Minimal before/after prompt comparisons

[tobs-code]

Looking for tiny examples:

• 1 natural language prompt
• 1 SoftPrompt-IR structured version
• short note on perceived difference (or none)

No benchmarks required.
Subjective impressions are fine.

[tobs-code]

I’ll summarize findings here once we have a few data points.