstrap collapses, so read the normal interval.
IsReliable is then false, so for 0/1 scores do not use it as a go/no-go — look at the other caveats one by one.
using Gil.Review;
var sample = ReviewSampler.DrawPilot(
[new PopulationStratum("answered", ["d1", "d2", "d3"]), new PopulationStratum("abstained", ["d4", "d5"])],
perStratum: 30,
seed: 7);
// Both strata are smaller than the pilot: taken whole.
Console.WriteLine(sample.SelectedCount); // 5
The package carries the native library for linux-x64, linux-arm64, linux-musl-x64, linux-musl-arm64, osx-x64, osx-arm64, win-x64 and win-arm64. A Native AOT application publishes it next to the executable.
Licensed under Apache-2.0.
| Product | Versions Compatible and additional computed target framework versions. |
|---|---|
| .NET | net10.0 is compatible. net10.0-android was computed. net10.0-browser was computed. net10.0-ios was computed. net10.0-maccatalyst was computed. net10.0-macos was computed. net10.0-tvos was computed. net10.0-windows was computed. |
-
net10.0
- No dependencies.
NuGet packages (1)
Showing the top 1 NuGet packages that depend on Gil.Review:
| Package | Downloads |
|---|---|
|
Eyu.Core
Eyu core contract — source-agnostic ontology inference ports (IStructureSource, IRecordSample, IOntologyProposer, IGroundingContract, IModelClient). |
GitHub repositories
This package is not used by any popular GitHub repositories.