Add hyperspectral IR sounder variables (AIRS, IASI, CrIS) to NNJAObsSat by negin513 · Pull Request #1046 · NVIDIA/earth2studio

negin513 · 2026-08-06T06:39:11Z

Description

Adds the hyperspectral IR sounders — airs, iasi, cris — as variables on the existing NNJAObsSat source, reading the NCEP aggregate BUFR products from the NNJA archive at s3://noaa-reanalyses-pds. One source now covers all NNJA satellite data (microwave + IR) with a single entry point, matching how UFS sat data is structured; a mixed request like ["atms", "cris"] returns a single frame.

All three sensors are returned in brightness temperature (K):

AIRS (Aqua, 2002–2023, airs/nasa/aqua route): stored as TMBR — no conversion needed. Per-channel wavenumbers read from the LOGRCW field in the BUFR.
IASI (MetOp-A/B/C, 2008–present): integer-coded scaled radiance (SCRA) with per-channel-group scale exponents (CHSF); Planck inversion via the shared radiance_to_bt.
CrIS (NPP/NOAA-20/NOAA-21, 2018–present, crisf4 FSR): float radiance (SRAD); Planck inversion. Per-band NFQF/NCQF quality flags packed into the quality column; cross-track position from FORN.

Channel selection follows the JPSS/METOP sensor_indices interface (validated at construction: non-empty, unique, on the instrument grid); default returns all archived channels.

New files

File	Purpose
`data/utils_ir.py`	IASI/CrIS spectral grids, radiance unit conversions
`lexicon/nnja.py`	`airs`/`iasi`/`cris` variable ids (extends existing lexicon)
`test/data/test_utils_ir.py`	Grid, conversion, and cross-consistency tests

Modified files

data/utils_ncep.py — adds decode_ir_sounder() alongside decode_microwave(), using the same ProcessPoolExecutor/batch pattern; widens quality to uint32 (AIRS ACQF is 24-bit, CrIS packed band flags are 28)
data/nnja.py — IR product routing with per-product naming conventions, platform maps, sensor_indices with validation
data/utils.py — radiance_to_bt: log1p → log(1+x) (argument ≥16 over Earth scenes; measurably faster, identical accuracy) and CODATA docstring correction

S3 paths (verified against live bucket)

AIRS: airs/nasa/aqua/YYYY/MM/bufr/airs_disc_final.YYYYMMDD.tHHz.bufr
IASI: iasi/mtiasi/YYYY/MM/bufr/gdas.YYYYMMDD.tHHz.mtiasi.tm00.bufr_d
CrIS: cris/crisf4/YYYY/MM/bufr/gdas.YYYYMMDD.tHHz.crisf4.tm00.bufr_d

Review history

All four correctness items from the review are addressed:

AIRS routed to airs/nasa/aqua (2002–2023) with the airs_disc_final filename convention
CrIS cross-track position emitted: scan_position is FORN (1–30) for CrIS, FOVN for all other sensors
CrIS quality decoded: per-band NFQF/NCQF packed as nfqf | ncqf<<19 per channel row; IASI comment reworded to name the footprint-level QGFQ-family scalars
quality widened to uint32

Plus the non-blocking dedup and hardening items: Planck inversion consolidated onto the shared radiance_to_bt; CrIS spectral grid pinned against jpss_cris (and IASI channel count against metop_iasi) by test; sensor_indices validation; 0-31-001 guard-spectrum marker handled explicitly with the second-31002 break scoped to AIRS; footprints missing their cross-track position dropped rather than failing the file; wavenumbers resolved once per footprint; documentation for elev (CrIS-only; SELV is platform altitude), scan_angle (always NaN for IR), per-sensor scan_position semantics, and realistic per-cycle row magnitudes in the memory warning.

Note: the ISD/IBTrACS/CFS-reforecast/OPERA obstore migrations previously stacked on this branch were split out to #1058; this PR is now NNJA IR only.

