BuilderOS 3.2.0

Changed: rounds are dry and show where they stand

pressure-testing was run live on a real decision and compared with Pocock's grilling; the gaps the run showed were decided in the project thread "Migliorare la skill pressure-testing" (1A 2A 3A 4A, plus a drier visual format).

• One dry shape for every round. 🎯 the claim in its strongest form and the assumption it rests on, open to correction; ❓ each question with options as A option → consequence; ➡️ one recommendation in one clause; 📍 decided, open, deferred and the next branch; ✍️ how to answer. The recommendation is the strongest option, not the safest.
• Three kinds of fact. A fact the session can retrieve is looked up; a fact only the person holds is one plain line with no options, and the decisions resting on it wait a round; a fact neither side holds gets the ways to close the gap as options.
• A plain close. The resolution table is followed by three lines: what was decided, what is still at risk, what will be tested and by when.
• Leaner. pressure-testing from 1,590 to about 1,080 words; the fact-gap template, working ahead and round recording moved to references/round-formats.md. The round cards in stack-decisions, design-direction, domain-model and the README use the same markers.

Added

• tests/scripts/rounds.test.mjs: the round template's shape, options with consequences and one-clause recommendations in every round example, plain person-only fact questions, no leftover old recommendation line, and a 1,100-word ceiling on pressure-testing.
• Eval case grill-a-decision: "grill me" gets one dry round with the opening claim, at most four questions, no guessed values for a fact only the user knows, and the closing count. Not run yet.