Gladius

Native image cutouts for the Roman Space Telescope.

Python import and command: gladius. Distribution name: roman-gladius (roman_gladius is the equivalent PyPI spelling).

Native detector cutouts from calibrated Roman images, with explicit parent-pixel mapping, separately retained parent metadata, exact stored astrometry, and optional display stacks. An independent alpha package built from the STScI Cookiecutter scaffold; it is not affiliated with or endorsed by STScI or Astropy.

Install and run

Python 3.11 or newer and Git are required. Gladius is not yet on PyPI; install from the public repository:

git clone https://github.com/AmberLee2427/gladius.git gladius
cd gladius
python -m venv .venv
source .venv/bin/activate
python -m pip install -e '.[roman,mast]'
gladius gui --workspace .

On Windows, activate with .venv\Scripts\activate instead. Open the printed GUI URL to add targets, choose images and processing, and download a YAML file or make cutouts. For the manual YAML workflow:

gladius init-config config.yaml
# Edit targets and inputs in config.yaml.
gladius validate config.yaml
gladius run config.yaml

For archive access, set inputs.mode: mast. Searches use your targets by default; mast_criteria adds optional archive constraints. Restricted data use MAST_API_TOKEN from the environment. Directory mode does not perform archive queries or downloads.

Try a complete offline example with no credentials or mission data:

gladius demo /tmp/my-gladius-demo

This creates two synthetic ASDF exposures, a YAML file, native ASDF/FITS cutouts, a display stack, an exposure GIF, a resolved configuration and a run manifest. Choose a new demo folder; existing demo inputs are not overwritten.

Or, with the roman and mast extras installed, try a real archive image. First sign in at the MAST token page, create an API token, and copy it for the hidden prompt below. The token identifies your account; restricted observations still require permission for that account:

gladius token             # Hidden prompt; saves MAST_API_TOKEN in .env here.
gladius demo mast-demo    # Queries MAST, downloads one parent, and makes cutouts.

The MAST demo writes mast-demo/demo.yaml and ASDF/FITS cutouts under mast-demo/output/. It rejects parents advertised above 512 MiB before download; use --max-parent-mib to change that limit. It uses an explicit 0.01-pixel WCS round-trip tolerance and makes no display stack. See the real-data quickstart.

Configuration

examples/config.yaml documents every option. Paths are relative to the configuration file. Unknown keys, duplicate YAML keys, invalid coordinates and inconsistent settings are rejected before input discovery. gladius config-schema schema.json writes JSON Schema for editor support.

• Select exposure-start dates in a UTC [start, end) interval, detectors 1–18, and filters. Empty detector/filter lists mean all; missing metadata needed for a requested selection causes an explicit error.
• Choose square cutout size, padding/strict edges and fully off-detector policy.
• Use supplied WCS, validate its numerical consistency, or require a separately supplied corrected GWCS. Validation does not establish absolute sky accuracy.
• Enable an autocorrelation elongation screen and decide whether failures enter display stacks. Native cutouts remain available regardless of quality flags.
• Retain full parent images or keep only their derived products. Local input files are never deleted. Archive parents are staged one at a time.
• Create median display stacks, exposure GIFs and configurable false-colour composites.
• Choose ASDF, FITS or both; configure bounded display stacks, resume and failure handling. Resolved defaults and dependency versions are recorded for each run.

See configuration, outputs, scientific conventions, and Python API.

Scientific contract

SCI and ERR are unresampled floating-point pixels with NaN padding. DQ flags stay integer; missing pixels are DO_NOT_USE (1), and existing saturation flags survive. Centering follows the commissioning convention floor(position) - (size - 1)//2, including odd sizes. Coordinates are zero-based (x, y) on the calibrated SCI array, with array[y, x] indexing. Origins can be negative.

The cutout WCS is a translation of the chosen parent WCS. The original WCS and metadata are retained under parent_wcs and parent_metadata; derived footprints are in cutout_metadata. Spatial arrays are cropped using the shared mapping, with descriptors recording their original paths, shapes and types. Non-spatial metadata arrays and custom fields are retained, including FITS header ordering, comments and duplicate HISTORY cards.

ASDF stores the exact GWCS. FITS provides SCI/ERR/DQ extensions, a measured TAN-SIP approximation where necessary, auxiliary arrays and a package-specific ASDFMETA extension retaining full metadata and exact WCS. These are derived products, not schema-valid Roman mission datamodels. Read them with read_cutout.

Current boundaries

The first version handles 2D calibrated detector images, not raw ramps, mosaics, spectral products or moving-target ephemerides. A supplied correction can be applied, but the separately authored Gaia solver is not bundled. The new independent autocorrelation method is provisional; its threshold must be calibrated, and the old commissioning threshold is not assumed equivalent. Stacks are resampled display products, not photometrically or PSF-matched science coadds. They do not propagate uncertainties. Execution is sequential and portable.

MAST staging is covered by mocked service tests; real archive access depends on service availability, permissions and input metadata. See TESTING.md for actual validation scope. No automatic upload, publication or external solver invocation occurs.

