Verification — tree-gen (Stream 1a)
Verification — tree-gen (Stream 1a)
Date: 2026-07-11 · Runner: Claude (agent, following SKILL.md)
Input: "Does remote work increase productivity?"
Output: verification/2026-07-11-run.tree.json · Gold: gold/remote-work.tree.json (from ../../claim-tree-annotation.md)
Diff vs gold
| Gold branch | Generated? | Notes |
|---|---|---|
whom — for whom / which roles | yes (whom, children whom-role/whom-work) | same distinctions, slightly different slugs & phrasing |
tasks — for what tasks (solo vs collaborative) | yes, matching ids tasks-solo/tasks-collab | near-identical |
time — time horizon (short vs long) | yes, matching ids | near-identical |
conditions — setup / management | yes, matching ids | near-identical |
| — | extra: measure (how is productivity measured?) | not in gold; a defensible extra distinction, not counted against |
Mechanical checks (python): valid JSON, unique kebab-case ids, root id q-root, kind consistent (question throughout), depth ≤ 3, leaves have children: [].
Verdict
Pass by the PLAN.md criterion — the generated tree covers all four gold major sub-questions (for whom / what tasks / what horizon / what conditions) with equivalent leaf distinctions. One extra branch (measure) beyond gold; acceptable per SKILL.md ("rephrasing is fine — coverage of the major distinctions is what matters").
Still to do per SKILL.md: try one new claim to check generalisation.
