REFUTE is a scientific critique and calibration benchmark for large language models. Paper-grounded claims are turned into model predictions, judged, and scored with Brier score and expected calibration error (ECE). An open Inspect AI task and interim preprint release are available.

AEGIS-SENTRY is an open-source orbital mechanics engine designed for real-time planetary defense modeling and Near-Earth Object (NEO) impact risk assessment. The engine dynamically ingests telemetry feeds from both NASA CNEOS Sentry-II (Impact Observation Sampling) and ESA NEOCC (Line of Variations) to evaluate risk divergence across 1-year, 10-year, and 50-year projection horizons. Key orbital capabilities include native 3D Keplerian state vector propagation, analytical Vokrouhlický diurnal Yarkovsky thermal drift modeling (da/dt), and b-plane gravitational keyhole proximity detection. Additionally, the software features dynamic mitigation solvers calibrated for Kinetic Impactor (utilizing DART empirical beta momentum enhancement factors), Gravity Tractor continuous force vectors, and Nuclear Standoff scenarios. Built as a zero-dependency architecture with native SSE streaming and OpenAPI 3.1 specifications, AEGIS-SENTRY provides an independent, accessible platform for planetary defense research and orbital perturbation modeling.

a0kit computes the relations of the de Sitter-Unruh modified-inertia framework for the MOND acceleration scale a0: the horizon-derived scale a0 = c H_Lambda / Z = c^2 sqrt(Lambda/32pi) (both the pure-dark-energy "canonical" and total-density "alt" footings), the Lambdaa0 and rho_DEa0 inversions (Lambda = 32pi a0^2/c^4), the interpolation kernel nu(y)=sqrt(1+1/y) (the functional form of Milgrom 1999), the a0-line identity g_obs^2 - g_bar^2 = a0 g_bar and its single-point inverse, the redshift law a0(z)/a0(0) = sqrt(rho_DE(z)/rho_DE0) under a CPL dark-energy history, and the deep-MOND baryonic Tully-Fisher flat velocity. Written in Python with numpy as the only dependency, the code validates its formulas with a self-test and a test suite, and produced and verified the numerical results reported in the framework's papers. a0kit computes the framework's relations but does not derive them: the value of a0, the coefficient Z = sqrt(32pi/3), and the sign of the inertial correction are posited, and the interpolation kernel is Milgrom's 1999 functional form. Both a0 footings are provided because which cosmic density sets the scale is not settled by the framework.

GalaxyPose is a Python toolkit that models galaxy trajectories and orientations as continuous functions of time from discrete cosmological simulation snapshots. It supports periodic-box-aware trajectory interpolation and quaternion-based orientation interpolation from rotation matrices or angular-momentum directions. These continuous models enable aligning stellar particle birth properties to the host-galaxy frame at formation time.

Numerical verification suite for Galileon-braided cyclic cosmology. 17 self-contained Python scripts proving spectral tilt (n_s = 0.961), bounce stability, super-horizon mode conservation, and cycle closure. Uses SciPy DOP853 integrator to solve Mukhanov-Sasaki and coupled Horndeski background equations during matter-dominated contraction (H < 0). Requires only NumPy and SciPy. Each script proves one claim from the accompanying paper (DOI: 10.5281/zenodo.21419162).

Browser-based suite of 104 interactive calculators for the Omega
Centauri (NGC 5139) intermediate-mass black hole debate, Fermi
Paradox analysis, and Macro Transcension Hypothesis. Tools cover
stellar kinematics (M-σ estimator, velocity dispersion), pulsar
acceleration constraints using TRAPUM 2026 data, Gaia DR4 precision
forecasting, ADAF/RIAF accretion spectral energy distributions with
JWST cross-checks, gravitational-wave horizon plotting (LISA, ET,
CE, PTA), Bayesian evidence aggregation, SETI sensitivity (radio
and IR excess), cosmological distances, and Kardashev-scale
engineering tools. All tools are hash-addressable (permalink-stable
state), browser-only (no backend, no PII), and 31 are exposed via a
Model Context Protocol (MCP) server (41 chained multi-tool workflows).
Curated measurement data is CC0; code is MIT; prose is CC BY 4.0.

Visualization of high-dimensional data is crucial to retrieve all the knowledge that is contained within a dataset. Effective and informative presentation of three-dimensional data via a two-dimensional medium is challenging, especially if the dataset more closely resembles a 2.5-dimensional (2.5D) entity due to sparse sampling. We present LStein (Linking Series to envision information neatly), a novel visualization approach implemented in Python, in an attempt to solve this challenge. Inspired by the astrophysical application of displaying photometric timeseries in multiple passbands with minimal loss of information, we compare our method to traditional approaches. While astronomy – specifically multi-passband visualization for lightcurves obtained with the Rubin Observatory – serves as the principal driver for the design, we demonstrate that LStein can be used in any context with 2.5D datasets from radio astronomy to machine learning hyperparameter search visualization. LStein provides a complementary visualization to traditional techniques. LStein can be installed from GitHub (https://github.com/TheRedElement/LStein).

The toolkit Radio Beam manipulates two-dimensional Gaussian beams associated with radio astronomical data. It extracts beam parameters from FITS headers, creates and edits beam definitions within the Astropy (ascl:1304.002) framework, and performs operations such as convolution, deconvolution, and brightness–temperature unit conversion using the beam area. The package also handles sets of beams for spectral cubes with channel-dependent resolution, identifies smallest common beams in a collection, and overlays beam shapes on Matplotlib plots for visualization.

redrock fits redshifts for spectroscopic data using a spectroperfectionism-based template-fitting approach developed for DESI spectra. Its rrdesi command-line tool analyzes input spectra with standard template sets or with archetype templates constructed from physical spectra combined with Legendre polynomials, solving for template and polynomial coefficients via bounded least-squares at a set of trial redshifts. An additional mode refines fits by selecting nearest-neighbor archetypes in chi-squared space and recombining them with Legendre polynomials, enabling flexible template construction for improved redshift estimation.

ELFO (Emission Line Fitting Optimization) improves emission-line fitting in integral-field spectroscopy data by exploiting spatial correlations between neighboring spectra. It wraps PyQSOFit (ascl:1809.008) to fit each spectrum individually, setting initial parameters from the fits of adjacent spaxels and selecting the most spatially smooth solution from multiple fitting orders. Designed for CSST-IFS data and validated on simulated CSST observations, ELFO enhances Hα emission-line fits in quasar spectra and can be adapted with minor changes to other emission lines and IFS datasets.