Overview
The artwork presents the same twelve cubes from two perspectives: the title card shows them mounted upright (each cube exposing U + F), while the physical installation tips every cube ~45° forward onto its F/D edge. Two cubes sit in gold cradles; they are provably the same cube state mounted in two different orientations.
The data layer is finished and hardened; the instruction layer is the open problem. Every bounded, gold-controlled extraction family tried so far has closed without producing letters, and the one derived candidate (DEAD) failed a live test.
The dataset
All twelve cubes, exactly as committed in derived/face-transcription.json, drawn in installation coordinates (rows top-to-bottom = distal to player; column 1 = installation left). Gold borders mark the gold-mounted pair (#5 C21, #8 C24).
| Face | Primary source | Status |
|---|---|---|
| F | display capture + straight-on photos | photo-confirmed |
| L | orbit photos + gallery overview | photo-confirmed |
| R | per-column orbit photos | photo-confirmed |
| U | 2026-08-08 top-down frames, independent re-read | 108/108 audit |
| D | teammate transcription + gallery + in-person read | photo-confirmed |
| Rear | legality-forced, then photographed head-on | 108/108 audit |
Interactive: 3-D reconstruction viewer (Girn Mode tab emits the machine-readable JSON; also committed at derived/girn-data/girn.json). The C24 black sticker is confirmed decoration and is excluded from everything.
Verified structure
The gold pair
Cubes #5 and #8 are the only center-normalized duplicate among all twelve (exact on all 54 facelets and every cubie coordinate), and their physical mounts are exact mutual inverse rotations — ±90° about the same horizontal axis the display is tipped on. In row-major order the gold positions split the ten black cubes 4–2–4.
Mounting orientations
Each cube's whole-cube mounting orientation is one of the 24 rotation-group elements. First-seen labels below; nine distinct classes, with structured repeats and the two inverse pairs marked.
#1 A inverse of #12 | #2 B | #3 C | #4 D |
#5 E inverse of #8 | #6 F | #7 G | #8 H inverse of #5 |
#9 D same as #4 | #10 B same as #2 | #11 C same as #3 | #12 I inverse of #1 |
Row-major composition of all twelve mounts is a 120° rotation. Cubes #1/#12 are the only non-gold inverse-mount pair.
Center/count token words
Reading each face's center color + count of center-colored stickers per cube gives a 12-token “word” per face. Under the pinned standard convention these compose to clean group elements — deliberate engineering that has resisted every letter mapping:
| Face word | Composes to | Notable rows |
|---|---|---|
| F | 120° about the body diagonal (the “rho” cycle R→U→F) | row 1 = identity |
| L | identity | — |
| Rear | identity | — |
| U | 180° about z | row 1 = identity |
| D | 180° about z | gold row = identity |
| R | 180° about (1,1,0) | — |
Live tests & channel
- Checker
- CPV BBS flag box (authenticated; tests via Davis)
- Official metadata
- Title + five stars only; no flavor text, no length hint
- Answer format
- All-caps flags; published siblings are unspaced phrases, 11–16 chars
- Tested
DEAD→ “incorrect” (2026-08-08) — falsified, never retried
Closed channels
| Channel | Scope | Result |
|---|---|---|
| Corner-bit mechanism | title-frame domain; 4,608-mechanism adversarial family | DEAD falsified live |
| Installation-frame corner bits | D∪F domain under title cycle | cannot run (no singleton transitions) |
| Per-cube token composition | 16 conventions | gold control fails structurally |
| Words executed on gold | 288 executions | zero black-state hits |
| Corner-orientation variants | rho-linked, preregistered | SPACE collision on cube 11 |
| Speffz mounting letters | 120 distinct strings, all offsets | nothing readable |
| Word scrambles (plain) | 894 notation-spellable words × 3 variants × 2 channels | zero hits |
| Word scrambles (modifier sweep) | 7,651,260 sequences | zero hits |
| Historical closures | QTM A1Z26 · Hamming mod-26 · element orders · sticker Speffz · Braille/semaphore/glyphs · trifid · sequential snapshots · move-word execution | all negative |
| Non-channels | QR card (organizer-confirmed) · C24 black sticker (decoration) · BBS console art | retired |
Outstanding questions & blockers
The piece supplies no A→Z rule anywhere — not on the installation, the title card, or the BBS entry. Every gold-controlled mapping from the verified structure to letters has been closed. Either the intended extraction is lateral (not a state decode), or it rests on a convention we have not identified.
Open questions
- Mechanism-style calibration (top ask): how were the other physical pieces' answers actually derived (e.g.
ORANGESCARF)? Deep decode vs lateral observation would immediately halve the search space. - Unexplained deliberate structure: why the gold ±90° inverse mounts; why #1/#12 invert; why the F word composes to the 120° cycle while L and Rear compose to identity; what the 4–2–4 split encodes.
- What the mounting orientations mean: nine distinct classes with structured repeats — the strongest unexplained signal in the piece.
Blockers & open scope
- Unexplored but unbounded without new signal: multi-word scrambles, non-dictionary move strings, wide-move letter maps, edge-based reductions. Each needs fresh evidence before a defensible sweep exists.
- Minor data nit: C14's U face is proved only via an independent re-read of a legacy frame (no genuinely new top-down covers it).
- Testing budget: unknown wrong-answer policy on the BBS; we spend tests only on motivated candidates.