Challenge
- Lane: Metadata Forensics (Day 04), GM id
metadata-forensics-day-04-picture-this - Prompt: "A dozen photographs have been resized, renamed, and shuffled. Reconstruct their true dimensions and find the image whose presentation does not match its structure."
- Question shape:
Determine jpeg.sof_dimensions_pixels for camera-001.jpg. - Modes: Single Solve, Time Trial, Mastery (all reported CORRECT by the Game Master on 2026-10-04).
Artifacts
| File | Size | SHA-256 |
|---|---|---|
picture_this.zip (challenge artifact, not published) | 1698 B | 27bf2675d0c7e6cbd642180cfb8edd38157f75de3c09f5001d5623487c821ba6 |
camera-001.jpg (only member) | 3025 B | bc79a152e217a8e4178a25fcefcaf02dfb2884d37ec4e8881e7d1909a3488318 |
Important scope note: the saved zip contains ONE image (camera-001.jpg), not the dozen described in the prompt. The dozen-image set (Mastery) is attempt-specific and was never saved locally, so per-image values for the other photos and the "mismatched presentation" outlier cannot be derived from available bytes.
Triage
Metadata forensics on a JPEG. The relevant ground truth is the SOF (Start Of Frame) segment, which holds the real pixel dimensions regardless of filename or any EXIF/JFIF claims. Solver: scripts/sof_dims.py (walks marker segments, reads SOFn, reports sizes/hashes; accepts a file, directory or zip).
Investigation
Command: wsl python3 scripts/sof_dims.py picture_this.zip
Output:
camera-001.jpg 3025 bc79a152e217a8e4178a25fcefcaf02dfb2884d37ec4e8881e7d1909a3488318
markers: DB@2(132) C0@136(17) C4@155(418) DA@575(12)
SOF 0xc0: 257x146 (WxH) prec=8 comps=3
Manual cross-check of the hex dump at offset 0x88: ff c0 00 11 08 00 92 01 01 03 01 22 00 02 11 01 03 11 01
ffc0= baseline DCT SOF0, length 0x11, precision 8- height =
0x0092= 146 - width =
0x0101= 257 - 3 components; component 1 sampling
0x22(2x2, i.e. 4:2:0), components 2 and 30x11
Structure observations for camera-001.jpg: no APP0/JFIF, no APP1/EXIF, no COM segment; only DQT, SOF0, DHT, SOS. So there is no competing metadata claim in this file; the SOF is the sole source of dimensions. Note the SOF values are non-multiples of 16 (257x146), consistent with a genuinely resized image (4:2:0 MCU grid 17 x 10 MCUs = 272x160 padded).
Flag Recovery
Derived value for the question shape jpeg.sof_dimensions_pixels, camera-001.jpg: width x height = 257x146 (this matches the "SOF=WxH e.g. 257x146" noted in the day's manager log from the live solve). The exact answer string format accepted by the GM (e.g. 257x146) was taken from that log, not re-verified here since no submission was made.
Not derived (data unavailable locally): the other eleven photos' dimensions and the presentation/structure-mismatched outlier (would be found by comparing each file's EXIF/JFIF-declared or filename-implied size against its SOF size using sof_dims.py on the full set). The original completion tokens were never captured and cannot be recovered from the GM afterward.
Flag
No completion token is available. Derived technical answer for the one available artifact:
camera-001.jpg SOF0 width=257 height=146 -> 257x146
Minimal Reproduction
unzip picture_this.zip
wsl python3 scripts/sof_dims.py picture_this.zip
Read the SOF ... WxH line per image; for a full set, flag the image whose EXIF/JFIF/declared size disagrees with its SOF size.
Lessons Learned
- The true size lives in SOFn, never in the filename or EXIF; walk segments rather than trusting tools that read thumbnails or declared tags.
- Save the per-attempt artifact set and the submitted answers/tokens at solve time; the GM does not return them later.