Skip to main content

nd_prepare

Repackages seven external nuclear data libraries as a local repository for nd_manager, the nuclear data package manager from Los Alamos National Laboratory (LANL). It is a separate companion tool, written against the Python standard library only: run it yourself, then point an installed nd_manager at the result.

Libraries

Each library yields up to four entries, named <NAME>_<CATEGORY>:

Library Key NEUTRON TSL PROTON DOSIMETRY
FENDL-3.2c FENDL32C ✓ —¹ ✓ —
FENDL-3.1d FENDL31D ✓ —¹ ✓ —
FENDL-2.1 FENDL21 ✓ —¹ — —
D1SDATA D1SDATA ✓² — — —
IRDFF-II IRDFF2 — — — ✓³
JEFF-3.3 JEFF33 ✓ ✓ ✓ —
JEFF-4.0 JEFF40 ✓ ✓⁴ ✓ —
  1. Neither FENDL-3 release publishes an S(α,β) sublibrary (checked both IAEA pages and the IAEA-NDS/FENDL-Processed repo), and FENDL-2.1 lists only neutron ACE data, so FENDL is transport-only.
  2. FENDL-3.1c-derived activation library for D1S shutdown dose rate. It is not published online: the zip must be supplied locally (see Source entries).
  3. Response functions for foil activation, not transport data.
  4. A selected subset of the TSL data: Be metal, D in D₂O, C in graphite, H in CH₂/H₂O/ZrH/ZrH1.5/ZrH2, O and U in UO₂, Zr in ZrH1.5/ZrH2. The full set is available from OECD-NEA on request.

To add a library, see CONTRIBUTING.md.

Installation

Requires Python 3.9 or later, curl on the PATH (used for every download; Windows 10 and later ships curl.exe), and an nd_manager installed separately. Only Linux is supported; it should work on Windows and macOS too, but that is untested. On Windows, keep cache/ and work/ in a short path: some JEFF file names are long and the 260-character limit applies.

pip install .        # or: pip install -e .  to edit the code or sources.json in place

This installs the nd-prepare command (also runnable as python3 -m nd_prepare), with sources.json bundled as package data. --manifest defaults to that bundled copy, not one in the current directory; without -e, edits to the checkout's sources.json need a reinstall. cache/, output/ and work/ are always relative to the current directory.

Usage

nd-prepare --all                # every library (a bare nd-prepare is an error)
nd-prepare --list               # keys and categories accepted by --only
nd-prepare --only FENDL31D
nd-prepare --only JEFF40:neutron JEFF40:tsl
nd-prepare --only IRDFF2:dosimetry

Archives and libraries.json go to output/. Downloads are cached in cache/, so re-runs fetch only what is missing. An unknown key or category is an error.

D1SDATA is built from local files, so a run that selects it (including --all or --download-only) stops with "local archive not found" unless its zips are in the current directory. Without them, select the other libraries: nd-prepare --only FENDL32C FENDL31D FENDL21 IRDFF2 JEFF33 JEFF40.

Option Effect
--download-only Fill cache/ and exit; a later run builds offline from cache.
--cleanup Once each entry is packed, delete its extracted payload and its cached downloads.
--cleanup-work Once each entry is packed, delete only its extracted payload; keep cache/.
--compression xz .tar.xz (smaller, slower) instead of .tar.gz.
--prune Remove libraries.json entries the manifest no longer produces, delete their archives, and exit.
--workdir DIR Extraction scratch directory (default ./work).
--manifest FILE Use another manifest.

Compression levels are low (gzip 6, xz preset 1): ACE data is dense ASCII and gains little from higher levels. Rebuilding an entry with the other compression points libraries.json at the new archive but leaves the old one in output/; --prune does not remove it, so delete it by hand.

