Your structure, made real and made auditable. Three axes — D (david), G (gap), A (avant) — each {−1, 0, +1}, composed three deep: 3×3×3 = 27. The whole state space fits on one screen, which is the first useful thing — every cell is colored by which attractor it falls into, computed exhaustively over all 27, no sampling. The three rotations {dga → gad → adg} are a genuine Z/3 symmetry of the dynamics here (verified), and the center (0,0,0) is a fixed point — the held seam, once more. Run a point and watch it fall; rotate the axes and watch the basin map permute.
Bridge-Burners LLC · Fiddler · D·G·A each ±1 · 27 states · Z/3 rotation · (0,0,0) held · anchor: AKASHA
each slice fixes A · within a slice: D →column (−1,0,+1) · G ↓row (+1,0,−1) · cell shows the triple
State
D · G · A0 · 0 · 0
step0
basincenter
ruleD'=sgn(G−A)…cyclic
All 4 attractors (exhaustive)
27 states → every fate known. That's the audit value: complete coverage, zero Monte-Carlo.
Useful or no — audited
UsefulA 3×3×3 ternary tensor is exactly a ternary 3D conv kernel (real in ternary nets / BitNet-style weights), and a 27-state space is small enough to enumerate completely — every trajectory's fate proven, not sampled. For audit/sim that exhaustiveness is the whole point.
Real-if-trueThe Z/3 rotation gives 3× parameter-sharing / equivariance — but only if your axes are genuinely interchangeable. Verified for this rule; you must verify it for yours.
ScaffoldThe names david/gap/avant carry meaning for you and zero compute for the engine — it sees axis 0,1,2. And 27 is a unit cell: the power is in tiling/stacking it, not in 27 alone.