Gamma pass rate drops sharply when fluence width hits 400 columns (resolution ~= 1.0 mm)

[ekamperi]

Gamma pass rate is stable (~100%) for resolutions around 1.0 mm, but drops sharply to ~78% exactly when the fluence grid width is 400 columns (resolution <= 1.0). A tiny change to 1.0001 mm (grid width 399) restores 100% pass. This suggests a discretization artifact tied to integer grid width / rounding of leaf/jaw positions.

To Reproduce
This is from a Varian TrueBeam logfile with HD120 MLC.

import matplotlib.pyplot as plt
from pylinac import load_log

f = "telemetry_1351-25_Plan_new_Arc1_20251215110840.bin"
l = load_log(f)

rv = []
avg_gamma = -1
for res in range(1, 20, 1):
    l.fluence.gamma.calc_map(resolution=0.1 * res)
    avg_gamma = l.fluence.gamma.pass_prcnt
    rv.append((0.1 * res, avg_gamma))
    print("Column %4d Res %.2f  Gamma%% %.2f" % (l.fluence.gamma.array.shape[1], 0.1 * res, avg_gamma))

x, y = zip(*rv)
plt.plot(x, y, marker='o')
plt.xlabel('Resolution')
plt.ylabel('Gamma pass %')
plt.show()

Expected behavior
The gamma passing rate shouldn't dip to 78% when for res < 1 and res > 1 it's ~100%.

Screenshots

[ekamperi]

And this is when zooming in around res=1.0:

[ekamperi]

This line apparently is relevant for the behavior I am seeing.

[crcrewso]

I've confirmed this behavior with Python 3.12 and 3.14, Pylinac 3.38.

I explored the range 0.1 to 4

There appears to be at least 4 nodes (0.991 to 1.0001), (1.1970 to 1.2001), (3.5913 to 3.5927), and (3.5950 to 3.5953)

Update:
Further inspection shows issues around 5.2, 7.8, and 9.2.

[crcrewso]

This line apparently is relevant for the behavior I am seeing.

Would it make sense to change 562 to

pos_offset = int(len(fluence_line)/2)