Register the result with nd_manager:

nd_manager config --add-repo ExternalND /path/to/output/libraries.json
nd_manager update
nd_manager list

Output layout

Inside each archive, everything except xsdir, README.md and docs/ sits under <NAME>_<CATEGORY>/, like LANL's own libraries (kidman.tgz, endf66.tgz). Each installed library therefore gets its own data directory, and filenames from different libraries cannot collide.

libraries.json accumulates across runs, so separate --only runs build one repository. As a consequence, entries for removed or renamed libraries stay listed (and offered by nd_manager) until you run --prune. Every build run warns about them (--download-only does not). Entries from another --manifest or added by hand also count as stale, so check the warning before pruning.

Size and licensing

A full run (with the D1SDATA zips in place) downloads roughly 24-26 GB, mostly JEFF; D1SDATA's 425 MB is local and not part of that. The main items:

  • JEFF-4.0 and JEFF-3.3 transport archives: about 1.3 GB each, 8 and 6 of them
  • JEFF-4.0 TSL archive: about 4.4 GB
  • four proton archives: under 1 GB combined
  • D1SDATA: about 425 MB
  • IRDFF-II: about 21 MB

Plan for at least 60 GB free for a full run (downloads, extracted payloads and archives); --cleanup-work lowers the peak. Use --only to download less.

Library License
JEFF-4.0, JEFF-3.3 CC-BY-4.0 (checked in the metadata on data.oecd-nea.org).
FENDL, IRDFF-II Only IAEA's general website terms; no explicit open-data license was found. Review them before sharing the archives.
D1SDATA Confidential. No published license; do not share the data or archives built from it outside your organization.

Data distribution

Every url source in sources.json is openly published (IAEA, OECD-NEA open records). Data obtained on request or under specific terms, such as the full JEFF-4.0 TSL set or D1SDATA, enters only through path sources and is never committed. D1SDATA is confidential: do not commit, publish or redistribute it, the archives built from it, or a libraries.json that serves them outside your organization.

Caveats when combining libraries

Duplicate table names

Each run warns when a table name is defined more than once, within or across the processed libraries. Installing such libraries together gives the merged xsdir duplicate entries, and MCNP® uses the first match.

ZAID conventions

ZAIDs for metastable states differ between libraries, and for Am-242 the conventions conflict:

Library Metastable rule Am-242 (ground) Am-242m
LANL libraries — 95642 95242
JEFF-3.3 MCNP: A + 300 + 100*m 95242 95642
JEFF-4.0 300 + 100*(m-1) + (A mod 100) 95242 95342

FENDL, D1SDATA and IRDFF-II have neither nuclide. nd_prepare neither renumbers nor checks these ZAIDs. MCNP reads the ZAID from the ACE header, so an xsdir-only rename would make MCNP fail, and rewriting the header would alter vendor data. When combining libraries:

  • do not mix JEFF-3.3 and JEFF-4.0 metastable nuclides;
  • do not trust the Am-242 ZAID alone: check the data file the xsdir entry points to (95-Am-242g-* or 95-Am-242m-*);
  • keep only one source for Am-242 in a given xsdir.

How it works

nd_manager install needs one top-level directory plus an xsdir. None of the providers ship that, so each library is rebuilt.

Transforms

  • Flatten vendor wrapper directories (fendl31d_ACE/, ace/, neutron/ace/, d1sdata/, ...) and rewrite the xsdir filenames to match (ready, fragments; merge-master strips each archive's top-level directory instead). Merged fragments (.xsd, .xdr, .dir) are deleted.
  • Strip the vendor header of master xsdirs (ready, merge-master): datapath=..., sometimes an atomic-weight-ratio table, then a directory line. A repository xsdir holds only entries. An atomic-weight-ratio table is saved as reference in docs/awr_<NAME>_<CATEGORY>, but nothing reads it. merge-master requires the directory line. ready also accepts an xsdir of entries only, but one with a header and no directory line is an error.
  • Drop entries whose file is missing from the payload (ready, merge-master). Dropped entries are logged by ZAID and do not fail the build unless none is left; some are expected (JEFF-4.0's xsdir lists far more TSL tables than are published as ACE).
  • Final pass on every xsdir, in all modes: reset every access route to 0, drop blank and control-character-only records, and end the file with a newline.
  • Line width (last step, after the <NAME>_<CATEGORY>/ filename prefix is added, which lengthens the lines): every entry is rebuilt as one logical line (its continuations joined) and word-wrapped again to at most 80 columns, the MCNP limit. An entry that fits is written on one line. A longer one is broken with a trailing + on each line but the last, and its continuation lines are indented by 10 spaces. A field that cannot fit in 80 columns fails the build.

