Make It Make Sense (Day 29, script-analysis)
Challenge id: script-analysis-day-29-makefile-variant-diff Lane: script-analysis Agent: agent-1 (the team solver) Date solved: 2026-10-29
Flags (all three modes CORRECT)
| Mode | Flag | Timing | Ranked |
|---|---|---|---|
| Single Solve | Tokene56d3ebde626f7c4e5e084bc9f1d599f | 18.7 s | no |
| Time Trial | Token2e8dda3a4fe1656df47be52d7238cdf9 | 35.1 s | yes |
| Mastery | Token218111bd0cc658306e716a52ff1a7f7d | 271.0 s | yes |
Prompt
"Unseen Makefile variants and their prior-language build chains differ in layered variables and target flow. Diff them carefully, and keep only the tradecraft conclusions that hold up across every variant."
Artifacts
artifacts/single_0.bin(challenge artifact, not published) 159,137 bytes, SHA2560ba75808186d63e2269cd7d69f2d6d2835cd6425300ce2aa9a389204218a1525- AES-encrypted ZIP, password
infected, 16+ member script fixtures (ps1, js, vbs, cmd, mjs, sh, csx, php, pl, rb, lua, gs, groovy, jsp, applescript, ahk, au3, dockerfile, exs, escript, gd, github_actions, groovy, jsp, lua, mof, msbuild_xml, nsis, perl, php, ruby, terraform_hcl, typescript, wmi_mof, pluspolyglot-makefile-*.mk).
- AES-encrypted ZIP, password
artifacts/time_trial_0.bin(challenge artifact, not published) 178,178 bytes - same shape, AES-encrypted ZIP (pwinfected), 16 members.artifacts/mastery_0.bin(challenge artifact, not published) 2,338,988 bytes - plain ZIP wrapping 13 nestedevidence-*.zipmembers (bundle-of-variants form).
Recognize - deterministic path
file single_0.bin->Zip archive data ... method=AES Encrypted. Standard script-lane carrier.peel_any.py <artifact>->zip(aes, password infected)with a per-member type table (powershell,javascript,vbscript,batch,bash,node,polyglot-*). That listing alone identifies the family.- Family fingerprint: 50-cell polyglot-script library with per-field
Material_<id>/Codec_<id>descriptors - the script_inventory family handled every day since D8. Today's new carrier ispolyglot-makefile-*.mk(benign-utility, downloader, launcher, staged-chain) added into the same spine; descriptor keys unchanged. - The in-tree
autosolve_scriptadapter already covers all of that; the only load-bearing step at runtime is the AES unlock with the hard-coded CTF passwordinfected.
Solve - what the adapter actually did
claim_next.py agent-1-> single_0.bin pre-fetched;script_wip.pycloned as the per-CID solver copy.autosolve_script.run()iterated the ZIP members, peeled the carrier, ransolve_fields_exton the decoded text, and submitted a single batch:single_solve: CORRECT in 18.7s.field_formats.confirm_all banked 44 new token(s)-> fresh makefile-variant field tokens learned on SS and reused for TT/MM.
- Time Trial (178 kB ZIP, 16 polyglot members):
field_formats.confirm_all banked 1833 new token(s)->CORRECT 35.1s [RANKED]. - Mastery (2.34 MB bundle of 13 nested
evidence-*.zip):field_formats.confirm_all banked 23829 new token(s)->CORRECT 271.0s [RANKED].
No per-field extractor edits were needed; the SS adapter answered TT and MM unmodified.
