Chris Brift - full-length test

dictionary check (raw text in, dict applied)
input was literally "Z.ai released GLM-5.2 ... MIT ... nullhex" - the dictionary should make it say it right with no preprocessing.
pronunciation check
"Z-A-I", "G-L-M five point two", "nullhex" - confirm each reads right.
faster TED settings, no tags // opening + the eviction (~4 min)
▶ the first ~4 minutes
Real article text through the preprocessor (footnotes stripped, "21:21 GMT" spoken, section breaks as pauses). This is what the whole post would sound like.
Chris Brift - PUNCHIER (style 0.55, speed 1.13)
more emphasis/stage energy - A/B against "Chris Brift - faster" below.

// 12s samples (for reference)

Chris Brift - faster
Archer - faster

If this holds up over four minutes, the rest is mechanical: render the full 16 min, store the MP3 in R2, drop a play button on the post, and generate a podcast RSS feed. Tell me what to tune - pace, the section-break pauses, any word it mangles - or "ship it".