Fail closed when nested validation tunables are not YAML numbers

bool is a subclass of int, so av_sync.tolerance_sec: true became 1.0s
OCR slack and ocr.min_confidence: true became 1.

Co-authored-by: jmjava <jmjava@gmail.com>

Author: cursoragent

milestones/README.md

@@ -5,11 +5,14 @@ repositories that install `docgen` and maintain their own demo bundle. The
 library no longer ships an in-repo dogfood; consumers are the integration test
 of record.
 
-**Active:** **[generation-numeric-tunables.md](generation-numeric-tunables.md)** —
-narration / scene-generation temperature and context-byte tunables must
-be YAML numbers (`true` used to become `1.0` / 1-byte context).
+**Active:** **[validation-numeric-tunables.md](validation-numeric-tunables.md)** —
+nested validation OCR / layout / av_sync / timing / story_end numerics
+must be YAML numbers (`true` used to become `1`).
 
 **Shipped:**
+- **[generation-numeric-tunables.md](generation-numeric-tunables.md)** —
+  narration / scene-generation temperature and context-byte tunables
+  must be YAML numbers (#117).
 - **[hint-segment-create-bool.md](hint-segment-create-bool.md)** —
   hint `docgen.segment.create` must be a YAML boolean (#116).
 - **[numeric-config-tunables.md](numeric-config-tunables.md)** —

milestones/generation-numeric-tunables.md

@@ -1,6 +1,6 @@
 # Milestone: LLM generation numeric tunables must be YAML numbers
 
-**Status:** Active  
+**Status:** Shipped  
 **PR:** [#117](https://github.com/jmjava/documentation-generator/pull/117)  
 **Depends on:** `milestones/hint-segment-create-bool.md` (PR #116),
 `milestones/numeric-config-tunables.md` (PR #115),

milestones/validation-numeric-tunables.md

@@ -0,0 +1,50 @@
+# Milestone: nested validation numerics must be YAML numbers
+
+**Status:** Active  
+**PR:** (this PR)  
+**Depends on:** `milestones/generation-numeric-tunables.md` (PR #117),
+`milestones/numeric-config-tunables.md` (PR #115)
+
+## Problem
+
+Top-level ``validation.max_drift_sec`` / ``max_freeze_ratio`` are
+already typed. Nested check tunables still go through ``int()`` /
+``float()``. A YAML **bool** is a subclass of ``int``, so:
+
+1. ``validation.av_sync.tolerance_sec: true`` became **1.0s** OCR
+   slack (vs default 3.0).
+2. ``validation.ocr.min_confidence: true`` became **1**.
+3. ``validation.story_end.max_early_sec: true`` became **1s**.
+
+A YAML list or string raised ``TypeError`` / ``ValueError`` inside
+validate, not ``ConfigError`` at load.
+
+## Goal
+
+Fail closed at ``Config.from_yaml``. Present values of:
+
+- ``validation.ocr.sample_interval_sec`` / ``min_confidence``
+- ``validation.layout.min_spacing_px`` / ``edge_margin_px``
+- ``validation.av_sync.tolerance_sec`` / ``min_anchors_per_segment`` /
+  ``max_anchors_per_segment``
+- ``validation.timing_sync.max_tail_gap_sec`` / ``max_end_overrun_sec``
+- ``validation.story_end.max_early_sec`` / ``max_early_ratio``
+
+must be YAML numbers (int or float, not bool). Missing keys keep
+defaults.
+
+## Done when
+
+- [x] Present nested tunables must be YAML numbers.
+- [x] Tests for bool ``tolerance_sec`` / ``max_early_sec``, list
+      ``min_confidence``, quoted ``max_tail_gap_sec``, and valid
+      numbers.
+- [x] `ruff check src/ tests/`
+- [x] `pytest tests/`
+- [x] `docgen benchmark` (no clock change)
+
+## Out of scope
+
+- ``validation.*.enabled`` / ``layout.check_overlap`` /
+  ``prefer_scene_spec_labels`` boolean gates.
+- Coercing numeric strings into numbers.

