moon-survival

Survival analysis toolkit for MoonBit, with native support for right-censored data. It is designed for medical follow-up, reliability engineering, software failure analysis, and user churn studies.

Project positioning

moon-survival provides reusable statistical building blocks for estimating time-to-event behavior across WebAssembly, Native, and JavaScript targets. The library is dependency-light and keeps data validation, estimators, model diagnostics, reporting, and reproducible simulation fixtures in one MoonBit module.

Core capabilities

• Right-censored and weighted survival observations with validation and summary statistics.
• Kaplan-Meier with Greenwood standard errors, confidence intervals, quantiles, median survival, restricted mean survival time, and curve export.
• Nelson-Aalen cumulative hazards with variance, confidence intervals, and survival transformations.
• Cox proportional hazards with Newton fitting, ridge stabilization, Breslow baseline hazards, predictions, residuals, and proportional-hazards diagnostics.
• Competing risks with Aalen-Johansen cumulative incidence, log-rank tests, pairwise comparisons, and RMST differences.
• Reliability models, power planning, bootstrap/jackknife resampling, grouped analysis, model selection, and calibration metrics.
• Parametric and AFT models, interval-censored Turnbull fitting, IPCW analysis, recurrent events, multi-state transitions, validation pipelines, and missing data workflows.
• Risk stratification, calibration, feature engineering, decision analysis, clinical endpoints, audit trails, reproducible exports, and trial operations.
• CSV ingestion/export, time-varying covariates, incremental accumulation, deterministic simulation, Markdown/CSV reports, ASCII visualization, and finite-state Markov cohort analysis.

Quick start

moon add Lyl66655/moon-survival
moon run --target native cmd/example

The example executable demonstrates Kaplan-Meier, Nelson-Aalen, Cox, and competing-risk workflows. A package can import the library package as:

import {
  "Lyl66655/moon-survival/src",
}

CLI and runnable commands

The repository provides two executable MoonBit entrypoints:

# Run the end-to-end example
moon run --target native cmd/example

# Run the deterministic performance fixture
moon run --target native --release cmd/benchmark

The benchmark command is the project CLI for repeatable local performance checks; the library itself is consumed through MoonBit package APIs.

Architecture

moon.mod                         module metadata and package version
src/moon.pkg                     public survival-analysis package
src/*.mbt                        estimators, models, diagnostics, reports
src/pkg.generated.mbti           generated public API summary
cmd/example/                     runnable API example
cmd/benchmark/                   deterministic benchmark executable
.github/workflows/check.yml      cross-platform quality checks
LICENSE                           Apache-2.0 license text

The source package is organized by responsibility: data and numerical utilities are the foundation; non-parametric estimators, regression models, reliability and competing-risk models build on them; diagnostics, reporting, visualization, and command entrypoints consume those public APIs.

Benchmark

The deterministic fixture contains 5,000 observations, 4,403 events, and 11.94% censoring. The following ranges were observed across three consecutive native release runs on 2026-08-20 with MoonBit 0.1.20260814 and moonc 0.10.8; timings vary by host:

Workload	Result
Kaplan-Meier, 5,000 observations	5,000 steps, 1–3 ms
Nelson-Aalen, 5,000 observations	5,000 steps, 0–1 ms
Cox, 500 rows and 3 covariates	4 iterations, converged, 54–58 ms
Exponential parametric fit, 5,000 observations	0–1 ms
IPCW weights and curve, 5,000 observations	5,000 steps, 28–32 ms
Two-group KM summary, 5,000 observations per group	2 curves, 2–6 ms

The fixture, event counts, censoring fraction, convergence status, effective sample size (4,951.7646), group survival at 30 (0.0034694), and RMST at 30 (7.4435516) are deterministic. Timings depend on the host machine.

Tests

moon fmt --check
moon check --target all --deny-warn
moon test --target wasm-gc --deny-warn
moon test --target native --deny-warn
moon info

The regression suite contains 119 passing tests covering empty data, all-censored data, ties, invalid events, malformed matrices, singular systems, CSV quoting, numerical extremes, model edge cases, interval and right censoring, IPCW, recurrent and multi-state data, validation, export, feature engineering, and deterministic performance fixtures.

CI

GitHub Actions runs on Ubuntu, macOS, and Windows. Each job installs the latest stable MoonBit toolchain from the official installer and runs formatting, all-target compilation checks, metadata generation, generated-source drift detection, WebAssembly GC tests, Native tests, JavaScript tests, and all-target builds.

License

Apache-2.0. See LICENSE.

Resources

• GitHub repository
• GitLink mirror
• Mooncakes package: Lyl66655/moon-survival