DBE-S v1 · pjrny/dbe-s

Simulable topological stack

Each layer is a falsifiable test that feeds the next. Q=1000 is inverse design — confinement × (1 − disruption) — not a claimed shot.

GitHub
Q_0D at fτ = 1
5.55
fτ* for Q_0D = 1000
2.00
P_disrupt (gated)
0.00
Q̄ at fτ*
≥ 1000†

Q = 1000 is an inverse-design target: find f_τ* such that a 0D D–T point with τ_E = f_τ* τ_0 satisfies Q_0D ≥ 1000. The stack can only move the time average Q̄ = Q_0D (1 − P_disrupt). Hitting Q̄ = 1000 requires near-ignition confinement AND P_disrupt → 0. It is not a performance claim of v1. †Q̄ ≥ 1000 here means P_disrupt = 0 on this 1-mode ODE at the inverse-design point — not a reactor shot.

  1. L0
    Artin OK
  2. L1
    pFlip 0.02
  3. L2
    2T strobe
  4. L3
    Optional compressed register
  5. L4
    Pd 0.00
  6. Σ
    Non-additive plant gain

Stack Ablation + Q̄ gate

PASS

in A–F (E optional)out Non-additive plant gain

Hypothesis
The integrated stack is non-additive: L0 flag on an L2 clock with L1 flip noise beats noisy PID on peak island width, raises Q̄ = Q_0D(1−P_d), and A–D,F all PASS. Q=1000 remains the 0D f_τ* target, reachable in Q̄ only if P_d → 0.
Null
The full stack does not beat noisy PID, Q̄ does not improve, or any of A–D,F is FAIL.
Metric
peak island full vs PID; Q̄_syn vs Q̄_PID; conjunction of layer verdicts; f_τ*
Baseline
noisy PID plant (no topological flag); layer ablations
Threshold
full < 0.95 PID peak AND Q̄_syn > Q̄_PID AND A,B,C,D,F all PASS
Observed
layers=A:PASS B:PASS C:PASS D:PASS F:PASS; E:INCONCLUSIVE; peak full=0.303 PID=3.176; Q̄ syn=10000 PID=0; f_τ*=2.00

Inverse design · log10 Q_0D vs fτ

Assumptions

  • Interfaces are classical: logical bits, not a quantum plant Hamiltonian
  • E (SYK) is optional and may be INCONCLUSIVE without failing the stack
  • Q=1000 remains a confinement-multiplier target, not an experimental claim
  • Uplink composition: Artin → pFlip=1/2 if FAIL; L1 pFlip; L2 strobe

Failure modes

  • A FAIL on any of A–D,F kills integration
  • Noisy PID already saturates the 1-mode island — then the stack has no plant gain
  • f_τ* ≫ 1 with no physical mechanism from L0–L3 is a FAIL of the research program, not a PASS of the ODE

Notes

  • Ablation x=0 open loop, x=1 noisy PID, x=2 L0–L2 gated.
  • Next versions must replace PID with a DIII-D-class ML controller as the baseline.

0 ms