xsdir entries are handled as whole records. A line ending in + continues onto the next (MCNP continuation format, used by FENDL-3.1d neutron), and dropping or rewriting an entry covers its continuation lines. A record has at least the 7 mandatory fields: ZAID, AWR, filename, access route, file type, address and table length.

Per-library fixes

Most of these were found by comparing each xsdir entry with its ACE header (table name, AWR, kT, NXS(1) table length).

Library / category Upstream issue Fix (manifest field)
FENDL-3.2c neutron, proton One data file and one .xsd fragment per isotope; no combined xsdir. Fragments are concatenated (fragments mode).
FENDL-3.1d neutron Two loose xsdir templates (_Linux.xsd, _Windows.xsd). _Linux.xsd is used; _Windows.xsd is deleted (drop_members).
FENDL-3.1d neutron Two K-39 files share the table name 19039.31c: 19K_039 (TENDL-2015) and 19K_039_ (TENDL-2010). 19K_039_ and its entry are discarded (discard_members).
FENDL-3.1d neutron The xsdir ends with a DOS end-of-file byte (0x1A). The byte is dropped in the final pass.
FENDL-3.1d proton Per-isotope fragments (.ace + .xdr); no combined xsdir. Fragments are concatenated (fragments mode).
FENDL-2.1 neutron One zip per isotope or natural element (71), each with an .ace and a one-entry .dir fragment; no combined xsdir. Fragments are concatenated from all archives (fragments mode).
FENDL-2.1 neutron Seven .dir fragments (He-3, He-4, Li-6, Li-7, Be-9, Cl-35, Cl-37) name an upper-case file (HE003.ace), but the zip holds He003.ace; on a case-sensitive file system the entries would be dropped. The xsdir filenames are rewritten to the real spelling (rename_xsdir_files); the ACE files are untouched.
JEFF-4.0 neutron One flat zip per temperature (8) plus a separate master xsdir. merge-master mode.
JEFF-4.0 neutron jeff4.xsdir (NEA record 0vhn5-d6t91) gives 14 table lengths for Cs-134 and Sm-150 (.40c-.46c) smaller than the ACE NXS(1). Those entries are taken from record 53ret-j3386 (xsdir_overrides).
JEFF-3.3 neutron, TSL Neutron is one tar.gz per temperature (6); each archive is wrapped in its own directory (ace_293/, ...), while the master xsdir uses bare names. The wrapper directories are stripped (strip_top_level_dir).
JEFF-3.3 neutron, TSL The master xsdir.jeff33 lists both neutron and TSL entries. Each sublibrary keeps only entries whose file is in its payload.
JEFF-4.0 TSL Data nested in data/, but jeff4.xsdir uses bare names. data/ is flattened.
JEFF-4.0 TSL Four table names cover 32 temperatures each (see below). .01 series discarded (discard_members); tables renamed (rename_xsdir_tables).
JEFF-3.3 proton The entries file names the data file tendlh, but the archive holds tendl17h. Access route is iaea. File renamed (rename_members); route reset in the final pass. Readme Ace-Readme.tendl17h becomes README.md (readme_member); graphs.tgz deleted (drop_members).
JEFF-4.0 proton The xsdir has no trailing newline. Added in the final pass.
D1SDATA Data in d1sdata/, xsdir_d1s loose at the zip root, trailing blank lines. d1sdata/ flattened; blanks dropped in the final pass.
D1SDATA The xsdir lists a Cl35.ace missing from the zip. Entry dropped.
D1SDATA Revised and new tables released after the main archive. Two patches applied (see Patches).
IRDFF-II The master IRDFF-II.xsd uses Windows backslash paths (unsupported); per-isotope .xsd fragments also ship. Fragments used; master deleted (drop_members).

