diff --git a/cli/trellis b/cli/trellis
index 8877211..3846134 100755
--- a/cli/trellis
+++ b/cli/trellis
@@ -267,28 +267,89 @@ def yaml_read_nested(text, parent, field):
     return None
 
 
+def parse_numeric_scalar(value):
+    """Parse a YAML scalar that should contain a numeric value."""
+    if value in (None, "", "null", "{}"):
+        return None
+    try:
+        return float(str(value).strip().strip('"').strip("'"))
+    except (TypeError, ValueError):
+        return None
+
+
+def extract_self_eval_score(text):
+    """Read the recorded self-eval score from supported spec shapes."""
+    for parent in ("self_eval", "perf_eval"):
+        for field in ("total", "score"):
+            score = parse_numeric_scalar(yaml_read_nested(text, parent, field))
+            if score is not None:
+                return score
+        score = parse_numeric_scalar(yaml_read_field(text, parent))
+        if score is not None:
+            return score
+
+    return parse_numeric_scalar(yaml_read_field(text, "score"))
+
+
+def parse_phase_statuses(text):
+    """Parse phase statuses from the phases block without counting unrelated statuses."""
+    lines = text.splitlines()
+    statuses = []
+    in_phases = False
+    i = 0
+
+    while i < len(lines):
+        line = lines[i]
+        stripped = line.strip()
+        indent = len(line) - len(line.lstrip())
+
+        if not in_phases:
+            if re.match(r'^phases:\s*$', line):
+                in_phases = True
+            i += 1
+            continue
+
+        if stripped and indent == 0:
+            break
+
+        match = re.match(r'^\s+-\s+id:\s*"?(phase\d+)"?', line)
+        if not match:
+            i += 1
+            continue
+
+        item_indent = len(match.group(0)) - len(match.group(0).lstrip())
+        status = "pending"
+        i += 1
+
+        while i < len(lines):
+            field_line = lines[i]
+            field_stripped = field_line.strip()
+            field_indent = len(field_line) - len(field_line.lstrip())
+
+            if field_stripped and field_indent <= item_indent:
+                break
+
+            status_match = re.match(
+                r'^\s+status:\s*"?(pending|in_progress|completed|failed|skipped)"?',
+                field_line,
+            )
+            if status_match:
+                status = status_match.group(1)
+
+            i += 1
+
+        statuses.append(status)
+
+    return statuses
+
+
 def count_phases(text):
     """Count phases and their statuses."""
-    total = 0
-    completed = 0
-    failed = 0
-    in_progress = 0
-    for m in re.finditer(r'^\s+-\s+id:\s*"?phase\d+"?', text, re.MULTILINE):
-        total += 1
-    for m in re.finditer(r'^\s+status:\s*"?completed"?', text, re.MULTILINE):
-        completed += 1
-    for m in re.finditer(r'^\s+status:\s*"?failed"?', text, re.MULTILINE):
-        failed += 1
-    for m in re.finditer(r'^\s+status:\s*"?in_progress"?', text, re.MULTILINE):
-        in_progress += 1
-    # Subtract the top-level status from phase counts
-    top_status = yaml_read_field(text, "status")
-    if top_status == "completed":
-        completed = max(0, completed - 1)
-    elif top_status == "failed":
-        failed = max(0, failed - 1)
-    elif top_status == "in_progress":
-        in_progress = max(0, in_progress - 1)
+    phase_statuses = parse_phase_statuses(text)
+    total = len(phase_statuses)
+    completed = sum(1 for status in phase_statuses if status == "completed")
+    failed = sum(1 for status in phase_statuses if status == "failed")
+    in_progress = sum(1 for status in phase_statuses if status == "in_progress")
     return total, completed, failed, in_progress
 
 
@@ -1073,18 +1134,16 @@ def cmd_start(args):
 
 def check_self_eval(text, task_id):
     """Warn if self-eval looks like a rubber stamp."""
-    # Look for self_eval or perf_eval score
-    score_match = re.search(r'(?:self_eval|perf_eval|score):\s*(\d+)', text)
-    if not score_match:
+    score = extract_self_eval_score(text)
+    if score is None:
         print(f"  {c(C_YELLOW, 'warn')}: no self-eval score found in spec")
         return
-
-    score = int(score_match.group(1))
     has_deviations = bool(re.search(r'deviations:', text, re.MULTILINE))
     has_improvements = bool(re.search(r'improvements:', text, re.MULTILINE))
+    score_display = int(score) if float(score).is_integer() else score
 
     if score >= 9 and not has_deviations and not has_improvements:
-        print(f"  {c(C_YELLOW, 'warn')}: self-eval {score}/10 with no deviations or improvements noted")
+        print(f"  {c(C_YELLOW, 'warn')}: self-eval {score_display}/10 with no deviations or improvements noted")
         print(f"         scores above 8 should document at least one deviation or improvement")
     elif score == 10:
         print(f"  {c(C_YELLOW, 'note')}: perfect 10/10 - are you sure? 10 means flawless with improvements beyond spec")
@@ -1112,7 +1171,7 @@ def cmd_complete(args):
     text = spec.read_text()
 
