Deep dives by topic: formulas, assumptions, common mistakes and worked examples.
Do not spend the whole context on input.
Why the same text has different token counts across models.