The backtrack board — real data

A decision walked backwards to the piece of context that caused it. There are only two kinds of decision point — an LLM decision (the model chose from what it was shown) or a deterministic decide() rule — and backtracking works on both. Pick any one below to walk it back.

✓ GENERATED — every score, rank, verdict, honesty line & custody pane computed from a real localizeContextBug → toBacktrackTrace run. Nothing hand-authored. (demo/gen-backtrack-board.ts)