Follow-on PRs

perf(data): vectorize NNJA IR decode and accumulate Arrow tables per batch #1059 (draft, stacked) — performance/memory: vectorized per-footprint conversion and Arrow-per-batch accumulation, per §5.1–5.2 of the review analysis
Emit ≤0-radiance rows with null observation instead of dropping (provenance; §4.4)
iasi_radiance/cris_radiance variables alongside BT (§2.2)
Optional airs_gdas variant exposing the airsev route (collocated AMSU-A + TOCC; §2.3)
Extract a shared MW/IR batch-decode harness (the two paths are deliberately parallel today)

Checklist

S3 paths verified against live noaa-reanalyses-pds bucket
Output schema shared with the microwave sensors (mixed MW+IR requests concat cleanly)
Unit tests — 67 passing (mocked BUFR decode, grid cross-consistency vs jpss_cris/metop_iasi)
Docs — NNJAObsSat entry in datasources_dataframe.rst renders the updated docstring
ruff clean
mypy — only pre-existing pandas stubs warning (identical to nnja.py, ufs.py, etc.)

Adds hyperspectral IR sounder support to the NNJA data sources: - earth2studio/data/utils_ir.py: Planck inversion, IASI/CrIS spectral grids, AIRS wavenumber table, ir32/ir48 channel presets (ported from healda/obs/ir_spectral.py) - earth2studio/data/airs_wavenumbers.csv, ir_channel_ranking.csv: bundled data tables copied from healda - earth2studio/data/utils_ncep.py: decode_ir_sounder() parallels decode_microwave(); handles AIRS TMBR (BT direct), IASI SCRA+CHSF (integer radiance with per-band scale), CrIS SRAD (float radiance) - earth2studio/data/nnja_ir.py: NNJAObsIRSat class, follows NNJAObsSat async fetch pattern; ir_channels param selects channel subset - earth2studio/lexicon/nnja_ir.py: NNJAObsIRSatLexicon S3 paths (verified against live bucket): AIRS: airs/airsev/YYYY/MM/bufr/gdas.YYYYMMDD.tHHz.airsev.tm00.bufr_d IASI: iasi/mtiasi/YYYY/MM/bufr/gdas.YYYYMMDD.tHHz.mtiasi.tm00.bufr_d CrIS: cris/cris/YYYY/MM/bufr/gdas.YYYYMMDD.tHHz.cris.tm00.bufr_d Compose with NNJAObsSat via RoutedObsSource (PR NVIDIA#974): RoutedObsSource({ ("atms", "mhs", "amsua", "amsub"): NNJAObsSat(), ("airs", "iasi", "cris"): NNJAObsIRSat(ir_channels="ir32"), })

copy-pr-bot · 2026-08-06T06:39:14Z

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

greptile-apps · 2026-08-06T06:41:57Z

Greptile Summary

Adds NNJAObsIRSat for AIRS, IASI, and CrIS observations, including channel presets, spectral metadata, radiance-to-brightness-temperature conversion, BUFR decoding, and public exports.

Adds ranked IR channel and AIRS wavenumber tables.
Extends shared NCEP decoding for hyperspectral sounders.
Migrates CFS reforecast, IBTrACS, ISD, and OPERA remote reads to obstore.

Confidence Score: 5/5

The PR appears safe to merge; no concrete blocking or independently actionable non-blocking defect was established.

The new IR source, spectral conversion utilities, BUFR integration, exports, and obstore migrations preserve the relevant source, schema, cache, and remote-read contracts on the reachable paths reviewed.

Important Files Changed

Filename	Overview
earth2studio/data/nnja_ir.py	Implements the NNJA hyperspectral IR dataframe source, request validation, S3 fetching, caching, task construction, and decode orchestration.
earth2studio/data/utils_ncep.py	Adds AIRS, IASI, and CrIS BUFR decoding and brightness-temperature conversion integration; no concrete changed-code failure was established.
earth2studio/data/utils_ir.py	Defines spectral grids, channel presets, radiance-unit conversions, and Planck inversion for the three supported sounders.
earth2studio/lexicon/nnja_ir.py	Adds sensor-name mappings for the new dataframe source.
earth2studio/data/cfs_reforecast.py	Migrates full-file HTTP fetching to obstore while preserving historical cache keys.
earth2studio/data/ibtracs.py	Migrates HTTP downloads and Last-Modified cache validation to obstore.
earth2studio/data/isd.py	Migrates station CSV and station-history S3 reads from s3fs to obstore.
earth2studio/data/opera.py	Migrates OPERA HTTP object reads to obstore while retaining URL-derived cache keys.