     # Check for exec results - warn if no criteria were executed
-    has_results = bool(re.search(r'result:\s*"?(pass|fail)"?', text, re.MULTILINE))
+    has_results = any(ac.get("result") in ("pass", "fail") for ac in parse_acceptance_criteria(text))
     if not has_results:
         print(f"  {c(C_YELLOW, 'warn')}: no exec results recorded. Run '{c(C_BOLD, f'trellis exec {args.task_id}')}' first")
 
@@ -2333,13 +2392,14 @@ def cmd_report(args):
         risks[risk] = risks.get(risk, 0) + 1
 
         # Self-eval scores
-        score_match = re.search(r'total:\s*(\d+)', text)
-        if score_match:
-            self_eval_scores.append(int(score_match.group(1)))
+        score = extract_self_eval_score(text)
+        if score is not None:
+            self_eval_scores.append(score)
 
         # Exec results
-        passes = len(re.findall(r'result:\s*"?pass"?', text))
-        fails = len(re.findall(r'result:\s*"?fail"?', text))
+        criteria = parse_acceptance_criteria(text)
+        passes = sum(1 for ac in criteria if ac.get("result") == "pass")
+        fails = sum(1 for ac in criteria if ac.get("result") == "fail")
         if passes or fails:
             exec_pass += passes
             exec_fail += fails
diff --git a/tests/review_gate_smoke.sh b/tests/review_gate_smoke.sh
index 5f989c2..a3f14ad 100755
--- a/tests/review_gate_smoke.sh
+++ b/tests/review_gate_smoke.sh
@@ -739,6 +739,218 @@ EOF
   assert_contains "$spec_text" 'Command timed out after 1s' "spec should record the configured timeout in result_output"
 }
 
