Where the refusal layer comes from, what removing it changes, and the four things it never touches.
The four-layer template, five adjustments that raise success rate, and how to tell when prompting is hopeless.
Register, key, base URL, model name, client config, verification. Plus the error-code table.
Three difficulty tiers across 4o, 4o-mini, o-series and GPT-5, with measured pass rates.
Why it refuses more than GPT, how Sonnet and Opus differ, and what's left after the layer comes off.
Pro vs Flash, whole-repository auditing, and the one job nothing else does as well.
Why it opens up more readily than GPT — and the keyword filter that prompts can't get past.