v0.3.0

Added

• Step 4 — Contest the findings, with references/contest.md. Every finding from Steps 2 and 3 is a hypothesis until it passes evidence (cite the moment), counter-evidence (find the moment that contradicts it; a rule founded on fewer than three moments is a hunch), and mechanism (would the fix have changed the cited moment, and does it conflict with a sibling skill's rule). Verdict per finding: confirmed, reframed, or falls — only survivors reach the Wiring section. A first pass here was two-of-seven wrong and missed the strongest item, which is what the step exists to catch.
• Step 1 — "Was it seen?": before the score is final, list every user-facing surface the session changed and who looked at it — a capture, a run, a person, or nobody. A surface nobody looked at is a finding whatever the tests said.
• A score cap of 6 for any session that shipped an unseen surface, overriding the scoring table in references/shapes.md. One session scored 7/10 on context economy alone and came back with nine defects from a single play.
• Contested section in the output — finding, verdict, and the moment it stood or fell on.

Changed

• "Write the notes" renumbered Step 4 → Step 5; Do NOT Load updated to match.
• Step 2: magnitude stays relative except where the session's own tool output printed a number (a suite runtime, a git log count) — cite it, with its source.
• Step 3: an unseen surface gets its lever from the gate of the artifact that shipped it, never a note.