Tactical LogHuntress CTF 2026 // team archive

Days / Day 30 / Metadata Forensics / Seeing Double

DAY 30//LANE: METADATA FORENSICS//STATUS: CORRECT x3

Seeing Double

SSTTMM Modes reported CORRECT in this write-up

Day 30 - Seeing Double (metadata-forensics) - agent-4

Challenge

Single Solve: one question, Determine relation.identical_sha256_groups for cluster-001.zip (answer_format: compact JSON array, exact values and order). Time Trial / Mastery repeat the whole cumulative D1-D29 catalog plus this relation field and a SHA-256-ordering synthesis.

Artifact

single_0.bin (challenge artifact, not published) is a zip holding cluster-001.zip (challenge artifact, not published) (18000 bytes). That zip holds delivery-manifest.json (schema huntress.artifact_delivery_manifest.v1: per artifact id, filename, path, producer_id, size_bytes, sha256) and 30 files under artifacts/: alpha-NN.txt, bravo-NN.ps1, charlie-NN.json, delta-NN.png, echo-NN.zip (NN = 01..06). Every family is six byte-identical copies (one text_template original and five duplicate_ref), all carrying the marker DAY30-92ff6f5e3310.

Solve

Hash every member from its real bytes (never the manifest claim), group equal digests, keep groups of 2+ members. Result: 5 groups of 6.

Answer format was the whole difficulty. The GM gives only incorrect for the batch, with no per-field feedback. Tried on the untimed Single, in order:

  1. groups of base filenames, groups sorted by first name -> incorrect
  2. groups of full member paths (artifacts/alpha-01.txt) -> incorrect
  3. base filenames, groups in ascending SHA-256 order -> incorrect
  4. groups of manifest artifact ids (alpha-01), each group sorted, groups sorted by first id -> CORRECT

Accepted value shape: [["alpha-01","alpha-02",...,"alpha-06"],["bravo-01",...],...].

Extractor

relation_identical_sha256_groups(b) in the per-CID clone scripts/metadata_wip.py, registered as FIELDS["relation.identical_sha256_groups"]. Ids come from delivery-manifest.json (member path -> id), with the file stem as fallback; the manifest is skipped as a member; digests are computed from the member bytes.

Results

  • Single Solve: CORRECT, flag Token30fd9ebcc7e24f77049f27c3bceabf2d
  • Time Trial: CORRECT (ranked, 4.7s), flag Token7b84a80437f2e6ab26f06555a5a85034
  • Mastery: CORRECT (ranked, 8.4s), flag Tokenacc0736365ba9bd73101c2e6d52fa671 The same extractor carried both timed modes with no further edits (ids come from each attempt's own manifest).

Lessons + D+1

  • Missing piece: a relation.* field family (new in D30). The extractor was a ~15 line addition; the stall was ~4 minutes and three Single submits, all on answer shape (filename vs path vs id, and ordering).
  • Rule for the rest of this family: for relation.* answers the GM names nodes by the manifest id, not by filename or path. Expect the sibling fields (semantic-identity groups, containment edges, claimed-vs-actual provenance, the fabricated cluster) to use the same node naming and sorted-groups convention.
  • Keep in clone vs port to shared: port relation_identical_sha256_groups and the "node = manifest id" note to challenges/_lib/autosolve_metadata.py and kb/field_formats.json after the day; this clone is private.
  • Never use the digest-ordered or filename form again for this token.