_{Reviews (1): Last reviewed commit: "style: black/ruff/mypy cleanup for NNJAO..." | Re-trigger Greptile}

for more information, see https://pre-commit.ci

aayushg55 · 2026-08-06T16:55:07Z

@negin513 Some initial high level comments:

I don't think we need a separate NNJAObsIRSat. It would be nice to match the UFS sat data in how the single "source" contains all the satellite data (IR + MW) and there is a single entry point. We can just have a separate decode path for MW and IR. But open to push back here
For channel selection, we shouldn't add the csv we created (the channel ranking file). Our HealDA specific things should be largely kept orthogonal to the E2S datasource. Channel selection should happen through the existing interface /api that exists for UFS, JPSS CRIS, and others. (i.e. no ir_channels argument). Default should return all channels
not sure if we need airs_wavenumbers.csv as of yet.

# Conflicts: # CHANGELOG.md

Review asked for a single satellite entry point matching the UFS pattern, no HealDA-specific channel presets, and no separate IR class: - NNJAObsSat now serves airs/iasi/cris alongside the microwave sensors: the IR products join _NNJA_SAT_PRODUCTS (identical URI layout) and _decode_file routes IR sensors to decode_ir_sounder with per-sensor platform narrowing. No constructor changes. - Dropped the ir_channels argument, ir_channel_ranking.csv, and the preset helpers; every published channel is returned and selection happens downstream. airs_wavenumbers.csv stays: it populates the schema's wavenumber column for AIRS, whose 281-channel grid is irregular and absent from the BUFR (IASI/CrIS grids are formulaic). - decode_ir_sounder now returns Arrow-typed frames via _rows_to_dataframe so mixed MW+IR requests concat with one dtype contract; per-file fetch timeout raised to 300s for the larger IR aggregates; CSVs added to the sdist include. - Removed nnja_ir.py and its lexicon; airs/iasi/cris live in NNJAObsSatLexicon as sensor::MNEMONIC routes. - Tests: IR task routing/URIs, pre-first_year archive gating, decode routing incl. satellite narrowing, and physically-verified AIRS passthrough + IASI CHSF/Planck + CrIS band-wavenumber decodes, plus utils_ir spectral-grid and Planck unit tests.

Verified against the live noaa-reanalyses-pds bucket and real BUFR decodes: - Route cris to cris/crisf4 (FSR, 2018-present). The previously routed cris/cris archive is the NSR product with different channel numbering, which the FSR conversion grid silently mis-converted for channels >= 714; it also stopped publishing in April 2020. crisf4 verified to carry 431 channels numbered on the 2211-channel FSR grid. - Correct archive coverage gates: airsev is 2007-2020 (not 2002+), mtiasi starts 2008 (not 2007). Products gain an optional last_year so out-of-coverage requests raise archive_unavailable at planning time instead of remote_file_missing after a wasted fetch. - Terminate the IR channel loop at the second replication block: airsev appends AMSU-A/HSB blocks reusing the TMBR descriptor, whose microwave values previously leaked into the output as airs rows. - decode_ir_sounder now raises on failed messages like decode_microwave so the folded source keeps one strict completeness contract. - Add sensor_indices constructor parameter mirroring the JPSS/METOP channel-selection interface (default None = all channels). - Skip planning IR tasks when the satellites filter excludes every platform of a sensor, avoiding a multi-GB fetch that would return zero rows. - Docstring/unit-notation corrections and pinned Planck constants in tests.

Exercises the full fetch pipeline for a request spanning both sensor families with mocked downloads/decodes: one Arrow-typed frame, shared schema and dtype contract across the concat.

negin513 · 2026-08-11T00:42:05Z

Thanks @aayushg55, good points — went ahead and restructured:

Dropped NNJAObsIRSat entirely. The IR sounders are now just variables on NNJAObsSat (airs, iasi, cris) and _decode_file routes them to the IR decode path internally. So one source for all the sat data like UFS, and a mixed request like ["atms", "cris"] comes back as a single frame.
Removed ir_channels and the ranking csv. Default is all channels now. I added a sensor_indices arg for narrowing since that's the interface JPSS_CRIS and the METOP sources already use, but you only get subsetting if you ask for it.
On airs_wavenumbers.csv, what would be the best way to handle it? Concidering missing wavwenumber data?

