fix: symmetric move detection in PR-review rules

[smochan]

Problem

A relocated function shows up in the graph diff as one node removed and another added sharing the same terminal name. A prior fix (PR #73, esc/pyvisHref) taught the removed-side removed-referenced rule to recognize a move and not flag it as a critical deletion — but the fix was one-sided. The added-side "new code" rules still misfired on the same relocation:

• formatQn was reported as a new-untested-hotspot after moving app.js → ui/helpers.js
• moved orphans were reported as new-dead-code

A relocated symbol is not new code; its coverage and reachability state predate the PR that moved it.

Fix

Make move detection symmetric:

• Extract the inline move check into a shared _is_moved(change, counterpart) helper plus precomputed added_names / removed_names sets in evaluate_rules.
• removed_referenced now uses the helper (behavior unchanged).
• new_dead_code and new_untested_hotspot now skip symbols whose terminal name + kind appears in the removed set.

Every built-in rule that treats a node as newly-added or removed code now accounts for moves. The generic added_node / removed_node whens have no built-in rule, so there is no remaining gap.

Tests

• test_new_untested_hotspot_skips_moved_symbol (regression)
• test_new_untested_hotspot_still_fires_for_genuinely_new (control — genuinely new high-fan-in untested fn still fires)
• test_new_dead_code_skips_moved_symbol (regression)

pytest tests/ -k review → 40 passed.

🤖 Generated with Claude Code

[github-actions]

codegraph PR review

Diff vs main · severity ≤ high

codegraph review (target: main)

Diff: +4 / -0 / ~0 nodes, +46 / -4 edges

Findings (0)

No findings.

_{Triggered by codegraph CI · last run}