Ear check: is the grit in the file?

Raw generations, converted once to 192k mp3 for this page. Nothing else touched them.

1 · New take 1, first 3 lines

2 · New take 2, opening

3 · New take 3, opening

4 · 02-SHIP, the shipped read (reference)

The exact generation settings, all of them

POST /v1/text-to-speech/naAstj4zS53vbYiQEiQp?output_format=pcm_24000
model_id: eleven_multilingual_v2
voice: Yair Pro (professional clone)
stability 0.20 · similarity 0.80 · style 0.90 · speaker boost ON · speed 1.1
no seed, no normalization override, no gain parameter (none exists)

measured: peaks -0.7 to -1.6 dB, zero clipped samples, all three takes
02-SHIP reference: peak -0.27 dB, same settings, same voice