The airsev aggregate encodes the channel centre wavenumber per channel as LOGRCW (descriptor 025076, log10 of the wavenumber in 1/m). Decoding it removes the need for the bundled CSV table entirely; verified on a real 2019 file that all 281 channels resolve with values matching the old table to <0.01 1/cm. The same real-file check exposed that the AMSU-A/HSB blocks trailing the AIRS channels are not always preceded by another replication descriptor, so the second-31002 terminator was insufficient — with LOGRCW decoded, leaked microwave channels would have carried plausible- looking wavenumbers. The declared replication count is now honored as the channel-block budget, which closes the leak for the real layout (exactly 281 channels decode, no extras). IASI/CrIS decode unchanged (verified byte-identical on a real crisf4 file). Also removes the previous dead SCCF wavenumber path and fixes the SCRA unit notation in the iasi_radiance_mw docstring.

negin513 · 2026-08-11T06:43:29Z

Update on the airs csv: turns out it's not needed at all. The airsev BUFR encodes the centre wavenumber per channel (LOGRCW, log10 of the wavenumber in m⁻¹), so I now read it straight from the file and deleted the csv. Checked against a real 2019 file, all 281 channels match the old table to <0.01 cm⁻¹.

Doing that also surfaced a decode bug. The AMSU-A/HSB blocks that follow the AIRS channels aren't always separated by another replication marker, so a few microwave channels were leaking through as AIRS rows. The decoder now uses the declared replication count as the block boundary.

aayushg55 · 2026-08-12T17:41:43Z

@negin513 Thanks for the restructure.

I had my agent review the NNJA IR based on our current work/findings in converting the same bufr to parquet. It produced a rather lengthy document which I think has some highly relevant information which would be hard to distill otherwise, so attaching in its full here as documentation/reference. The summary and the suggested changes are both at the top, everything after that is the additional evidence.
e2studio_pr1046_nnja_ir.md

The four correctness items, repeated here:

AIRS route. airsev covers 2007-02-21 → 2020-12-07, but airs/nasa/aqua in the same bucket
covers a longer 2002-08-31 → 2023-01-18. The two are structurally identical and mostly agree on the infrared
values. Worth defaulting to the NASA route, or exposing both somehow. (We use the NASA route)
CrIS FORN is decoded into scalars and then never emitted. CrIS is the one sensor where
FOVN is not the cross-track position — FOVN is 1–9 within the field of regard, FORN is 1–30
across track, and all 270 pairs occur. So CrIS currently comes out of this source with no
cross-track position at all, which is what edge-of-scan screening and scan-angle bias correction
key on. jpss_cris.py already carries the two as separate columns.
CrIS quality is hard-coded to "none". crisf4 carries five quality fields. NFQF is the one
that flags the rows damaged by the NOAA-21 neon event, and those radiances are finite, so a
consumer can't spot them any other way. The long layout makes this tractable — resolve the band
for each channel and pack that band's flags.
quality is uint16 but AIRS ACQF is a 24-bit flag table. A value above 65535 doesn't
truncate, it raises, and that happens outside the per-message try, so one such value fails the
whole request. uint32 covers it.

The rest is mainly nice to have code deduplication and some comments worth correcting.

There are also some points on performance/memory, but this is not really specific to this implementation but rather generally applicable for all these raw high-resolution MW/IR sounders data sources, and particularly for IR which has hundreds of channels. These would probably be best addressed in future PRs. TLDR, as is, 48hr of satellite data might consume a lot of memory.

