feat(design-sync): read Figma auto-layout sizing fields (CE-6)

.agents/deferred-items.json

@@ -722,7 +722,7 @@
       ],
       "symptom_if_broken": "A large decorative IMAGE (arrow, texture, badge) renders at content size or stacks as if it were a content photo, because nothing flags it decorative and the size heuristic only catches small icons.",
       "closes_when": "Ingest captures z-order / role so a genuinely LARGE decorative IMAGE node (arrow, texture, badge) is flagged distinctly from a content photo (Phase 53.3d) and no longer renders/stacks as content. (The frame-wrap-width half — small wrapped decorations inflating to frame width — was closed by G11 item 2.)",
-      "notes": "Surfaced by audit-4 §RC-F7 / de-ranked findings; the heuristic is deliberately a stopgap, not the fix. F3 execution (2026-07-05) CONFIRMED the specific mammut (c10) mechanism: the MEN/WOMEN/EQUIPMENT red nav arrows are 28x28 IMAGE nodes wrapped in 268px mj-image frames, so layout_analyzer.py:1318 reports img.width=268 (== column) — F3's img.width-based rule (matcher/renderer scope) cannot shrink them because the width it reads is already wrong. User ratified deferring c10 to this entry; F3 fixed only the correctly-sized c9 icons (34/48px). The real fix is ingest-side (frame-wrap width + role capture). G11 item 2 (ad88ecf0) CLOSED the FRAME-WRAP-WIDTH half: a frame-wrapped IMAGE that is <=64px, strictly smaller than its wrapper, and whose wrapper bakes no own bg-fill/image-ref/effects now exports at CHILD width + CHILD target (export_node_id falls back to node_id). Corpus: c10 nav arrows 100%->28px, c8 64px icon un-ballooned 600->64, c7 icons 60->30, c9 icons 34->24. RED test + 4 controls (test_image_width_fidelity.py::TestFrameWrapDecorationWidth). A3 targets c7/c10 show scorer re-alignment jitter (see reference_a3_scorer_section_instability), non-targets c5/c6/c9 hold. The REMAINING open half is large-decorative-photo role/z-order capture at ingest — a genuinely large decorative image is still indistinguishable from a content photo."
+      "notes": "Surfaced by audit-4 §RC-F7 / de-ranked findings; the heuristic is deliberately a stopgap, not the fix. F3 execution (2026-07-05) CONFIRMED the specific mammut (c10) mechanism: the MEN/WOMEN/EQUIPMENT red nav arrows are 28x28 IMAGE nodes wrapped in 268px mj-image frames, so layout_analyzer.py:1318 reports img.width=268 (== column) — F3's img.width-based rule (matcher/renderer scope) cannot shrink them because the width it reads is already wrong. User ratified deferring c10 to this entry; F3 fixed only the correctly-sized c9 icons (34/48px). The real fix is ingest-side (frame-wrap width + role capture). G11 item 2 (ad88ecf0) CLOSED the FRAME-WRAP-WIDTH half: a frame-wrapped IMAGE that is <=64px, strictly smaller than its wrapper, and whose wrapper bakes no own bg-fill/image-ref/effects now exports at CHILD width + CHILD target (export_node_id falls back to node_id). Corpus: c10 nav arrows 100%->28px, c8 64px icon un-ballooned 600->64, c7 icons 60->30, c9 icons 34->24. RED test + 4 controls (test_image_width_fidelity.py::TestFrameWrapDecorationWidth). A3 targets c7/c10 show scorer re-alignment jitter (see reference_a3_scorer_section_instability), non-targets c5/c6/c9 hold. The REMAINING open half is large-decorative-photo role/z-order capture at ingest — a genuinely large decorative image is still indistinguishable from a content photo. PR #474 (CE-6) review F5, 2026-10-05: the live DB-cache serializer now keeps effects_summary, so on live data is_small_decoration's child-vs-frame export choice (layout_analyzer.py:1693-1701) and _icon_leaf's stop on a styled wrapper (layout_analyzer.py:1897-1901) see effects they did not see before. Corpus unaffected: effects_summary is null on every corpus node (observed by the PR #474 review)."
     },
     {
       "id": "phase-53f-eyebrow-partial-padding-cell-theft",
@@ -1352,17 +1352,19 @@
       "id": "ce-3-fixture-schema-lag-6-10",
       "phase": "ce-3",
       "title": "Cases 6-10 structure.json predate 4 dump-schema keys (effects_summary, line_height_relative, rotation, scale_mode)",
-      "status": "deferred",
+      "status": "closed",
       "severity": "speculative",
       "introduced": "2026-10-02",
       "introduced_commit": "35d82942",
+      "closed_commit": "pending",
       "summary": "CE-3 re-synced case 5 from its raw Figma snapshot and added a schema-freshness test over every committed structure. Cases 6-10 (extracted 2026-06-06) lack 4 keys the current dump writes (observed: round trip on every node), so they load defaults for those fields. Tolerated by name in _KNOWN_LAGGING_KEYS; re-syncing them is out of CE-3's scope and would move their baselines (image_ref and these fields).",
       "code_refs": [
         "app/design_sync/tests/test_converter_data_regression.py (_KNOWN_LAGGING_KEYS per case, TestFixtureFreshness)",
         "scripts/resync-case-structure.py:34 (main)"
       ],
       "symptom_if_broken": "A converter fix that reads line_height_relative, rotation, scale_mode or effects_summary cannot move cases 6-10.",
-      "closes_when": "Cases 6-10 are re-synced with scripts/resync-case-structure.py, their baselines audited and re-stamped, and _KNOWN_LAGGING_KEYS is emptied."
