[mia-autodev]

Closes #58: For aws-chammi-75-multichannel-pretraining: dataset aws-chammi-75-multichannel-pretraining has no snippet that reads it.

Implemented by snippets: Claude Code running in a sandboxed job on the cluster. The agent's own
summary is the first comment.

• Merge: approve this PR, or comment /approve <head sha> (needed when running under a personal token; a
  trusted worker posts it when the output of this head is approved).
  snippets merges once an approval matches the current head and checks are not failing.
• Request changes: submit a review with Request changes, or comment /revise <what to change>.
  Up to 3 revision rounds.
• Stop: close this PR.

[mia-autodev]

snippets: round 1 pushed 39e2a984c9f2 to snippets/issue-58 · $2.67

Added snippets/png-channels/, a reader for greyscale PNGs stored one file per channel, and committed it as 39e2a98 on snippets/issue-58 (not pushed). The smoke test passes on the real CHAMMI-75 data.

What it does

• One URL is read as a yx image; several URLs are the channels of one field of view, stacked as cyx in binding order.
• Formats: 8- or 16-bit grey PNG, from a plain URL or a zip member, decoded whole with imagecodecs (access: file). RGB, palette, alpha and sub-8-bit PNGs are refused at open with a clear message.
• No params, and no new dependency (numpy, imagecodecs are on the allowlist).
• Only the first file's header is read at open. A later channel that is missing or has a different shape or bit depth fails when it is read.

The example binding

examples/aws-chammi-75-multichannel-pretraining.yaml binds the two fields the record inspected, each with all of its channel files instead of the draft's channel 0 alone:

raw	field	array
0	idr0001 (..._channel-0.png, -1)	cyx 2×1040×1376 uint8
1	hpa0018 (<uuid>-ch0.png … ch3)	cyx 4×512×512 uint8

I found the sibling channel files by listing each field's key prefix on chammi-data.s3.amazonaws.com, the allowlisted host. All URLs are on that host, not the s3.us-west-2 one.

Testing

Smoke summary, run as CLAUDE.md prescribes with MIA_HARNESS_ALLOWED_HOSTS=chammi-data.s3.amazonaws.com:

{"ok": true, "shapes": {"raw": [[2, 1040, 1376], [4, 512, 512]], "labels": []},
 "warnings": [], "errors": [], "bytes_fetched": 784737}

• Real-data crops: harness crop --n 4 made 4 crops, drawn from both channels.
• harness verify: with the issue's Record block as given, all 4 crops fail with "crop.json has no licence", because that block has no license field. With a placeholder licence added to a local copy of the record, all 4 pass.
• CI checker: ci/check_pr.py --base origin/main passes.
• Existing suites: harness tests 225 passed, CI-checker tests 21 passed.
• Edge cases: 15 local tests over the harness range server pass (uint8/uint16, channel order, interlaced and multi-block files, a server without Range, stored and deflated zip members with a glob, and the refusals). These are not committed, because CI rejects a snippet PR that touches harness/tests.

Limits to know about

• Two fields out of about 2.8 million files. The bucket has no directory index and the snippet does not enumerate it, so a binding names each field URL by URL. Covering the dataset at scale needs the file list from an S3 listing or the bucket's metadata CSV, which is outside what a snippet can do.
• Channel names and voxel size are not set. Neither is in a PNG, and the record sets neither.

[mia-autodev]

Review judge: approve

The PR adds a new png-channels snippet that reads 8/16-bit greyscale PNGs over HTTP(S) or from zip members, as yx for one file or as cyx when several per-channel files are stacked. It also adds the requested CHAMMI-75 example binding, which lists two fields with explicit global-host URLs. The code follows the spec's interface, I/O and dependency rules and only adds files under a new snippet directory. The one open question is whether the extra channel URLs exist, because the record does not verify them.

item	severity	note
snippets/png-channels/snippet.py	ok	It decodes greyscale PNGs with imagecodecs.png_decode, giving yx for one URL and cyx for several, which is what the issue asks for. It imports only struct, numpy, imagecodecs and mia_harness.api, and all I/O goes through io.open/io.fetch/io.expand, with the NeedsFetch fallback the spec describes.
snippets/png-channels/snippet.py	ok	open_array reads only the 33-byte signature+IHDR header, not voxel data, as spec §2 requires. Its signature matches the required annotation, and access is reported as 'file' at runtime.
snippets/png-channels/snippet.py	ok	Only grey PNGs (colour type 0) at 8 or 16 bits are accepted. Each channel is checked against the first file's shape and dtype, and a decode that adds an alpha sample from tRNS is caught. Error messages name the URL.
snippets/png-channels/snippet.yaml	ok	Name matches the directory, formats are [png], containers are [http, zip], params are empty, requires is [numpy, imagecodecs] (both shown in the spec's allowlist example), and tested_on matches the example binding.
snippets/png-channels/examples/aws-chammi-75-multichannel-pretraining.yaml	ok	The binding is at the path the issue asks for. It uses explicit URLs on the global chammi-data.s3.amazonaws.com host with no glob over the missing S3 index, and binds no labels, consistent with the record.
snippets/png-channels/examples/aws-chammi-75-multichannel-pretraining.yaml	suspect	The record verifies only idr0001 channel-0 and hpa0018 ch0. The channel-1 and ch1..ch3 URLs depend on a prefix listing the agent says it ran, which the evidence does not show; the crops status will confirm them.
snippets/png-channels/README.md	ok	Documents glob support (needed by spec §2), access mode and known limits, and explains why snippets_commit is the pre-snippet base commit.
diff	ok	Only adds files under one new snippets/png-channels/ directory. No protected paths or existing snippets are touched.
ci.json	suspect	CI was not awaited (shadow), so neither the check nor the real-data crops result is in evidence.

confidence 0.8 · model claude-opus-5-5 · prompt f5f9c39cc98e · policy v1 · $0.23

Shadow mode: this verdict is recorded and compared with the human label; it decides nothing.