- Route AIRS to airs/nasa/aqua (2002-2023) instead of airs/airsev (2007-2020); the NASA route is structurally identical and spans five more early years plus the instrument's final two. Adds a per-product naming convention since the NASA files are named airs_disc_final.YYYYMMDD.tHHz.bufr rather than gdas.*.tm00.bufr_d. Verified against the live bucket. - Emit CrIS FORN (cross-track, 1-30) as scan_position; FOVN is the 3x3 within-FoR detector index (1-9) and is not the cross-track position. - Decode CrIS per-band NFQF/NCQF quality flags: resolve each channel's band and pack nfqf | ncqf << 19 into the quality column. NFQF is the flag that fires on real sensor failures (e.g. the NOAA-21 neon event) whose damaged radiances are otherwise finite and plausible. - Widen quality to uint32: AIRS ACQF is a 24-bit flag table, and a value above 65535 raised ArrowInvalid outside the per-message try, failing the whole request.

- Drop C1/C2/brightness_temperature from utils_ir.py in favor of the shared radiance_to_bt / PLANCK_C1 / PLANCK_C2 in utils.py, already used by the MetOp and JPSS granule sources for the same instruments. - Switch the shared inversion from np.log1p to np.log(1 + x): the argument never falls below ~16 over Earth scenes, so log1p buys no accuracy and is measurably slower. - Correct the radiance_to_bt docstring, which claimed CODATA 2018 but printed the 2014 values. - Add a cross-consistency test asserting wavenumber_cm_inverse("cris") matches jpss_cris._CRIS_WAVENUMBER_APOD element for element, and that the IASI channel count matches metop_iasi, so the grids cannot drift apart silently.

- Validate sensor_indices at construction (non-empty, unique, integer, within the instrument grid) so an empty or out-of-range selection raises instead of silently returning an empty frame after downloading a multi-GB aggregate; matches the JPSS/METOP sources' behavior. - Recognize DRF8BIT (0-31-001) as the crisf4 guard-spectrum block boundary and scope the second-DRF16BIT break to AIRS, so the guard protection is explicit and a nested replication block in a future IASI/CrIS product cannot silently truncate footprints. The comment now states the channel budget is the primary terminator with markers as backstops. - Drop footprints missing their cross-track position (FORN for CrIS, FOVN otherwise) instead of emitting None into the non-nullable scan_position column, matching the microwave path. - Resolve IASI/CrIS channel wavenumbers once per footprint instead of once per channel, removing a one-element numpy allocation per channel from the hot loop; hoist the conversion imports to module scope. - Log a debug count of IASI channels skipped for missing CHSF bands so a narrower-than-expected band table is distinguishable from no data. - Documentation: elev is CrIS-only (SELV is platform altitude, not a substitute); scan_angle is always NaN for IR; per-sensor scan_position semantics; real per-cycle row magnitudes in the memory warning; CHSF is a channel-group exponent, not per-band.

Revert the ISD, IBTrACS, CFS reforecast, and OPERA obstore migrations to their main state; they now land via NVIDIA#1058 so this PR is NNJA IR only.

aayushg55

Thanks @negin513 for following through with all of the changes!

- Correct the quality column metadata: the microwave path populates it from the per-channel 0-33-081 flags, so "MW: null" was wrong; only IASI leaves it null. - Move the utils_ir module docstring into a comment block below the license header per e2s-002 (only license headers above imports). - Raise on an uninitialized BUFR decoder worker in the IR batch fn, matching the microwave path, instead of miscounting it as per-message decode failures.

negin513 · 2026-08-12T22:11:40Z

/ok to test f4f0679

negin513 · 2026-08-12T22:14:28Z

Thanks @aayushg55 . All issues are addressed and started a perf PR on #1059 for vectorized conversion.

negin513 · 2026-08-12T22:14:58Z

Will bring this in once all tests pass.

negin513 · 2026-08-12T23:56:12Z

/ok to test cbfb1f6

negin513 added 7 commits August 4, 2026 20:16

Migrate ISD data source to obstore

c232fcb

Migrate IBTrACS data source to obstore

63f3bfe

Migrate CFS reforecast data source to obstore

c508d13

Migrate OPERA data source to obstore

d9f9e00

Update CHANGELOG for obstore data source migrations

9bd7eb1

style: black/ruff/mypy cleanup for NNJAObsIRSat

fc205b4

negin513 changed the title ~~feat(data): add NNJAObsIRSat for AIRS, IASI, CrIS from NNJA archive~~ wip(data): add NNJAObsIRSat Aug 6, 2026

