Day 16 - Same Same, But Different (script-analysis)
GM id: script-analysis-day-16-obfuscation-equivalence | lane: script-analysis | fields per attempt: 1095 Agent: agent-1 (Opus 5 / high). All three modes CORRECT.
Result
| mode | result | flag | time |
|---|---|---|---|
| Single Solve | CORRECT | Tokend2a56ced8f957119291573bd3566e173 | 5.7 s |
| Time Trial | CORRECT | Token60f099bdb00ab526b3c0806cf3b53d35 | 59.8 s |
| Mastery | CORRECT (GM ledger) | already_solved - the GM reported the mode already solved on its own ground truth ("The original solve time is recorded, but the player's name is currently unavailable."); board.py reconcile 16 re-added it from GM truth | 16.2 s |
Mastery was submitted in full (5 sets x 1095 = 5475 fields derived from that attempt's own bytes) before the GM returned already_solved, so the answer path was exercised end to end; the solved state comes from the GM, not a local assumption.
Challenge
GM prompt: "Several obfuscation families have scrambled the same logic independently, or so they claim. Strip each version down to its semantics, prove which ones are truly equivalent, and expose the variant that quietly changed the plot."
Artifact: artifact_0.bin (challenge artifact, not published), 113162 bytes, WinZip-AES zip (123 members, password infected) - 22 .sh, 14 .py, 12 .js, 10 .vbs, 9 .php, 8 .mjs, 5 .cmd, 5 .lua, 5 .pl, 5 .rb, 4 .ps1, 4 .bat, 4 .mof, 4 .ts, 4 .au3, 4 .applescript, 4 .proj. Every fixture is inert; nothing was executed.
The "equivalence" framing is prompt-level only. Underneath it is the same per-field script-triage inventory the lane has answered since Day 1: each fixture carries its own self-describing descriptor (Material_<id> / Codec_<id> plus load/decode/recover/inspect/plan_<id>), and the semantics the prompt asks you to "strip down to" are exactly the descriptor's steps graph, stage-count, role and ATT&CK mapping. Two fixtures that decode to the same ordered behavior graph are the equivalent ones; the "variant that quietly changed the plot" is the fixture whose decoded graph has a different terminal step (e.g. dispatch_process where its siblings end in return_configuration), which the descriptor states plainly once decoded. No semantic diffing of obfuscated source was needed - the graph is in the bytes.
Method
claim_next.pyclaimed the challenge and ran the banked lane adapterchallenges/_lib/autosolve_script.pybefore any hand analysis.- The adapter peeled the AES zip (
pyzipper, passwordinfected), walked all 123 members, decoded each carrier (base64 / hex / fragmented hex / gzip / XOR / char-code arrays) and read the descriptor out of each fixture. - Field answers came straight from the decoded descriptors - scenario/role enums, ATT&CK ids, campaign ids, urls, wallet, mutex, file paths, JScript function/invocation/property style, Batch expansion + control form, VBScript transform + control form, stage counts and
>-joined step graphs. - One considered submit per mode under the exact answer ids.
field_formats.confirm_allbanked 56 new tokens at Single and 1094 more at Time Trial. - Time Trial and Mastery were opened only by AUTOSOLVE after its readiness gate ("prior mode already CORRECT live on this challenge"), never by hand.
Today's new layer
The novelty diff against Day 15 is 69 new question keys, no removals - i.e. a pure addition on top of the cumulative inventory:
1. Ruby (.rb) as a carrier - 58 of the 69 keys. Five new fixture groups:
| field group | role | steps | stage-count | ATT&CK | obfuscation |
|---|---|---|---|---|---|
polyglot-ruby-benign-utility-* | benign_utility | decode_configuration>return_configuration | 2 | none | plain_hex |
polyglot-ruby-downloader-* | downloader | acquire_remote>dispatch_process | 2 | T1105 | base64 |
polyglot-ruby-launcher-* | launcher | decode_embedded>write_file>dispatch_process | 3 | T1059 | hex_base64 |
polyglot-ruby-staged-chain-* | staged_chain | profile_host>decode_configuration>dispatch_language | 3 | T1059 | fragmented_hex |
host-ruby-rack-ruby-* | staged_chain | acquire_remote>dispatch_process | 2 | T1059 | base64 |
Ruby maps to the parent T1059 (Command and Scripting Interpreter) - ATT&CK has no Ruby sub-technique, so the bare parent is the accepted value, and the transfer step still wins the precedence for the downloader (T1105). Campaign ids are RUBY-<8 hex>; urls are https://ruby-<10 hex>.example.invalid/stage; the host container profile is ruby_rack (vs standalone for the four polyglots). next-language values chained out to powershell, jscript, batch and vbscript.
2. Two new self-describing bash-advanced features - 14 keys, bringing the set to 16 features:
| field group | feature | primitive | role | steps | ATT&CK |
|---|---|---|---|---|---|
bash-advanced-ansi-c-quoting-* | ansi_c_quoting | ansi_c_escape_string | advanced_syntax | recognize_syntax>reconstruct_intent | T1059.004 |
bash-advanced-regex-conditional-* | regex_conditional | bash_regex_match | advanced_syntax | recognize_syntax>reconstruct_intent | T1059.004 |
Both cost zero extra work: the feature and primitive strings are read verbatim out of the fixture's own material, the role/steps/stage-count are fixed for the family, and Bash is T1059.004. This is the sixth straight day the "bash-advanced features are self-describing" hypothesis held.
Why no adapter extension was needed
autosolve_script.py already generalises over carrier x transform x representation (the universe test is 1445/1445 PASS). Ruby's carrier idioms ("...".unpack, hex string literals, base64 via Base64.decode64, fragmented hex concatenation) all fall inside the existing codec table, and the descriptor key names are unchanged, so the new fixtures resolved on the first pass - Single Solve landed in 5.7 s with 1095/1095 accepted. The theme ("prove which are equivalent") added no new field shape.
Takeaways
- A new language in this lane is now routinely free: the adapter reads the descriptor, and the only genuinely new knowledge is the interpreter -> ATT&CK mapping. For Ruby that is the parent
T1059(no sub-technique exists). - A prompt that sounds like a new analysis task ("prove equivalence", "find the divergent variant") can still be the same inventory underneath - check the answer-id diff before believing the framing.
- 69 new keys, 0 removed: the Day-8-onward pattern (one carrier/theme + two bash-advanced features per day) held again, which is why the pre-warmed adapter could finish Single in under 6 seconds.