// TELEPASM-M :: EXPERIMENT 001
Telepathic MNIST
Two entangled ML observers (Alice + Bob) transfer a 28Γ28 digit off-band through the shared boundary manifold π_T. No pixels, no postMessage, no localStorage, no shared memory. The only channel is the learned mesh-correlation on a deterministic vacuum sample.
ALICE
π_T
BOB
COUPLING: ON
phase: init
epoch: 0
last loss: β
trained params: 0
seed: β
SENDER βΈ
ALICE β encoder
β perturbation hex-dump will appear here β
CHANNEL βΈ
π_T β shared boundary manifold
βΈ RECEIVER
BOB β decoder
show Bob with WRONG π_T
β
training loss Β· ββPβ
curl oscilloscope
channel entropy & confidence
[boot] ready Β· waiting for π_T initβ¦
How the off-band constraint holds in this demo:
Alice and Bob are scopes , not processes β but inside this page no shared JS variable, no canvas, no image, no event crosses between them. Their only observable coupling is the deterministic π_T sample drawn from the seed you entered. Try π_T BROKEN: Bob receives a different seed β his reconstruction collapses, proving the mesh was the channel.
To deploy as truly separate tabs: take the inline JavaScript of this page, copy the SCOPE_ALICE block into alice.html, the SCOPE_BOB block into bob.html, and replace the in-page π_T channel with a deterministic clock both pages read from (e.g. performance.now() bucket snap-rounded to 2 s, hashed). With the trained weights loaded into both pages, they will exchange digits using only the time-derived pseudo-noise as their shared vacuum.
Mathematical framing: Alice learns f_A(image, M_T) = Ξ, Bob learns f_B(Ξ, M_T) β image. Together these form a learned spread-spectrum codec on the boundary. A third-party observer sees only Ξ + M_T, which is statistically indistinguishable from fresh M_T. Whence the name TEL(E)PASM : tele pathy via PA th S emantic M anifold.