Development and provenance

python -m pip install -e '.[dev,roman,mast]'
python -m pytest
python -m ruff check src tests docs/conf.py
python -m sphinx -W -b html docs docs/_build/html
python -m build

Historical source snapshots are local reference material and are excluded from distributions. TEMPLATE.json records the Cookiecutter template and exact commit; NOTICE.md explains license scope and attribution. The copied external Gaia solver has not been incorporated into the package.

New package code is BSD-3-Clause. Template attribution is retained separately. See contributing, conduct, AI disclosure, and citation. The project has not yet been published to PyPI or assigned a DOI.

Local web GUI and target CSVs

The separate Nexus notebook queries real MAST data and requires an explicit path to your credential file. It is not part of the docs site. Install its plotting/kernel dependencies with pip install -e '.[roman,mast,notebook]'.

For restricted MAST access, run gladius token in your workspace to save the token through a hidden prompt. Alternatively copy .env.example to .env and set MAST_API_TOKEN=your_token_here. The GUI loads that file when the server starts; restart the server after editing it. Existing shell variables take precedence. For gladius run config.yaml, place .env beside the YAML instead. Use --env-file /path/to/credentials.env with token, run, gui, or the MAST demo to select a different credential file. Keep .env private and out of Git; the repository already ignores it. GitHub Actions secrets do not populate this local file. See the authentication instructions.

gladius gui --workspace . --port 8768

Open the printed local URL. The GUI builds the same validated configuration used by the CLI, imports target CSVs, downloads self-contained YAML, and runs the pipeline with live progress. It includes a YAML editor/importer for existing configurations. Relative paths resolve from the directory shown in the builder; downloaded YAML contains absolute paths. The server binds only to loopback. Keep its terminal open during runs; closing a browser tab does not cancel a run.

gladius validate config.yaml --targets targets.csv
gladius run config.yaml --targets targets.csv

--targets replaces YAML targets before validation, so the YAML may omit targets when that option is supplied. The CSV path is relative to the CLI working directory; other paths retain their YAML-relative meaning.

id,ra,dec,catalogue
OGLE-2019-BLG-0011,270.123,-29.456,OGLE

Use decimal ICRS degrees (ra,dec) or zero-based pixels (x,y), never both. ra_deg/dec_deg and name/target_id aliases are supported. IDs can be omitted to generate stable row-order IDs such as target-000001. Extra columns are retained as strings under each target's metadata, including in cutout provenance. Invalid rows and duplicate IDs fail explicitly. The GUI preview shows the first 50 targets; all rows are included in the configuration and run.

Each GUI submission saves its resolved YAML and progress record in .gladius-gui/ under the workspace; scientific outputs still go to the configured output directory. Only one GUI-owned run is active at a time. Reloading the page reconnects to the last run while the server remains open. After a server restart, run history remains on disk rather than being automatically restarted.

On Nexus, start the GUI in a JupyterLab terminal with gladius gui --workspace /path/on/nexus/to/work --nexus and open the printed proxy link. Gladius detects a supplied JupyterHub public URL automatically; --public-url provides an explicit override. See the Nexus and SSH instructions.

Python API

The CLI and GUI use the same importable functions:

from pathlib import Path

from gladius import RunConfig, load_config, load_environment, read_targets_csv, run

# Folder containing this example's YAML, targets, and optional .env file.
workspace = Path.cwd()

# Python callers explicitly load credentials for restricted MAST access.
# Reads workspace/.env if present; existing environment variables take precedence.
load_environment(workspace)

# Existing YAML, with an optional CSV override: returns a validated RunConfig.
config = load_config(workspace / "config.yaml", targets_path=workspace / "targets.csv")
manifest_path = run(config, progress=print)

# Or construct the configuration directly:
config = RunConfig(
    targets=read_targets_csv(workspace / "targets.csv"),
    inputs={"mode": "mast"},
    extraction={"size": 150, "no_overlap": "skip"},
).resolved(workspace)
# manifest_path = run(config)

MAST sky searches are generated from target RA/Dec, with Level 2 calibrated images and a configurable 6 arcminute candidate radius by default. Additional criteria are optional for sky targets; pixel-only targets need an archive constraint such as a program ID. See the configuration guide for authentication, query limits and overlap conventions. make_cutout() also accepts in-memory SCI/ERR/DQ arrays, a WCS and a Target; see the Python API documentation.

Live archive tests in GitHub Actions

Add MAST_API_TOKEN under Settings → Secrets and variables → Actions. The Live Roman MAST download workflow runs on default-branch code changes, or manually from Actions. It downloads one product capped at 20 MiB and reports whether it tested a calibrated image/cutout or a small associated product. The manual calibrated-image option requires full extraction coverage within the same cap. Downloaded science data are not uploaded as artifacts. Normal tests remain offline. See the configuration guide for the exact coverage and limits.

Publishing releases

Pushing a matching v<version> tag runs the checked PyPI release workflow. Store the PyPI upload token as the repository Actions secret PYPI_API_TOKEN. See release instructions before pushing a tag.