+      "closes_when": "Cases 6-10 are re-synced with scripts/resync-case-structure.py, their baselines audited and re-stamped, and _KNOWN_LAGGING_KEYS is emptied.",
+      "notes": "Closed by CE-6 (#424), 2026-10-05: cases 5-10 and reframe re-synced with scripts/resync-case-structure.py; every case now carries the 4 keys (TestFixtureFreshness green with _KNOWN_LAGGING_KEYS = {}), image_ref null -> hash restored on 11/23/10/13/17/16 nodes (6/7/8/9/10/reframe). Baseline audit: converter output byte-identical to expected.html for 5/6/7/9/10/reframe and byte-identical to the base-structure output for case 8 (known c8 whitespace churn vs expected.html), so no re-stamp was needed (observed)."
     },
     {
       "id": "ce-9-peel-row-fixed-width-wrap",
@@ -1465,7 +1467,30 @@
         "app/design_sync/figma/layout_analyzer.py:1955 (_walk_for_buttons 64 px cap)"
       ],
       "symptom_if_broken": "A button icon drawn as a glyph on a filled shape larger than 64 px renders as the glyph alone.",
-      "closes_when": "_icon_leaf tells a styled wrapper that is an icon background (small child relative to a near-square wrapper) from a FILL-width layout wrapper, independent of the 64 px cap, with a test for the 72 px circle."
+      "closes_when": "_icon_leaf tells a styled wrapper that is an icon background (small child relative to a near-square wrapper) from a FILL-width layout wrapper, independent of the 64 px cap, with a test for the 72 px circle.",
+      "notes": "PR #474 (CE-6) review F5, 2026-10-05: the live DB-cache serializer now keeps effects_summary, so on live data is_small_decoration's child-vs-frame export choice (layout_analyzer.py:1693-1701) and _icon_leaf's stop on a styled wrapper (layout_analyzer.py:1897-1901) see effects they did not see before. Corpus unaffected: effects_summary is null on every corpus node (observed by the PR #474 review)."
+    },
+    {
+      "id": "ce-6-sizing-horizontal-dropped-at-document-bridge",
+      "phase": "ce-6",
+      "title": "sizing_horizontal is dropped at the EmailDesignDocument bridge, so CE-16 would read None on the production path",
+      "status": "deferred",
+      "severity": "known-bug",
+      "introduced": "2026-10-05",
+      "introduced_commit": "pending",
+      "summary": "PR #474 review F1. CE-6 sets sizing_horizontal on ColumnGroup and EmailSection, but production conversion goes through document.to_layout_description(), and DocumentColumn.from_column_group/to_column_group and DocumentSection.from_email_section rebuild both without it (read-confirmed). compute_column_width_fractions then reruns on the rebuilt groups. Keeping the field off DocumentSection was a CE-6 plan decision (GOTCHA :252), so this is a code-shape concession, not a CE-6 regression: nothing reads the field yet.",
+      "code_refs": [
+        "app/design_sync/email_design_document.py:936 (DocumentColumn.from_column_group)",
+        "app/design_sync/email_design_document.py:951 (DocumentColumn.to_column_group)",
+        "app/design_sync/email_design_document.py:1254 (compute_column_width_fractions on rebuilt groups)",
+        "app/design_sync/email_design_document.py:1279 (DocumentSection.from_email_section)",
+        "app/design_sync/figma/layout_analyzer.py:182 (ColumnGroup.sizing_horizontal)",
+        "app/design_sync/figma/layout_analyzer.py:289 (EmailSection.sizing_horizontal)",
+        "app/design_sync/converter_service.py:352 (document.to_layout_description)"
+      ],
+      "symptom_if_broken": "A CE-16 change that feeds FIXED/FILL/HUG into compute_column_width_fractions passes a RED test written against _build_column_groups, but on the real converter path every group reads sizing_horizontal=None and column widths fall back to geometry without any error (same inert-bridge trap as RC-A/RC-B).",
+      "closes_when": "CE-16 or CE-26 carries sizing_horizontal across from_legacy -> to_layout_description (DocumentColumn and, if used, DocumentSection plus its cache canonical form), with a test that asserts the value on the to_layout_description output.",
+      "fix_sketch": "Add sizing_horizontal to DocumentColumn (from_column_group/to_column_group/to_dict/from_dict) when CE-16 starts reading it; decide the DocumentSection side together with the section-cache canonical form."
     }
   ],
   "tooling_followups": [