Build notes

Decisions, diagrams, and what I learned.

• Solvability check: can every target card actually be built? →

  Checking a timed card game's fairness exhaustively, instead of by feel.

  Math Attack

• Credit spend gates: no blind re-rolls →

  Gating generation spend, and turning every failed generation into a written rule.

  Agentic workflows and custom skills

• Resolved errors should stay resolved →

  The loop that turns a failure in one session into a rule for every session after.

  Agentic workflows and custom skills

• The cheaper model made things up →

  Choosing a vision model by testing both on the same 99 slides.

  NTULearn toolkit