+case_complete_nested_exec_and_self_eval() {
+  local repo task_id output archive_path spec_text
+  repo="$(new_repo)"
+  task_id="complete-nested-exec-and-self-eval"
+  write_changed_file "$repo"
+
+  cat > "$repo/.ai/specs/active/$task_id.yaml" <<EOF
+spec_version: "1.1"
+task_id: "$task_id"
+created: "2026-03-26T00:00:00Z"
+updated: "2026-03-26T00:00:00Z"
+status: "in_progress"
+
+task:
+  title: "Nested exec and self-eval"
+  summary: "Ensure trellis complete recognizes nested acceptance results and self_eval totals"
+  size: "small"
+  risk_level: "low"
+
+phases:
+  - id: "phase1"
+    name: "Nested result shape"
+    objective: "Exercise nested acceptance result parsing"
+    changes:
+      - file: "app.txt"
+        action: "update"
+        lines: "1"
+        content_spec: "Smoke change"
+    acceptance_criteria:
+      - id: "ac1_1"
+        type: "test"
+        description: "Nested result should count as executed"
+        command: "printf '1 example, 0 failures\\n'"
+        expected: "0 failures"
+        result:
+          status: "pass"
+          timestamp: "2026-03-26T00:00:00Z"
+          output: "1 example, 0 failures"
+    status: "completed"
+
+planning_log:
+  - timestamp: "2026-03-26T00:00:00Z"
+    actor: "user"
+    summary: "Bootstrap smoke fixture"
+
+self_eval:
+  completeness: 3
+  architecture_fidelity: 3
+  spec_alignment: 1
+  validation_depth: 1
+  total: 8.8
+  notes: "Nested score fixture"
+  second_pass_performed: false
+
+deviations: []
+EOF
+
+  write_review_v3 "$repo" "$task_id" "pass" "executor" "completed" "pass" "pass" "pass" "pass" "pass" \
+    "No issues found — checked callers of app.txt." \
+    "No issues found — checked AGENTS.md and CONVENTIONS.md." \
+    "No issues found — checked hardcodes and null handling." \
+    "None." \
+    "None."
+
+  capture output bash -lc "cd '$repo' && PATH='$CLI_ROOT':\"\$PATH\" trellis complete '$task_id'"
+  assert_not_contains "$output" "no exec results recorded" "complete should recognize nested acceptance results recorded by trellis exec"
+  assert_not_contains "$output" "no self-eval score found in spec" "complete should recognize nested self_eval totals"
+  archive_path="$(archive_spec_path "$repo" "$task_id")"
+  [ -n "$archive_path" ] || fail "complete should archive the nested exec fixture"
+  spec_text="$(cat "$archive_path")"
+  assert_contains "$spec_text" 'status: "completed"' "archived spec should remain completed"
+}
+
+case_report_nested_exec_and_self_eval() {
+  local repo task_id output
+  repo="$(new_repo)"
+  task_id="report-nested-exec-and-self-eval"
+
+  mkdir -p "$repo/.ai/specs/archive/2026-03"
+  cat > "$repo/.ai/specs/archive/2026-03/$task_id.yaml" <<EOF
+spec_version: "1.1"
+task_id: "$task_id"
+created: "2026-03-26T00:00:00Z"
+updated: "2026-03-26T00:00:00Z"
+status: "completed"
+
+task:
+  title: "Report nested parsing"
+  summary: "Ensure trellis report counts nested execution results and decimal self-eval totals"
+  size: "small"
+  risk_level: "low"
+
+phases:
+  - id: "phase1"
+    name: "Nested result shape"
+    objective: "Exercise nested acceptance result parsing in report"
+    changes:
+      - file: "app.txt"
+        action: "update"
+        lines: "1"
+        content_spec: "Smoke change"
+    acceptance_criteria:
+      - id: "ac1_1"
+        type: "test"
+        description: "Nested pass result should count"
+        command: "printf '1 example, 0 failures\\n'"
+        expected: "0 failures"
+        result:
+          status: "pass"
+          timestamp: "2026-03-26T00:00:00Z"
+          output: "1 example, 0 failures"
+    status: "completed"
+
+planning_log:
+  - timestamp: "2026-03-26T00:00:00Z"
+    actor: "user"
+    summary: "Bootstrap smoke fixture"
+
+self_eval:
+  completeness: 3
+  architecture_fidelity: 3
+  spec_alignment: 1
+  validation_depth: 1
+  total: 8.8
+  notes: "Nested score fixture"
+  second_pass_performed: false
+
+deviations: []
+EOF
+
+  capture output bash -lc "cd '$repo' && PATH='$CLI_ROOT':\"\$PATH\" trellis report"
+  assert_contains "$output" "avg: 8.8/10  (1 scored)" "report should read decimal self-eval totals from the nested block"
+  assert_contains "$output" "1 passed / 0 failed  (100% pass rate)" "report should count nested acceptance results"
+}
+
+case_status_phase_count_ignores_top_level_status() {
+  local repo task_id output
+  repo="$(new_repo)"
+  task_id="status-phase-count"
+
+  mkdir -p "$repo/.ai/specs/archive/2026-03"
+  cat > "$repo/.ai/specs/archive/2026-03/$task_id.yaml" <<EOF
+spec_version: "1.1"
+task_id: "$task_id"
+created: "2026-03-26T00:00:00Z"
+updated: "2026-03-26T00:00:00Z"
+status: "completed"
+
+task:
+  title: "Status phase counts"
+  summary: "Ensure trellis status only counts statuses from the phases section"
+  size: "small"
+  risk_level: "low"
+
+task_notes:
+  status: "completed"
+
+phases:
+  - id: "phase1"
+    name: "One"
+    objective: "One"
+    changes:
+      - file: "app.txt"
+        action: "update"
+        content_spec: "One"
+    acceptance_criteria:
+      - id: "ac1_1"
+        type: "custom"
+        description: "One"
+        result:
+          status: "pass"
+    status: "completed"
+  - id: "phase2"
+    name: "Two"
+    objective: "Two"
+    changes:
+      - file: "app.txt"
+        action: "update"
+        content_spec: "Two"
+    acceptance_criteria:
+      - id: "ac2_1"
+        type: "custom"
+        description: "Two"
+        result:
+          status: "pass"
+    status: "completed"
+  - id: "phase3"
+    name: "Three"
+    objective: "Three"
+    changes:
+      - file: "app.txt"
+        action: "update"
+        content_spec: "Three"
+    acceptance_criteria:
+      - id: "ac3_1"
+        type: "custom"
+        description: "Three"
+        result:
+          status: "pass"
+    status: "completed"
+
+planning_log:
+  - timestamp: "2026-03-26T00:00:00Z"
+    actor: "user"
+    summary: "Bootstrap smoke fixture"
+EOF
+
+  capture output bash -lc "cd '$repo' && PATH='$CLI_ROOT':\"\$PATH\" trellis status '$task_id'"
+  assert_contains "$output" "phases: 3 done  (3 total)" "status should count only phase statuses and not subtract the top-level spec status"
+  assert_not_contains "$output" "1 pending" "status should not invent a pending phase when all phase statuses are completed"
+}
+
 case_all() {
   case_smoke_bootstrap
   case_review_pass_topology
@@ -752,6 +964,9 @@ case_all() {
   case_non_mutating_review
   case_exec_resume_nested_results
   case_exec_timeout_override
+  case_complete_nested_exec_and_self_eval
+  case_report_nested_exec_and_self_eval
+  case_status_phase_count_ignores_top_level_status
 }
 
 main() {
@@ -769,6 +984,9 @@ main() {
     non-mutating-review) case_non_mutating_review ;;
     exec-resume-nested-results) case_exec_resume_nested_results ;;
     exec-timeout-override) case_exec_timeout_override ;;
+    complete-nested-exec-and-self-eval) case_complete_nested_exec_and_self_eval ;;
+    report-nested-exec-and-self-eval) case_report_nested_exec_and_self_eval ;;
+    status-phase-count-ignores-top-level-status) case_status_phase_count_ignores_top_level_status ;;
     all) case_all ;;
     *)
       fail "unknown case: $action"