v2.19.0: Experiment 17 -- term-order x noise interaction (real positive result)
Experiment 17: term-order x noise interaction check
Experiment 16's own caveat asked whether term-order sensitivity behaves
differently under noise than it does noiselessly. It does.
Method: same original-vs-reversed K=10+10 SYK term ordering as
Experiment 16, but with a depolarizing Kraus channel injected after
each protocol phase (noise_p=0.01), delta averaged over 6 noisy
trials per order (common random numbers between mu signs to isolate
the sign effect from noise-realization variance).
Result -- the first candidate in this whole project that stays
significant as the sample grows, instead of regressing to nothing:
| n | r | p |
|---|---|---|
| 6 | +0.811 | 0.050 |
| 20 | +0.587 | 0.0065 |
| 30 | +0.396 | 0.030 |
| 50 | +0.340 | 0.0158 |
For comparison, Experiment 16's noiseless version of the identical test
collapsed from p=0.34 (n=6) to p=0.13 (n=30) over the same range --
that one was an honest negative result. This one is not: the point
estimate shrinks as expected with n, but stabilizes around r~0.34-0.40
and stays under p=0.05 at every step from n=6 to n=50.
25/50 (50%) of the n=50 sample are still wrong-signed, so this is a
modest, partial structural handle on the sign -- not a resolution of
it. But it is the first of nine independent candidates tested across
this project to show a real, replicated correlation.
Full write-up: https://tatopenn-cell.github.io/Dense-Evolution-Discovery/wormhole_syk_teleportation/