RMC-BestFit v2.0.1

RMC-BestFit v2.0.1 updates the desktop application, headless API, and RMC.BestFit model library. It uses RMC.Numerics 2.2.0 and WPF Framework 1.0.5 and includes scientific corrections and reliability improvements.

Highlights

• Flood-frequency analysis: restores magnitude-ordered Hirsch–Stedinger plotting positions; improves low-outlier processing, seasonal point-process annualization and simulation, mixture identification, and composite/coincident-frequency uncertainty sampling.
• Bulletin 17C: improves bootstrap diagnostics and optimizer failure handling, preserves point estimates when covariance computation fails, and restores distribution-aware GMM delta-method quantile-variance reporting with explicit units and unavailable-result messages.
• Estimation and uncertainty: corrects information criteria, profile likelihoods, PSIS-LOO diagnostics, covariance failure reporting, rating-curve likelihoods, and spatial GEV likelihood, prediction, cross-validation, and uncertainty handling.
• Time series: corrects transform fitting, differencing, ARIMAX conditioning, simulation, and forecast reconstruction; derives default training windows from model requirements and preserves fitted parameters, bounds, and priors when restoring or editing analyses.
• Desktop reliability: repairs mixture row editing, analysis refresh, source-data invalidation, POT diagnostics, theme-change selections, SQLite connection disposal, and HEC-DSS imports across missing storage blocks.
• API and assisted workflows: adds source provenance, input chronology, plot-source export, and explicit low-outlier controls. REST and MCP requests now enforce browser-origin restrictions.
• Documentation: expands source-backed example guides, Python plotting support, technical-reference and verification-report drafts, and traceable verification records. The NuGet package now includes IntelliSense XML documentation.

Desktop installation and dependencies

Download RMC-BestFit.v2.0.1.zip, extract the complete archive, and run RMC-BestFit.exe. Keep its libraries/ and runtimes/ folders together. The desktop requires Windows x64 and the .NET 10 Desktop Runtime x64.

The archive includes the published RMC.Numerics 2.2.0 and RMC.Wpf.Framework Core, Models, Controls and Support 1.0.5 assemblies, plus the managed/native HEC-DSS and SQLite payloads. The separate RMC.BestFit 2.0.1 NuGet package depends on Numerics 2.2.0; release publication starts its NuGet publishing workflow.

bestfit-frequency packages

The release packaging workflow produces four distributions of the maintained bestfit-frequency skill: standalone, OpenAI marketplace, direct OpenAI plugin, and Claude plugin, each with a SHA-256 sidecar.

Plugin version 0.3.2 is separate from application/API version 2.0.1. These packages require a terminal-capable session with .NET 10, Python, dependency access, and a running local API. They do not contain application binaries or deploy a hosted service. Client installation and directory publication are separate steps.

See the installation guide.

Upgrading

• Teams sharing projects should upgrade together. Version 2.0.0 can display newer mixture results but cannot reliably recompute them and can reset fitted ARIMAX state.
• Re-run time-series analyses that display the pre-v2.0.1 results or incompatible-parameters warning. Check the priors of covariate models previously reopened, edited, and saved in 2.0.0.
• Saved analyses are not automatically re-estimated. Re-run or reprocess affected analyses to adopt corrected plotting positions, information criteria, uncertainty sampling, and other numerical changes.
• Keep a backup of older projects, particularly Bulletin 17C analyses saved before 2.0 Beta-4.

Report snapshots

The attached technical reference and verification report retain their reviewed 27 September 2026 source checkpoints. Their original metadata and PDF bytes are preserved; their report date is not a claim that all numerical studies were rerun for this release. Current source documentation and detailed release notes include subsequent corrections. SHA256SUMS covers the attached desktop ZIP, model package and reports.

Desktop download verification

Built from source commit 0d478ef05676e802574a2bb4e0ae7d58426f6f5d.

SHA256: 300d30a3ed7f58fd67c5b1aea1a135796ca1e58bc1d10d3e6882c16945e6e67c