LoginSign Up
← Back to all sparks
R

reliagrowr

ANALYTICS
Velocity0.0

R package reliagrowr by paulgovan — release notes from GitHub.

github.com/paulgovan/reliagrowr ↗

A reliability growth package put its models behind an MCP server for AI assistants to call.

reliability-engineeringr-packagerepairable-systemsmcpai-toolingstatistical-modeling
◆Current state
ReliaGrowR fits reliability growth models to failure data — Crow-AMSAA and Duane, with maximum likelihood estimation, confidence bounds, prediction, and reliability demonstration test planning. The last year widened it well past growth curves into repairable systems: parametric non-homogeneous Poisson process fitting with automatic change point detection, non-parametric mean cumulative function estimation, and system exposure calculation. The most recent release adds goodness-of-fit statistics and exposes the package's functions as Model Context Protocol tools.
◆Where it's heading
Two arcs run in parallel. The statistical one is a steady march from plotting a growth curve to modelling recurrent failures properly — segmented NHPP models that detect their own change points, Nelson-Aalen estimation, Cramér-von Mises and Kolmogorov-Smirnov statistics for judging the fits. The interface one is newer and more unusual: the package now ships an MCP server, and its sibling plotting package followed with one two weeks later, so this is a deliberate direction across the maintainer's reliability suite rather than a single experiment. Naming and S3 conventions were cleaned up early, which is what made a uniform tool surface plausible later.
◆Prediction
Given the sibling packages moved to MCP within weeks of each other, the remaining tools in the suite are the obvious next candidates; on the statistical side, goodness-of-fit having just arrived suggests model comparison and selection helpers are the natural follow-on.

◆Recent moves

  1. 2mo ago

    Reliability growth models exposed as MCP tools

    ⚡ SPARK

    rga_mcp_server() makes the package's models callable by AI assistants over the Model Context Protocol — a new class of caller for a domain statistics package. Shipped alongside gof() for Cramér-von Mises and Kolmogorov-Smirnov statistics and predict_duane() with confidence bounds, so the same release adds both a way to judge fits and a way for something other than a human to request them.

    View source ↗
  2. 4mo ago

    Repairable systems analysis arrives: NHPP, MCF, exposure

    The largest statistical expansion in the log, and the one that stops this being a growth-curve package: parametric NHPP fitting in Power Law and Log-Linear forms with piecewise support and automatic change point detection, Nelson-Aalen mean cumulative function estimation, and system exposure across repairable systems. Overlay functions let multiple fits share a plot, which is the practical need once several models compete for the same data.

    View source ↗
  3. 4mo ago

    Maximum likelihood fitting and failure simulation

    An MLE fit method for rga() gives an alternative to the regression-based estimates the package started with, and sim_failures() supports simulation study workflows. Confidence intervals for the Duane model and a failure-allowed parameter for demonstration test planning round it out — incremental statistical depth ahead of the bigger repairable-systems push.

    View source ↗
  4. 8mo ago

    ReliaGrowR 0.3.2

    This release publishes release notes byte-identical to the 0.3 tag from a month earlier, listing the same plotting and printing options and the same data management vignette. Nothing in the feed distinguishes it from its predecessor, so it reads as a patch shipped without its own notes.

    View source ↗
  5. 9mo ago

    More plotting and printing options for RGA and Duane models

    Presentation work on the two core model types plus a vignette on reliability data management. Modest, but this is the release whose notes the following 0.3.2 tag reused verbatim, so it is the one that carries the actual changes.

    View source ↗
  6. 10mo ago

    S3 methods replace the ad hoc plotting functions

    duane_plot() becomes duane() with a separate S3 plot method and plot_rga() is replaced the same way — a breaking but conventional cleanup that gives the package the uniform object-and-method structure the later tidy extractors and tool servers depend on. P-P and Q-Q goodness-of-fit plots and demonstration test planning arrive alongside.

    View source ↗