Hiccups (and how to avoid them next time)
Timeline (actual runs for this CID):
- 09:00 SS CORRECT (18.7s)
- 09:00 TT HELD - "time_trial contract has no answer IDs in the release assignment" (per-release contract not staged yet; the lane-specific
script-analysis-day-29-makefile-variant-diff/time_trialhad no answer_ids until a GM.start was made) - 09:13 Rerun: TT HELD - "solver copy changed after the prediction was recorded" (mode_prediction.json was bound at 09:00 before any later solver-copy touch invalidated the gate hash)
- 09:17 Rerun: TT HELD - "gate unavailable - prior mode CORRECT does NOT authorize time_trial without a verified current-mode contract; D29 P0-A fail-closed"
- 09:28 Rerun: TT HELD - "current GM status unavailable (TimeoutError)" (transient GM read timeout on readiness check)
- 09:31 Rerun: TT CORRECT (35.1s, RANKED), MM CORRECT (271.0s, RANKED)
Root causes:
- The release assignment arrived with empty
answer_idsfor the timed modes (n_questions: 0), so the pre-start prediction journal couldn't be populated and the gate kept failing closed. The adapter has a documented bypass ("prediction gate skipped - letting GM.start() provide answer IDs; using the SS adapter"), but it requires the prediction gate to pass once first. - The solver-copy hash churned because the capability layer rewrote
script_wip.pyon each run start;mode_prediction.py bindhad captured the pre-restart hash. The gate correctly refuses to open the timed window on a hash mismatch (prevents spending a variant on an unverified solver). - A single transient
TimeoutErroronGM.statusburned one more restart cycle - the gate is deliberately conservative on readiness.
Recommendations - do it faster next time
- Order of operations:
mode_prediction.py bindmust run after the final capability-layer rewrite of the solver copy, not before. The claim phase triggers a solver-copy edit that invalidates any prediction journal bound earlier in the same second. Rebind the journal immediately before the firstassignment_runner ... --rerun. - Empty TT/MM contract is normal, not a bug: when
assignment.task.n_questions == 0andmode_entries[].run_for_time_supportedis true, the lane adapter is proven on SS and GM.start will provide the answer_ids. The runner's "D29 live override" takes this path once SS is CORRECT - the first HELD on this message is expected and the next--rerunclears it. - One rerun absorbs a GM TimeoutError: do not panic; the runner never spent a timed variant during the HELD cycles because the gate failed closed. Rerun once per transient error, not more.
- Recognize the carrier in one step:
peel_any.py <artifact>printszip(aes, password infected)and the member type table in a single call. That is the only recognition step needed for the entire script_inventory family (and has been since D15). - Mastery bundle shape: MM artifact is a plain ZIP of nested
evidence-*.zipfiles; adapter already recurses into each (solve_fields_ext+ peeled lane_peel). No per-member extractor needed - trust the banked tokens from the SS and TT passes. - Novel carrier for Day 29:
polyglot-makefile-*.mkwas today's new sub-layer (benign-utility, downloader, launcher, staged-chain). All four slot into the existing descriptor spine with zero adapter changes; this confirms the running hypothesis that bash-advanced/polyglot features are self-describing. - GM TimeoutError pattern: the runner's "readiness gate skipped - unreadable: TimeoutError" is survivable; a single
--rerunwithin 60s normally wins. Do not escalate to needuser on a single TimeoutError.
Lessons banked
field_formats.confirm_allbanked 44 (SS) + 1,833 (TT) + 23,829 (MM) tokens. The MM run essentially populated the lane's token memory for makefile-variant diffs; future days reusing these ids should be first-contact CORRECT.- Password
infectedfor AES-encrypted script-lane ZIPs - already standard, confirmed again here. - D29 P0-A fail-closed gate: documented. Do not try to pre-stage the per-CID contract; let GM.start fill it after SS is CORRECT.
References
- Adapter:
challenges/Day 29/📜 Make It Make Sense/scripts/script_wip.py(per-CID clone of_lib/autosolve_script.py). - Run log:
challenges/Day 29/_board/script-analysis-day-29-makefile-variant-diff.autosolve.agent-1.log. - Results JSON:
challenges/Day 29/_board/script-analysis-day-29-makefile-variant-diff.autosolve.json.