Larger Relative Tolerance Variation in `splinempe`

[ric-evans]

In test-run-realistic (splinempe, run00136662-evt000035405932-BRONZE.pkl), the measured variation is seen at 0.1689 for a single pixel. This is higher than the accepted 0.1.

[
         [
            1,
            1261.4054977969026,
            NaN,
            NaN
         ],
         [
            1,
            1079.0788843076734,
            NaN,
            NaN
         ],
         [
            0.0,
            0.16896504614322816,
            NaN,
            NaN
         ],
         [
            true,
            false,
            true,
            true
         ]
      ],
...

[tianluyuan]

It looks like the XYZT reconstructions are even more susceptible to this issue with reproducibility. Copying over results from one CI test still led to failures on the subsequent run. The CI jobs were run on the same machine so it's possible there's some randomness in the seed chain that's causing this.

[ric-evans]

It looks like the XYZT reconstructions are even more susceptible to this issue with reproducibility. Copying over results from one CI test still led to failures on the subsequent run. The CI jobs were run on the same machine so it's possible there's some randomness in the seed chain that's causing this.

as seen in #281