A Twisted Perspective — data layer & open questions

Gold Bug 2026, puzzle 7 (TWS), 5-star. Complete reconstruction of the twelve-cube installation, the verified structure inside it, and what still blocks the flag.

Overview

Stickers reconstructed
648/648
every facelet photo-supported
Legal cube states
12/12
unique legal completion each
Audit discrepancies
0
U 108/108 · Rear 108/108
Channels closed
10+
1 live falsification (DEAD)

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.

Title card: twelve cubes mounted upright
The title card: upright, exposing U + F.
Installation: twelve cubes tipped forward on their mounts
The installation: the same states, twisted onto the F/D edge.
Where this stands

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).

#1 · C11
U
L
F
R
Rear
D
#2 · C12
U
L
F
R
Rear
D
#3 · C13
U
L
F
R
Rear
D
#4 · C14
U
L
F
R
Rear
D
#5 · C21 gold
U
L
F
R
Rear
D
#6 · C22
U
L
F
R
Rear
D
#7 · C23
U
L
F
R
Rear
D
#8 · C24 gold
U
L
F
R
Rear
D
#9 · C31
U
L
F
R
Rear
D
#10 · C32
U
L
F
R
Rear
D
#11 · C33
U
L
F
R
Rear
D
#12 · C34
U
L
F
R
Rear
D
Provenance by face class
FacePrimary sourceStatus
Fdisplay capture + straight-on photosphoto-confirmed
Lorbit photos + gallery overviewphoto-confirmed
Rper-column orbit photosphoto-confirmed
U2026-08-08 top-down frames, independent re-read108/108 audit
Dteammate transcription + gallery + in-person readphoto-confirmed
Rearlegality-forced, then photographed head-on108/108 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:

Whole-frame composition of each face's token word
Face wordComposes toNotable rows
F120° about the body diagonal (the “rho” cycle R→U→F)row 1 = identity
Lidentity
Rearidentity
U180° about zrow 1 = identity
D180° about zgold row = identity
R180° 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

Bounded families exhausted (full records in problem-5/investigation/)
ChannelScopeResult
Corner-bit mechanismtitle-frame domain; 4,608-mechanism adversarial familyDEAD falsified live
Installation-frame corner bitsD∪F domain under title cyclecannot run (no singleton transitions)
Per-cube token composition16 conventionsgold control fails structurally
Words executed on gold288 executionszero black-state hits
Corner-orientation variantsrho-linked, preregisteredSPACE collision on cube 11
Speffz mounting letters120 distinct strings, all offsetsnothing readable
Word scrambles (plain)894 notation-spellable words × 3 variants × 2 channelszero hits
Word scrambles (modifier sweep)7,651,260 sequenceszero hits
Historical closuresQTM A1Z26 · Hamming mod-26 · element orders · sticker Speffz · Braille/semaphore/glyphs · trifid · sequential snapshots · move-word executionall negative
Non-channelsQR card (organizer-confirmed) · C24 black sticker (decoration) · BBS console artretired

Outstanding questions & blockers

The core blocker: no letter mapping

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.