JEFF-4.0 TSL UO₂ tables. ouo2.00t, ouo2.01t, uuo2.00t and uuo2.01t each name 32 files, one per temperature (128 of 316 entries), so a temperature cannot be selected by table name. The .01 series is the same evaluation processed 1 K away from the .00 one (suspected processing bug), so its 64 files are discarded. The remaining 64 tables are renumbered in the xsdir only (ACE headers untouched) to ouo2.00t-ouo2.31t and uuo2.00t-uuo2.31t, ascending with temperature (285 K to 1700 K), so 62 of them get a new name. The same index means the same temperature in both materials.

sources.json

Each library entry has neutron, tsl, proton and dosimetry keys, null for categories it does not publish. date and production are set once per entry and merged into each category, so the sublibraries of one release cannot drift apart. mode selects the build strategy:

Mode Used for What it does
ready FENDL-3.1d neutron, D1SDATA, JEFF-4.0 TSL, JEFF proton Applies flatten_dirs and rename_members, renames xsdir_member to xsdir, then applies the common transforms.
fragments FENDL-3.2c, FENDL-3.1d proton, FENDL-2.1, IRDFF-II Extracts all archives into one directory, concatenates the fragments matched by fragment_glob, prefixing each filename with its fragment's directory (none for a fragment at the archive root), deletes the fragments, then applies flatten_dirs and rename_xsdir_files. A fragment may hold several entries and + continuation lines; any other non-blank line is an error.
merge-master JEFF neutron, JEFF-3.3 TSL Extracts all archives into one directory (strip_top_level_dir optionally strips each archive's top-level directory), then builds the xsdir from the separate xsdir_source.

Optional fields:

Field Modes Effect Error if
flatten_dirs ready, fragments Moves the contents of these directories (possibly nested, e.g. neutron/ace) to the payload root. A leading <dir>/ is stripped from the xsdir filenames that have it, so the vendor xsdir may use either prefixed names (FENDL-3.1d's fendl31d_ACE/) or bare ones (JEFF-4.0 TSL's data/). directory missing, name collision, leftover content
rename_members ready Renames payload files. file missing, target already exists
discard_members ready Deletes files before the xsdir is filtered, so their entries go too. file missing
rename_xsdir_files fragments Maps an xsdir filename to the real spelling of the data file, after flatten_dirs. The data file is not modified. new file not in the payload; old file not in the xsdir
rename_xsdir_tables ready Maps a data file name to the new table name of its xsdir entry, after missing-file entries are dropped. The ACE file is not modified. file not in xsdir, name still repeated; two files mapped to one name fail validation
drop_members any Deletes files after the build and patches. file missing
readme_member any Moves a vendor readme to README.md at the archive root. file missing, README.md already exists
strip_top_level_dir merge-master true strips each archive's top-level directory on extraction (JEFF-3.3's ace_293/, ...).
xsdir_overrides merge-master A second xsdir source (url/path + md5). Only the table names in its entries (e.g. 55134.40c) are taken from it, replacing the same-named entries in place; changes are logged. name missing from either xsdir
cache_name any url source File name under cache/<NAME>_<CATEGORY>/, for URLs that share a basename (both JEFF-4.0 xsdirs are jeff4.xsdir). not a plain file name

The manifest is validated before any download. An unknown field (e.g. drop_member), or one the mode does not read (e.g. discard_members in merge-master), is an error. So is a library key that is not letters and digits only, and a member path that is absolute or contains ...

Source entries

A source in archives, xsdir_source, xsdir_overrides or patches gives either a url (downloaded and cached under cache/) or a path relative to the current directory, for unpublished data. For example, D1SDATA uses:

{"path": "d1sdata_YTF5QD_v1_0.zip", "kind": "zip", "md5": "f2f268a2b95d06046dac72631223d554"}

Run nd-prepare from the directory holding that file. Keep that directory outside the git checkout: local sources are not public, and .gitignore excludes *.zip only as a safety net.