[pre-commit.ci] auto fixes from pre-commit.com hooks

0473e52

for more information, see https://pre-commit.ci

negin513 requested a review from aayushg55 August 6, 2026 15:27

negin513 added 4 commits August 10, 2026 17:14

Merge branch 'main' into obstore-independent-four

c880f69

# Conflicts: # CHANGELOG.md

Add end-to-end mixed microwave+IR fetch test for NNJAObsSat

a447c3d

Exercises the full fetch pipeline for a request spanning both sensor families with mocked downloads/decodes: one Arrow-typed frame, shared schema and dtype contract across the concat.

negin513 requested a review from NickGeneva August 11, 2026 05:36

negin513 added 4 commits August 12, 2026 12:06

Merge branch 'main' into obstore-independent-four

b3dce67

negin513 mentioned this pull request Aug 12, 2026

Optimize and migrate ISD, IBTrACS, CFS reforecast, and OPERA data sources on obstore #1058

Merged

6 tasks

Split obstore migrations out to their own PR

8ab5021

Revert the ISD, IBTrACS, CFS reforecast, and OPERA obstore migrations to their main state; they now land via NVIDIA#1058 so this PR is NNJA IR only.

negin513 changed the title ~~wip(data): add NNJAObsIRSat~~ Add hyperspectral IR sounder variables (AIRS, IASI, CrIS) to NNJAObsSat Aug 12, 2026

negin513 mentioned this pull request Aug 12, 2026

perf(data): vectorize NNJA IR decode and accumulate Arrow tables per batch #1059

Merged

6 tasks

aayushg55 approved these changes Aug 12, 2026

View reviewed changes

negin513 removed the request for review from NickGeneva August 12, 2026 22:15

Update NNJAObsSat lexicon test for AIRS, IASI, CrIS entries

cbfb1f6

negin513 merged commit 505b394 into NVIDIA:main Aug 13, 2026
15 checks passed

negin513 deleted the obstore-independent-four branch August 13, 2026 00:25

Navigation Menu

Uh oh!

Add hyperspectral IR sounder variables (AIRS, IASI, CrIS) to NNJAObsSat - #1046

Add hyperspectral IR sounder variables (AIRS, IASI, CrIS) to NNJAObsSat#1046
negin513 merged 20 commits into
NVIDIA:mainfrom
negin513:obstore-independent-four

negin513 commented Aug 6, 2026 •

edited

Loading

Uh oh!

copy-pr-bot Bot commented Aug 6, 2026

Uh oh!

greptile-apps Bot commented Aug 6, 2026

Confidence Score: 5/5

Important Files Changed

Uh oh!

aayushg55 commented Aug 6, 2026

Uh oh!

negin513 commented Aug 11, 2026 •

edited

Loading

Uh oh!

negin513 commented Aug 11, 2026

Uh oh!

aayushg55 commented Aug 12, 2026 •

edited

Loading

Uh oh!

aayushg55 left a comment

Uh oh!

negin513 commented Aug 12, 2026

Uh oh!

negin513 commented Aug 12, 2026

Uh oh!

negin513 commented Aug 12, 2026

Uh oh!

negin513 commented Aug 12, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

negin513 commented Aug 6, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Description

New files

Modified files

S3 paths (verified against live bucket)

Review history

Follow-on PRs

Checklist

Uh oh!

copy-pr-bot Bot commented Aug 6, 2026

Uh oh!

greptile-apps Bot commented Aug 6, 2026

Greptile Summary

Confidence Score: 5/5

Important Files Changed

Uh oh!

aayushg55 commented Aug 6, 2026

Uh oh!

negin513 commented Aug 11, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

negin513 commented Aug 11, 2026

Uh oh!

aayushg55 commented Aug 12, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

aayushg55 left a comment

Choose a reason for hiding this comment

Uh oh!

negin513 commented Aug 12, 2026

Uh oh!

negin513 commented Aug 12, 2026

Uh oh!

negin513 commented Aug 12, 2026

Uh oh!

negin513 commented Aug 12, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

negin513 commented Aug 6, 2026 •

edited

Loading

negin513 commented Aug 11, 2026 •

edited

Loading

aayushg55 commented Aug 12, 2026 •

edited

Loading