praxis

Pull a real GW event from GWOSC and produce a full Siren brief — what it taught us, what it's still hiding, and what's next.

Run the full scientific-method loop on a question, dataset, or claim.

Writes and runs reproducible scientific analysis code. Use to fetch data, run parameter estimation or simulation-based inference, make diagnostic plots, and verify numbers. Handles the GW stack (gwosc, gwpy, bilby, pycbc) and the SBI stack (sbi, lampe, jax/torch).

Maps a result in one field into what it constrains or enables in adjacent fields. Use when an investigation's finding could matter beyond its home domain — e.g. a GW event bearing on cosmology, the nuclear equation of state, fundamental physics, or multimessenger astronomy.

The gravitational-wave domain authority. Use for anything requiring correct GW physics: choosing waveform models and priors, interpreting parameter-estimation posteriors, detector and noise behavior, source classification (BBH/BNS/NSBH), population inference, and GW conventions. The core of the Siren pack.

Turns numbers into physical meaning. Use after results are produced (and ideally after the skeptic) to say what physically follows, what does not, and how strongly. Strictly separates measurement from inference from speculation.

Reads scientific papers and data releases (arXiv, ADS, journal pages, GWOSC docs) and extracts a structured, citation-grounded summary. Use to establish background for an investigation, to evaluate a claim against prior work, or whenever the user shares a paper/preprint.

How to fetch real gravitational-wave data from the LVK / GWOSC public archive — event catalogs, parameter-estimation summaries, and strain. Use whenever an investigation needs actual numbers for a GW event instead of recalled values. Endpoints are public, no-auth, and CORS-enabled.

How to read a scientific paper or preprint rigorously and extract a structured, citation-grounded brief — claim, method, assumptions, limitations, dependencies. Use when establishing background, evaluating a claim, or whenever the user shares a paper, arXiv id, or DOI.

The master workflow for data-driven natural science. Trigger whenever the user wants to investigate a phenomenon, analyze a dataset, evaluate a result, reproduce or extend a paper, or asks "what does this data tell us / what should we do next". Routes the work through the scientist roles (reader, domain expert, analyst-coder, interpreter, skeptic, teacher, next-steps planner) following an explicit, falsifiable loop instead of answering from memory.

When and how to use simulation-based inference (SBI / likelihood-free inference) instead of stochastic sampling, and how to validate it before any scientific use. Use for inference problems with an intractable or expensive likelihood, amortized inference over many events, or when classical sampling is too slow — and always before trusting an SBI posterior.