fix(IMP-05): complete V4 fallback evidence and dedup qualifiers

Refs #5

- Add runtime template_id dedup in lookup_v4_match_with_fallback with
  first-occurrence reservation; duplicate ranks become audit evidence,
  not new fallback candidates.
- Add Step 9 candidate_evidence as the primary per-unit evidence field
  while keeping fallback_chain as a compat alias for legacy readers.
- Add Step 20 fallback_selection_count and selection_paths derived from
  comp_debug.v4_fallback_summary with defensive defaults; top-level
  overall enum unchanged.
- Tighten synthetic fallback tests for duplicate handling (rank-1 reject A
  + rank-2 use_as_is A + rank-3 distinct B → rank-3 wins) and add tests
  for candidate_evidence + alias equality and Step 20 qualifier presence
  with defensive defaults.
- Verify with pytest (10 passed) and smoke_frame_render --self-check
  (11/11 partials, IMP-04 F17 calibration intact).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

src/phase_z2_pipeline.py

@@ -491,6 +491,12 @@ def lookup_v4_match_with_fallback(
         return None, trace
 
     first_skip_reason: Optional[str] = None
+    # IMP-05 L4 dedup (Codex #14 ordering — Claude #16 placement precision) :
+    # first occurrence claims template_id for the chain regardless of decision
+    # (selected/non-direct/rejected/missing-contract/capacity-skipped). Defensive
+    # against V4 anomaly where same template_id appears at multiple ranks with
+    # different labels — first label/reason is preserved, later duplicates skip.
+    seen_template_ids: set[str] = set()
     for i, judgment in enumerate(judgments, start=1):
         match = _v4_match_from_judgment(section_id, judgment, rank=i)
         status = to_phase_z_status(match)
@@ -515,6 +521,15 @@ def lookup_v4_match_with_fallback(
             "reason": None,
         }
 
+        # IMP-05 L4 dedup — duplicate check BEFORE rank evaluation.
+        # First occurrence reserves template_id even if non-direct/rejected/skipped.
+        # Later rank with same template_id is skipped as duplicate, audit fields preserved.
+        if match.template_id in seen_template_ids:
+            candidate_trace["reason"] = "duplicate_template_id"
+            trace["candidates"].append(candidate_trace)
+            continue
+        seen_template_ids.add(match.template_id)
+
         if status not in MVP1_ALLOWED_STATUSES:
             candidate_trace["reason"] = f"phase_z_status_not_allowed:{status}"
         elif get_contract(match.template_id) is None:
@@ -1350,6 +1365,13 @@ def compute_slide_status(sections: list[MdxSection],
     else:
         overall = "PARTIAL_COVERAGE_WITH_VISUAL_REGRESSION"
 
+    # IMP-05 L3 (Codex #10 D4 / #17 idea F / Claude #21 idea J) — Step 20 qualifier fields.
+    # Additive only — top-level overall enum unchanged. Defensive defaults so non-fallback
+    # paths (empty v4_fallback_summary) do not crash and report 0 / [] cleanly.
+    _v4_fb_summary = comp_debug.get("v4_fallback_summary", {}) or {}
+    _fallback_selection_count = _v4_fb_summary.get("fallback_selection_count", 0)
+    _selection_paths = _v4_fb_summary.get("selection_paths", [])
+
     return {
         "rendered": True,
         "visual_check_passed": visual_passed,
@@ -1361,6 +1383,9 @@ def compute_slide_status(sections: list[MdxSection],
         "selection_path": "fallback_used" if fallback_selections else "rank_1",
         "fallback_used": bool(fallback_selections),
         "fallback_selections": fallback_selections,
+        # IMP-05 L3 qualifier fields — grouped near existing fallback fields for readability.
+        "fallback_selection_count": _fallback_selection_count,
+        "selection_paths": _selection_paths,
         "visual_fail_reasons": list(overflow.get("fail_reasons") or []),
         "adapter_needed_count": len(adapter_needed_units),
         "adapter_needed_units": adapter_needed_units,
@@ -2706,7 +2731,11 @@ def run_phase_z2_mvp1(
             "selection_path": unit.selection_path,
             "fallback_used": bool(unit.selection_path and "fallback" in unit.selection_path),
             "fallback_reason": unit.fallback_reason,
-            "fallback_chain": selection_trace.get("candidates", []),
+            # IMP-05 L2 (Codex #10 D4 / #16 idea A) — Step 9 per-unit candidate evidence.
+            # candidate_evidence is the primary field for future frontend / AI consumers.
+            # fallback_chain is kept as a compat alias for any pre-IMP-05 reader.
+            "candidate_evidence": selection_trace.get("candidates", []),
+            "fallback_chain": selection_trace.get("candidates", []),  # compat alias; prefer candidate_evidence
             "v4_candidates": [
                 {
                     "template_id": c.template_id,