[mia-autodev]

Crops at 39e2a984c9f2f1b3fe960d8adfaf42ce817df0bb: passed, 16 of 16 verified (at least 8 needed); the crop job fetched 1737277 bytes.

16 crops

raw	overlay	caption
		i27354089-e1794579687471-b0-000 · z=None c=0 · (732, 554)
		i27354089-e1794579687471-b0-001 · z=None c=0 · (532, 271)
		i27354089-e1794579687471-b0-002 · z=None c=1 · (144, 1026)
		i27354089-e1794579687471-b0-003 · z=None c=1 · (618, 1028)
		i27354089-e1794579687471-b0-004 · z=None c=0 · (223, 897)
		i27354089-e1794579687471-b0-005 · z=None c=0 · (127, 372)
		i27354089-e1794579687471-b0-006 · z=None c=1 · (369, 704)
		i27354089-e1794579687471-b0-007 · z=None c=0 · (424, 623)
		i27354089-e1794579687471-b0-008 · z=None c=1 · (366, 458)
		i27354089-e1794579687471-b0-009 · z=None c=1 · (196, 804)
		i27354089-e1794579687471-b0-010 · z=None c=0 · (483, 829)
		i27354089-e1794579687471-b0-011 · z=None c=1 · (218, 291)
		i27354089-e1794579687471-b0-012 · z=None c=0 · (447, 1111)
		i27354089-e1794579687471-b0-013 · z=None c=1 · (170, 512)
		i27354089-e1794579687471-b0-014 · z=None c=1 · (694, 470)
		i27354089-e1794579687471-b0-015 · z=None c=0 · (49, 1020)

Binding (sha256 4f2180e0705743b2057181f4c6cd5db3c434aae19186f434d9974e652b405b9c):

record: aws-chammi-75-multichannel-pretraining
record_commit: 3f75709f1d2778fe86fd9c436b2b5ff948a72c7b
snippets_commit: 39e2a984c9f2f1b3fe960d8adfaf42ce817df0bb
raw:
- snippet: png-channels
  urls:
  - https://chammi-data.s3.amazonaws.com/CHAMMI-75/CHAMMI-75_train/idr0001/idr0001-plate_100A-converted/Meas_01%282013-01-14_11-13-43%29_001001001_series-0_z-9_t-0_channel-0.png
  - https://chammi-data.s3.amazonaws.com/CHAMMI-75/CHAMMI-75_train/idr0001/idr0001-plate_100A-converted/Meas_01%282013-01-14_11-13-43%29_001001001_series-0_z-9_t-0_channel-1.png
  params: {}
- snippet: png-channels
  urls:
  - https://chammi-data.s3.amazonaws.com/CHAMMI-75/CHAMMI-75_train/hpa0018/00070df0-bbc3-11e8-b2bc-ac1f6b6435d0-ch0.png
  - https://chammi-data.s3.amazonaws.com/CHAMMI-75/CHAMMI-75_train/hpa0018/00070df0-bbc3-11e8-b2bc-ac1f6b6435d0-ch1.png
  - https://chammi-data.s3.amazonaws.com/CHAMMI-75/CHAMMI-75_train/hpa0018/00070df0-bbc3-11e8-b2bc-ac1f6b6435d0-ch2.png
  - https://chammi-data.s3.amazonaws.com/CHAMMI-75/CHAMMI-75_train/hpa0018/00070df0-bbc3-11e8-b2bc-ac1f6b6435d0-ch3.png
  params: {}
labels: []
reasons:
- 'png-channels instead of the draft''s tiff-zip-member: every raw is a single-channel
  8-bit grey PNG (record technical.arrays: format png, axes yx, uint8, deflate), which
  tifffile refuses (''not a TIFF file: header=b''\x89PNG''''); png-channels decodes
  PNG and has no params'
- 'raw 0 is the record''s sampled idr0001 field with both of its channel files as
  cyx 2x1040x1376 instead of the draft''s channel-0 alone: the dataset stores one
  PNG per channel of a field (record path_pattern <field>_channel-<c>.png), and a
  listing of the field''s key prefix on chammi-data.s3.amazonaws.com (?list-type=2&prefix=...001001001_)
  returns exactly channel-0 and channel-1'
- 'raw 1 is the record''s probed hpa0018 field (00070df0-..., <uuid>-ch<c>.png) with
  its four channel files ch0..ch3 as cyx 4x512x512, so the binding covers both file
  layouts and both shapes the record saw (shape_varies: true); the same prefix listing
  returns exactly ch0..ch3'
- 'all urls on chammi-data.s3.amazonaws.com, not the s3.us-west-2 host of the record''s
  hpa0018 url: the record notes say the regional host is not on the allowlist and
  that the global host serves the same keys'
- 'explicit urls, no glob: the bucket has no HTTP directory index (harness list on
  the prefix returns 404), so the channels of a field are listed one url each, in
  channel-number order'
- 'two fields only: the other ~2.8M PNGs are not named in the record (listing_complete:
  false) and each field is its own array, since shapes and channel counts differ by
  study'
- 'no voxel size and no channel names: neither is in a PNG, and the record sets neither
  (they differ by source study)'
- 'no labels bound: record annotations.present is false (self-supervised pre-training
  set)'

[mia-autodev]

/approve 39e2a98