Errors and Orbit Correction

简体中文 · Previous: Nonlinear dynamics · Next: Results and evidence

Errors & Orbit Correction is the durable CPU v0.9 workspace. It applies one reviewed immutable portable-FODO profile to the current v0.2 project and produces error, BPM response/SVD correction, and finite-seed tolerance evidence.

Fixed reviewed profile

Item	Configuration
Seed ensemble	master 20260821, explicit seeds 0…7
Error scales	0, 1, 8
Components	dx/dy alignment, roll, normal-field relative
BPMs	8 reviewed stable occurrences with offset/gain/resolution
Correctors	8 H + 8 V, hard limit ±400 µrad
Response	central h=1 µrad and h/2
SVD	cutoff 1e-6, required rank 8/8, condition ≤ 1e8
Yield	target 0.90, Wilson 95%, passing/failing bracket
Scale	24 ensemble trials + one single-seed PyAT comparison, CPU only

These are read-only reviewed choices, not a generic orbit-correction knob editor.

Configure a v0.9 target

Set all three together:

v0.9 dedicated state directory: /absolute/dedicated/path
v0.9 exact source revision:     3b14dbcc5e525235a44db32ad96e20be5c33e3ef
v0.9 build ID:                  source:3b14dbcc5e52

The Engine operator should pre-create a dedicated formal state directory. The executable must point to the same exact build. Do not reuse a v0.8/5253 target.

Run

1. Open protocol/orbit_error_model_v0_1/fixtures/portable-fodo-reference-ring.studio-project.json or an appropriately reviewed v0.2 project.
2. Select the v0.9 target and open Calculate → Errors & Orbit Correction.
3. Wait for exact capability negotiation.
4. Optionally show advanced read-only configuration and raw evidence.
5. Click Run durable orbit correction v0.9 and confirm the screening boundary.
6. Use Detach, Reconnect, or Cancel remote job as needed.
7. Reopen a completed job from Retained formal results.

Read the result

• Horizontal / Vertical Orbit: pre/post closed orbit at BPMs against canonical lattice s.
• Correctors: H/V command in µrad against ±400 µrad; saturation fails the gate.
• Response & SVD: rank, condition, singular values, and h/h/2 convergence.
• Yield & Tolerance: per-scale/seed success, residual/max orbit, saturation, Wilson interval, and bracket.
• Evidence: four artifacts, Engine origin, checks, assumptions, applicability, provenance, and copyable reviewed configuration.

A solver return is not success. Response convergence, SVD rank/condition, residuals, and hard limits must all pass. Non-converged, deficient, ill-conditioned, or saturated data remains diagnostic without qualification evidence.

Interpretation boundary

Results are planar reference-δ 4D ideal/reference-model SCREENING, with MACHINE PREDICTION: FALSE. An eight-seed Wilson interval is not machine tolerance, and PyAT agreement is not measurement validation.

Strong coupling/skew, dispersion/chromaticity correction, LOCO, survey/measured response, hysteresis, power-supply dynamics, drift, DA/FMA/lifetime under errors, GPU, and SLURM are absent. Correction success cannot establish DA, LMA, Touschek lifetime, commissioning readiness, or an operable machine.