Every source requires an md5, checked on each download, cache hit and local read. A mismatch aborts the run, and a bad cached download is deleted so the next run downloads it again. An interrupted download resumes on the next run, or starts over if the server cannot resume it. Where the checksums come from:

  • JEFF: published by the vendor (data.oecd-nea.org record API);
  • FENDL, IRDFF-II: cross-checked by downloading twice, with curl and wget;
  • D1SDATA: computed from the single local file.

Patches

patches adds data released after a library's main archive. D1SDATA has two: a revised Nb93.ace, and a revised Cu65.ace plus new Re185.ace and Sb121.ace. Each patch is a source (path or url, plus kind and md5) with a files list, applied after the base payload is built:

Field Meaning
member The file's path inside the patch archive.
dest Destination relative to the payload root after flatten_dirs, without ... An existing file is overwritten.
xsdir_fragment (optional) A member holding one xsdir entry for the file. It replaces the entry with the same ZAID, or is appended if there is none. Omit it when the ZAID and table length are unchanged (D1SDATA's Nb93 patch).

D1SDATA's fragments are fill-in-the-blank templates, with the literal words filename and route in those fields. Both fields are overwritten outright (the route is always 0), not pattern-matched.

Code layout

File (src/nd_prepare/) Role
cli.py Arguments and orchestration
common.py Shared constants and logging
prepare.py Per-entry build, libraries.json, prune
builders.py The three modes, patches, nesting
xsdir.py xsdir text handling
sources.py Download, md5
extract.py Safe zip/tar extraction
pack.py Reproducible archives; every file is stamped with the library's date from sources.json
manifest.py Manifest loading, validation and entry selection
sources.json The manifest, bundled as package data

End-to-end verification

The scripts in tests/ target POSIX systems; nd-prepare itself does not.

tests/verify_full_run.py is a pre-release check, not a unit test suite. It:

  • runs the whole manifest through nd-prepare (the full ~24-26 GB download);
  • installs every archive into a throwaway nd_manager profile (this needs a sibling nd_manager checkout, or --nd-manager-path);
  • checks that every installed file is under its library's own directory, and that each xsdir, per library and merged, ends with a newline, has no blank lines, and gives every entry route 0 and a file that exists.

Libraries whose local path sources are missing from --workdir (D1SDATA) are skipped. To run it:

  • Smoke test: --only IRDFF2:dosimetry.
  • Full run on a cluster: tests/run_verify_full.slurm submits it to an offline compute node.
  • Less disk: --cleanup-work is forwarded to nd-prepare, which deletes each entry's unpacked payload as it goes (the cache is kept).

Independent md5 and xsdir validation

Three more scripts check the installed result against the vendor data. They use only the Python standard library, never nd_prepare, nd_manager or sources.json, so a bug in the tool cannot hide in the check:

python3 tests/validate_md5.py           --workdir DIR [--jobs N] [--only LIB ...]
python3 tests/validate_xsdir.py         --workdir DIR [--strict] [--detail] [--only LIB ...]
python3 tests/validate_downloads_md5.py --downloads DIR --data DIR [--jobs N] [--only LIB ...]
  • validate_md5.py: every installed file is byte-identical to a member of a vendor archive.
  • validate_downloads_md5.py: every file nd_manager installed is byte-identical to its member in the archive nd_manager kept (keep downloads). It streams each archive in memory, one worker per archive.
  • validate_xsdir.py: every xsdir entry matches the vendor xsdir field by field, except the file path and a placeholder route, and agrees with the header of its ACE file. xsdir_all, the merged xsdir that nd_manager writes to MCNP_DATA/, holds exactly the per-library entries.

What validate_md5.py and validate_xsdir.py expect is written by hand in tests/validation_expected.json (its keys are described in tests/README.md): where each original xsdir is, whether renames are allowed (JEFF-4.0 TSL), and the counts from a reviewed full run. A changed count is a warning, or an error with --strict; once reviewed, validate_xsdir.py --write-expected-counts updates them. A library listed there but not installed is an error, unless --only is given. That includes D1SDATA when its zips are missing.

validate_md5.py and validate_xsdir.py write their reports to DIR/validation/, and tests/run_verify_full.slurm runs both after verify_full_run.py. validate_downloads_md5.py only prints its result; it needs nd_manager's kept downloads and is run by hand.

