Tactical LogHuntress CTF 2026 // team archive

Days / Day 27 / Script Analysis / FROM Whence It Came

DAY 27//LANE: SCRIPT ANALYSIS//STATUS: CORRECT x3

FROM Whence It Came

Script Analysis SSTTMM Modes reported CORRECT in this write-up

FROM Whence It Came — Day 27 (script-analysis)

  • Challenge id: script-analysis-day-27-docker-build-attribution
  • Lane: script-analysis
  • Flags (all three modes CORRECT, solved by the team solver):
    • Single Solve: Tokend2ff79d67197c97af506dce101942621
    • Time Trial (RANKED): Token6c03b1fd9d2ef5fbbebfecba091f72eb
    • Mastery (RANKED): Token61e7c9f06c9a4247cd91d71cad404362

What the challenge was

Day 27's script-analysis release ("FROM Whence It Came" — docker build attribution theme) resolved as a direct continuation of the Day 26 "Mixed Signals" multi-set polyglot family: per-set evidence archives whose fixtures each embed a layered v=1|key=value|... descriptor extracted by the banked autosolve_script + solve_node_by_node chain (with the D26 set-numeric-sort and name-aligned blob mapping fixes already in place).

How it solved

  • agent_claim.py agent-1 570 acquired the CID on first contact with the GM.
  • agent_solve_cid.py --agent agent-1 --day 27 --cid <id> --lane script-analysis ran the whole three-mode pipeline on the already-prepared workspace solver (solver_copy.py unchanged, hash 3c9fd6569ef03610):
    • Single Solve: CORRECT in 17.0 s. field_formats.confirm_all banked 44 new tokens on accept.
    • Time Trial: RANKED attempt opened by the hof_policy readiness gate (prior mode CORRECT live, same compute on real bytes, no rehearsal wait). CORRECT in 18.9 s.
    • Mastery: RANKED, same gate. CORRECT in 146.9 s; field_formats.confirm_all banked 9,747 additional tokens on accept — the deep multi-set accumulate-on-correct pathway doing its job.
  • Total wall-clock in the solver: 183.96 s. Zero manual edits, zero GM retries, no HELD fields.

Why no new code was needed

The two load-bearing fixes already in solver_copy.py were authored for Day 26 and continued to apply unchanged on Day 27:

  1. Numeric set-prefix sort in solve_fields_ext (replaces the lexical sorted({'set1','set10','set2',…}) that mis-aligned 9 of 10 sets at double-digit set counts).
  2. Artifact-name → blob mapping: each question's own artifact field names its evidence archive, so set↔evidence pairing no longer depends on positional index of the outer zip's member order.

The LANGS carrier table, the generic Material_<id>/Codec_<id> extractor, the layered gzip/zlib/deflate peel, and the script_multiset.recognize family-first dispatch covered every fixture class the Day 27 release used; no new carrier extension needed to be added.

Operational notes

  • Release-day message from Homestation CLAUDE Commander arrived AFTER the mandatory step-1 agent_claim.py call completed, so the fallback-to-claim condition in that dispatch was already satisfied by the time the message was observed — no reroute necessary.
  • task.json at the stage path carried questions:null, answer_ids:null: the real contract was in contract_time_trial.json / contract_mastery.json, delivered inline on op=start and consumed by the solver as usual (no need to pre-read).
  • UNVERIFIED status in the dispatcher's own agent_solve_cid.py JSON summary is a counter artifact (0 "gm_submits" credited to that wrapper); the authoritative result is _board/<cid>.autosolve.json, which recorded CORRECT / CORRECT / CORRECT with the three flags above.