The Two Literacies

The first literacy: SPEC

Specification is the discipline of externalizing intent before generating anything. A working spec names five things:

• Goal — what the output is for, and what done looks like.
• Context — what the system cannot know unless you tell it.
• Constraints — what must and must not appear.
• Format — the shape the output has to take to be usable.
• Verification — how you will check it, decided before you generate.

Writing the spec is not overhead. It is a load-reduction move: the thinking is done once, on paper, instead of repeatedly and badly in the middle of a generate-and-hope cycle.

The second literacy: CATCH

Audit is the perceptual discipline of recognizing failure before it ships. AI failure is not random — it clusters in recognizable patterns: the fluent fabrication, the confident omission, the subtly shifted claim, the stale fact, the wrong-but-plausible number, the borrowed structure that doesn't fit, the answer to a question you didn't ask. CATCH trains the habit of auditing output against source, data, and spec — every time, in a fixed order — so that the patterns become visible before the work leaves your hands.

The Loop

The two literacies compound only as a practiced sequence. Specify before you generate; audit before you accept; save what survived the audit so the next spec starts further ahead. The Loop is what turns two skills into a working discipline — and what makes the discipline hold across a real working week.