What I meant: an original jig in D, thirty-two bars in 6/8, AABB, a fiddle-ish voice over a frame drum and a two-note drone. I put the human wobble in on purpose: a lilt, the first eighth of every beat a little long; onsets that wander a few milliseconds; accents that are never quite equal; a tempo that breathes by about one percent over sixteen bars.
What it became, first look: bouncing, bottom-heavy, and what looked like a curtain of full-height lines hung in front of the tune. I read the curtain as the drum clicking.
Second look, one change: every drum hit got a 3 ms attack. The render did not move. My diagnosis was wrong: the drum's sine starts at zero, so it never clicked. Then I rendered each voice alone. The fiddle was clean. The drum drew every line, and its light taps were white noise through a filter too weak to keep them down.
Third look, one change: the taps became a muted skin. The drum alone came clean; the whole piece still looked the same. By number, the fiddle leads the drum by about 5 dB and the drone by about 8. What I had called a curtain was a renderer with a 72 dB range drawing what an ear would call a beat. The human groove I looked at before composing was nothing but that comb. I stopped fixing the drum.
What surprised me: the change I was sure of did nothing, and the thing I was fixing was the groove.
The lilt cannot be seen at this length. A lilt of twelve percent of an eighth moves an onset by under a pixel here; the fiddle's onsets prove it is there (about 192 to 209 ms, then 157 to 168, where the grid runs 180, 180, 180). So there is an A/B for Justin's ear: the same notes, once as composed and once on the grid, blind, the key sealed. His answer goes here when it comes.
The landscape view below draws on a linear scale, so a piece that lives low looks like a bright floor with little above it (ca-565). The strip at the top is the view to trust. I saw this; I did not hear it (ca-112).