rang-core

the rang-core mark, an interferogram with one amber source

ra-ng aims to be the home for agentic radio astronomy, MeerKAT first. rang-core is its first package: flagging, calibration and imaging on one Rust core, driven from Python and one command line. The PyPI distribution and command are rang-core, the Python package is rang, and the imager is gemclean. “Rang” began as Radio Astronomy Next Generation; rang is Dutch for rank, the first thing the identifiability survey measures.

Important

Status, November 2026: a tested research demonstrator, not an operational calibrator. No novelty is claimed: the goal is better calibration and imaging at comparable cost, or comparable quality at much lower cost, and neither has been shown. Two questions are closed as null, as tested: a gain-TV-guided threshold flagging policy (research/flagging-verdict.md), and beam and pointing (research/b1-3-pointing-selfcal.md). Three learned flaggers stopped or failed as frozen (The research story, section 7). The reference imager GeMClean is faster than WSClean on 16 threads, not always on one, at 2.0–2.4 times the memory on the real pilot with its current defaults (research/g7-defaults.md). The acceptance on real data (E2) has reported and misses some gates, as does native data (E2-N, 17 of 25 rows); the comparison with CARACal (E3) has measured E2-B; E2-A was still running at the release and follows in 0.3.1; Known limits lists what is known.

Where to start

You are

Start with

Then

New to radio astronomy

Learn radio interferometry with rang: the introductory notebooks I1–I9, from the radio sky to a CLEANed image, polarisation and pulsars, each under a minute on a laptop

From voltages to an image, in one page, the Glossary, and an agent in a checkout that answers from the notebooks and reports (Answer a beginner’s question)

An astronomer with a MeerKAT MS

Installation (pip install rang-core), then Quickstart: A1 small to a calibrated image with rang-core run in under a minute

How-to guides for astronomers: your own L-band MS, calibration, flagging, checking the result, and the Known limits

An algorithm scientist or developer

Developer guide: the Rust core and the Python layer, building, testing

Claims: propose, freeze, measure: the claims rule, oracles and frozen tolerances; Contributing for the pull-request route

What is in it

  • A Rust core, reached through the compiled extension rang._core: a Measurement Set reader and casacore table writer, zarr ingest with HTTP range fetch, QuartiCal’s full-Jones solver, CASA’s K, G, B, F and polarisation solves and applycal, Tricolour’s SumThreshold strategies, WSClean’s imaging weights and w-stacking gridder (CPU, and GPU in a source build), ducc0’s wgridder, CASA’s flux standards and a component predictor.

  • GeMClean, a reference imager: gemclean takes a Measurement Set to FITS with WSClean-style options, using the exact image-domain value at each minor-cycle step (research/gemclean.md).

  • One command line (Command line): rang-core flag, calibrate, apply, image, selfcal and run take a config.yaml and your Measurement Set, write back into it and resume where a run stopped; rang-core research STAGE VERB replays each research chain. Every call writes a versioned receipt.

  • The tools’ file formats: CASA calibration tables, QuartiCal stores, sky models, H5parm, killMS solutions, CASA flag versions and MVFv4 archives.

  • Oracles. Each port is gated on reproducing the upstream tool’s output, within a tolerance written before the run.

  • Notebooks: nine introductory ones, the MS-to-image tutorial, five on the chain on real data and two on the research questions (Tutorials); the research story keeps the negative results.