Trademarks and affiliation

MCNP® and Monte Carlo N-Particle® are registered trademarks owned by Triad National Security, LLC, manager and operator of Los Alamos National Laboratory for the U.S. Department of Energy. nd_prepare is an independent ENEA project and is not affiliated with or endorsed by Triad National Security, LLC, Los Alamos National Laboratory, the IAEA, or the OECD Nuclear Energy Agency. Data library names (FENDL, JEFF, IRDFF, D1SDATA) are used only to identify the data sets that nd_prepare downloads.

What nd_prepare knows of the xsdir format (the 80-column limit, + continuation lines, the entry fields), of the nd_manager format and of the ACE format comes from publicly available sources, listed in References. This repository contains no MCNP® source code, executables or data libraries.

References

  • Colin Josey, Jeremy Conlin, LANL Nuclear Data Manager Format Specification v1.0 (2022), doi:10.2172/1880470
  • Jeremy Conlin, Wim Haeck, Paul Romano, A Compact ENDF (ACE) Format Specification (2026), doi:10.2172/3427064
  • Los Alamos National Laboratory, LA-UR-25-26264 (2025): the fields of an xsdir entry
  • Joel Kulesza, Terry Adams, Jerawan Armstrong, Simon Bolding, Forrest Brown, Jeffrey Bull, Timothy Burke, Alexander Clark, Robert Forster III, Jesse Giron, Tristan Grieve, Colin Josey, Roger Martz, Gregg McKinney, Eric Pearson, Michael Rising, Clell Solomon Jr., Swaminarayan Sriram, Travis Trahan, Colin Weaver, Stephen Wilson, Anthony Zukaitis, MCNP® Code Version 6.3.1 Theory & User Manual (2024), doi:10.2172/2372634

Metadata

Release files for nd-prepare 0.9.1

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for nd-prepare 0.9.1
File Size Uploaded Provenance Actions
nd_prepare-0.9.1.tar.gz 79.7 kB With provenance Details for nd_prepare-0.9.1.tar.gz

Built distribution (wheel)

Table of built distributions (wheels) for nd-prepare 0.9.1
File Interpreter ABI Platform Actions
nd_prepare-0.9.1-py3-none-any.whl Python 3 none any Details for nd_prepare-0.9.1-py3-none-any.whl

Total release size: 126.6 kB

Release files / nd_prepare-0.9.1.tar.gz

Download URL nd_prepare-0.9.1.tar.gz
Size 79.7 kB
Tags Source
SHA-256 checksum
How to use checksums
ec3816779e3c8265906748e20e59fd95a78ab57a67c59020b397991291956be8
BLAKE2b-256 checksum
How to use checksums
789601e7147ee4f6688c2ffd2635c105fd81f674d7267bf4f35cf2826ef4de43
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 19, 2026.

Transparency log

Release files / nd_prepare-0.9.1-py3-none-any.whl

Download URL nd_prepare-0.9.1-py3-none-any.whl
Size 46.9 kB
Tags Python 3
SHA-256 checksum
How to use checksums
7435adfcb38f90f82fe7721970de23883c3c76c81ccae0677a9b93375c60ff69
BLAKE2b-256 checksum
How to use checksums
8799700286449a83dc419496cc2c35c0ce8642781975ce9336cf508d3178d2c4
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 19, 2026.

Transparency log

Security for nd-prepare 0.9.1

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

Full provenance

Both files in this release have been uploaded with provenance metadata.

Select from 2 files

Attestation for nd_prepare-0.9.1.tar.gz

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 19, 2026.

Transparency log

A note on package security

PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.

To learn more about protecting yourself from malicious actors, visit our package security documentation.

Release history Release notifications | RSS feed

This release

0.9.1 This release

2 release files Full provenance

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page