src/docgen/config.py

@@ -607,6 +607,42 @@ def __post_init__(self) -> None:
                 label="validation.narration_lint.post_tts_deny_patterns",
                 source=src,
             )
+        for nkey in ("sample_interval_sec", "min_confidence"):
+            if ocr.get(nkey) is not None:
+                require_yaml_number(
+                    ocr[nkey], label=f"validation.ocr.{nkey}", source=src
+                )
+        layout = self._sub_block(validation, "layout", label="validation.layout")
+        for nkey in ("min_spacing_px", "edge_margin_px"):
+            if layout.get(nkey) is not None:
+                require_yaml_number(
+                    layout[nkey], label=f"validation.layout.{nkey}", source=src
+                )
+        for nkey in (
+            "tolerance_sec",
+            "min_anchors_per_segment",
+            "max_anchors_per_segment",
+        ):
+            if avs.get(nkey) is not None:
+                require_yaml_number(
+                    avs[nkey], label=f"validation.av_sync.{nkey}", source=src
+                )
+        ts_sync = self._sub_block(
+            validation, "timing_sync", label="validation.timing_sync"
+        )
+        for nkey in ("max_tail_gap_sec", "max_end_overrun_sec"):
+            if ts_sync.get(nkey) is not None:
+                require_yaml_number(
+                    ts_sync[nkey],
+                    label=f"validation.timing_sync.{nkey}",
+                    source=src,
+                )
+        story = self._sub_block(validation, "story_end", label="validation.story_end")
+        for nkey in ("max_early_sec", "max_early_ratio"):
+            if story.get(nkey) is not None:
+                require_yaml_number(
+                    story[nkey], label=f"validation.story_end.{nkey}", source=src
+                )
 
     def _source_label(self) -> str:
         return self.yaml_path.name if self.yaml_path else "docgen.yaml"

tests/test_config.py

@@ -948,3 +948,65 @@ def test_from_yaml_generation_numeric_tunables_allowed(tmp_path: Path) -> None:
     assert c.raw["narration_from_source"]["max_context_bytes"] == 90000
     assert c.raw["manim_scene_generation"]["temperature"] == 0.4
     assert c.raw["manim_scene_generation"]["max_whisper_words_in_prompt"] == 12
+
+
+def test_from_yaml_bool_av_sync_tolerance_raises(tmp_path: Path) -> None:
+    p = tmp_path / "docgen.yaml"
+    p.write_text("validation:\n  av_sync:\n    tolerance_sec: true\n", encoding="utf-8")
+    with pytest.raises(
+        ConfigError, match="validation.av_sync.tolerance_sec must be a YAML number"
+    ):
+        Config.from_yaml(p)
+
+
+def test_from_yaml_list_ocr_min_confidence_raises(tmp_path: Path) -> None:
+    p = tmp_path / "docgen.yaml"
+    p.write_text(
+        "validation:\n  ocr:\n    min_confidence:\n      - 40\n",
+        encoding="utf-8",
+    )
+    with pytest.raises(
+        ConfigError, match="validation.ocr.min_confidence must be a YAML number"
+    ):
+        Config.from_yaml(p)
+
+
+def test_from_yaml_bool_story_end_max_early_sec_raises(tmp_path: Path) -> None:
+    p = tmp_path / "docgen.yaml"
+    p.write_text("validation:\n  story_end:\n    max_early_sec: true\n", encoding="utf-8")
+    with pytest.raises(
+        ConfigError, match="validation.story_end.max_early_sec must be a YAML number"
+    ):
+        Config.from_yaml(p)
+
+
+def test_from_yaml_string_timing_sync_tail_gap_raises(tmp_path: Path) -> None:
+    p = tmp_path / "docgen.yaml"
+    p.write_text(
+        'validation:\n  timing_sync:\n    max_tail_gap_sec: "3.0"\n',
+        encoding="utf-8",
+    )
+    with pytest.raises(
+        ConfigError, match="validation.timing_sync.max_tail_gap_sec must be a YAML number"
+    ):
+        Config.from_yaml(p)
+
+
+def test_from_yaml_validation_nested_numerics_allowed(tmp_path: Path) -> None:
+    p = tmp_path / "docgen.yaml"
+    p.write_text(
+        "validation:\n"
+        "  ocr:\n    sample_interval_sec: 3\n    min_confidence: 50\n"
+        "  layout:\n    min_spacing_px: 12\n    edge_margin_px: 18\n"
+        "  av_sync:\n    tolerance_sec: 2.5\n    min_anchors_per_segment: 3\n"
+        "  timing_sync:\n    max_tail_gap_sec: 4.0\n    max_end_overrun_sec: 1.5\n"
+        "  story_end:\n    max_early_sec: 30\n    max_early_ratio: 0.4\n",
+        encoding="utf-8",
+    )
+    c = Config.from_yaml(p)
+    assert c.ocr_config["sample_interval_sec"] == 3
+    assert c.ocr_config["min_confidence"] == 50
+    assert c.layout_config["min_spacing_px"] == 12
+    assert c.av_sync_config["tolerance_sec"] == 2.5
+    assert c.timing_sync_config["max_tail_gap_sec"] == 4.0
+    assert c.story_end_config["max_early_sec"] == 30