Questions tested, not merely repeated.
Independent investigations with visible methods, observations, and limitations. Public or synthetic inputs only.
complete
Agent cost scales with the square of steps, not linearly
A cost-model experiment showing why a multi-step AI agent is far more expensive than the same number of independent calls — because each step re-sends the whole growing transcript as input.
Reproducible
complete
Does prompt caching actually cut LLM cost? A cost-model experiment
A first-principles cost model showing how much prompt caching saves as a function of two variables you control: the share of input that is a stable prefix, and how often it repeats.
Reproducible
complete
Why two 60-character titles truncate differently in search
A demonstration that search snippets are cut by rendered pixel width, not character count — so a title full of wide letters truncates sooner than a same-length title of narrow ones.
Reproducible