FUSION-SEAM-001 — Disruption prediction — the cross-machine domain gap

Posted by:

|

On:

|

FUSION-SEAM-001RECOVERY_EXACT · FALSIFIER_AWAITING_DATAS2

FUSION-SEAM-001 · Disruption prediction — the cross-machine domain gap

LIVE CANDIDATE · STANDING SUPPORTED — FALSIFIER STAKED, NOT YET RUN

context
Predictors trained on one tokamak fail on another; the field admits no physical model of the root cause. The classical pipeline trains on normalized scalar magnitudes |x(t)| — the machine-specific normalization IS the address, absorbed into the scalars, and the mode’s rotation state is discarded when amplitude envelopes are taken.
schema
MN-RECOVERY-CERTIFICATE-1.1
certificate_id
MN-CERT-FUSION-SEAM-001-20260830
job_id
FUSION-SEAM-001
title
Disruption prediction — the cross-machine domain gap
author
Derrick E. Muncy
method
MuncyNautics Seam Recovery Method v1.0
source_records
Nucl. Fusion 66 056038 (2026); Nat. Comms. Phys. 2023 · Fusion Seam Recovery — Live Candidates (project record)
species
S2 — The Address
governing_equation_as_projected (the seam)
alarm: |B_mode(t)| > threshold_machine  (x → |x|; φ, dφ/dt and the q-surface seat discarded; the address absorbed into a dimensional threshold)
complete_governing_equation (the repair)
B_mode(t) = r(t)·e^{iφ(t)} seated at q = m/n; lock ⇔ dφ/dt → 0 with r > 0 (a zero-object 0_[φ] in rotation, seat kept); alarm: dφ/dt < ε AND Δφ_coils locked — dimensionless, machine-free
lost_channels
address (machine normalization) and phase (rotation state)
structured_residual
{ modulus: the machine-specific thresholds (retired by the dimensionless criterion), phase: dφ/dt carried explicitly, transition: the lock event 0_[φ] with seat kept — residual = the unexecuted cross-machine deployment test }
correction_rule
Correction = reconstruction from the recovery: restored arguments, seats, phases and arms in the equation itself. No raw appended constants; any finite term must be derived from the recovered channels, never fitted.
phase_retained
True — the repair carries phase, seat, rank and both arms explicitly
falsifier
Train a phase-channel-only predictor on JET; deploy on ASDEX-U with no target data. If warning-time distribution degrades as severely as the magnitude-trained baseline, the recovery is wrong.
standing
SUPPORTED — recovery identified and demonstrated cold; not yet run against the machines' datasets
promotion_authority
YELLOW — no GREEN until the staked falsifier is executed
generated_utc
2026-08-31T03:13:23.100Z

Posted by

in