PSTH with numpy.histogram shows periodic “gaps” at bin boundaries even after integer time conversion—how to bin robustly?

I’m computing peristimulus time histograms (PSTHs) in Python. Each trial aligns spikes to a reach-start timestamp and bins them into fixed-width bins. I see vertical “gaps” (low counts) at exact bin boundaries (e.g., every 20 ms). I thought it was floating-point, so I converted everything to integers and even tried microseconds + half-open bins, but the stripes persist.

Minimal example (synthetic data that reproduces the effect on my machine):

import numpy as np

# ---- synthetic spikes: uniform + weak locking every 20 ms ----
rng = np.random.default_rng(0)
n_trials = 200
pre_s, post_s, bw_s = 1.0, 4.0, 0.02   # 20 ms
TICK = 1_000_000                        # microseconds
pre, post, bw = int(pre_s*TICK), int(post_s*TICK), int(bw_s*TICK)
edges_rel = np.arange(-pre, post+1, bw, dtype=np.int64)

# trial starts (ms), here zeros for simplicity
reach_ticks = np.zeros(n_trials, dtype=np.int64)

# build spikes per trial with slight bin-boundary bias
spike_ticks = []
for _ in range(n_trials):
    # uniform spikes
    base = rng.integers(-pre, post, size=300)
    # add a few spikes perturbed around multiples of 20 ms
    lock = (np.arange(-pre, post, 20_000) + rng.integers(-200, 200, size=(pre+post)//20_000))
    spikes = np.concatenate([base, lock])
    spikes.sort()
    spike_ticks.append(spikes)

# histogram per trial (half-open [left,right) bins)
H = []
for rs, rels in zip(reach_ticks, spike_ticks):
    edges_abs = rs + edges_rel
    h, _ = np.histogram(rels, bins=edges_abs)
    H.append(h)
H = np.asarray(H, float) / bw_s  # Hz

# show that the middle bin near  -0.5 s dips relative to neighbors
centers = (edges_rel[:-1] + edges_rel[1:])/(2*TICK)
mid = np.argmin(np.abs(centers + 0.5))
print("Means around -0.5s:", H.mean(0)[mid-1:mid+2])