feat(render) Campaign FW1: NINE of ten fixtures conformant - the degrade arm

The moving-fixture divergence was the building degrade ladder:
GfxObjDegradeInfo::get_degrade @0x0051e4b0 (Ghidra-verified - BN's
FPU-flag pseudo-C misread BOTH arm selection and one formula) slides
each level's threshold from ideal toward MAX as the multiplier
approaches 1, and the live client runs the positive arm at ~0.99, so
level 0's portal-bearing BSP survives to ~max_dist (48 for the Holtburg
cottages), not ideal (24). The recon note's "deg_mul = -0.99" was a
sign misread; the negative arm's threshold slides toward MIN and
contradicts the fixtures from both directions. With the two-arm port,
holtburg-walkout, holtburg-transitions, and holtburg-walkabout pass
every pairable frame - the walkout-F2 microscope's prediction
(103,100 | 100x3 | 124 through the cottage exit views) landed exactly.

Also this round, falsified and reverted: a BN-driven swap of the
portal walker's negative/in-plane arms (Ghidra shows side 1 = negative
EMITS, side 2 = in-plane does not - the original port was correct; the
swap broke four fixtures). The walker docs and unit tests now pin the
Ghidra-verified truth table.

Parked with findings: foundry-entry F67 (right flood set, one
plane-side classification at the +/-eps boundary orders 11d before
116/118) and doorway-still (retail shows zero floods at a pose one
meter from walkout-F2's flooding pose; multi-portal clip boundary).

Suites: Walk 124/3 skips; hermetic 6,687/0.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

src/AcDream.App/Rendering/Walk/WalkBuilding.cs

@@ -77,23 +77,38 @@ public sealed class WalkBuilding
     /// client; registry-configurable).</summary>
     public const float DefaultDegradeDistance = 100f;
 
+    /// <summary>The live capture client's <c>Render::deg_mul</c> arm:
+    /// magnitude 0.99 pinned by the walkout-F2 fixture (building a9b4001e,
+    /// effective 30.3, level 0 ideal/max 24/48 — retail floods, so its
+    /// threshold ≈ max ⇒ the POSITIVE arm; the recon note's "−0.99" sign was
+    /// a misread — the negative arm's threshold ≈ min contradicts the
+    /// fixture from both directions). Re-dump at the next retail session.</summary>
+    public const float DefaultDegradeMultiplier = 0.99f;
+
     /// <summary>
-    /// <c>GfxObjDegradeInfo::get_degrade</c> @0x0051e4b0 (the live client's
-    /// arm: <c>auto_update_deg_mul</c> with a non-positive multiplier →
-    /// thresholds are the levels' IDEAL distances): effective =
-    /// max(0, distance − degradeDistance); the FIRST level with
-    /// effective &lt; IdealDist wins; no match → the LAST level. The
-    /// positive-bias arm (threshold = ideal − (ideal−max)·mul) is a port
-    /// follow-up if a fixture ever pins a positive bias.
+    /// <c>GfxObjDegradeInfo::get_degrade</c> @0x0051e4b0 (BN flag-mush on
+    /// the FPU compares — cross-checked against the Ghidra decomp, which is
+    /// clean): effective = max(0, |distance| − degradeDistance), then the
+    /// FIRST level whose threshold exceeds effective wins; no match → the
+    /// LAST level. The threshold depends on the multiplier's sign:
+    /// mul ≥ 0 → <c>ideal − (ideal − max)·mul</c> (ideal→max as mul→1);
+    /// mul &lt; 0 → <c>ideal + (ideal − min)·mul</c> (ideal→min as mul→−1).
+    /// mul is <c>Render::deg_mul</c> when <c>auto_update_deg_mul</c> is on,
+    /// else <c>s_rUserSuppliedDegradeBias</c>.
     /// </summary>
     public WalkBspNode? SelectDrawingBsp(
-        float viewerDistance, float degradeDistance = DefaultDegradeDistance)
+        float viewerDistance,
+        float degradeDistance = DefaultDegradeDistance,
+        float degradeMultiplier = DefaultDegradeMultiplier)
     {
         if (DegradeLevels.Length == 0) return DrawingBsp;
         float effective = MathF.Max(0f, MathF.Abs(viewerDistance) - degradeDistance);
         foreach (WalkBuildingDegradeLevel level in DegradeLevels)
         {
-            if (effective < level.IdealDist)
+            float threshold = degradeMultiplier >= 0f
+                ? level.IdealDist - (level.IdealDist - level.MaxDist) * degradeMultiplier
+                : level.IdealDist + (level.IdealDist - level.MinDist) * degradeMultiplier;
+            if (effective < threshold)
                 return level.DrawingBsp;
         }
         return DegradeLevels[^1].DrawingBsp;
@@ -130,10 +145,16 @@ public static class WalkBuildingPortals
 
     /// <summary>
     /// <c>BSPTREE::build_draw_portals_only</c> @0x00539860 + the node/portal
-    /// walkers: dispatch the root, then walk plane-side ordered — the child
-    /// OPPOSITE the viewer first, so portals emit far-to-near. PORT nodes
-    /// emit every in_portal on the POSITIVE and NEGATIVE arms; the IN_PLANE
-    /// arm (|d| ≤ ε) visits the positive child and emits NOTHING.
+    /// walkers (<c>BSPNODE</c> @0x0053c100, <c>BSPPORTAL</c> @0x0053d870;
+    /// side arms verified against the GHIDRA decomp 2026-08-30 — BN's
+    /// FPU-flag pseudo-C reads the negative/in-plane split ambiguously, and
+    /// a "corrected" swap of these arms broke four fixtures before being
+    /// falsified; Ghidra: <c>d ≤ ε → side=1 unless −ε ≤ d → side=2</c>):
+    /// dispatch the root, then walk plane-side ordered — the child OPPOSITE
+    /// the viewer first, so portals emit far-to-near. PORT nodes emit every
+    /// in_portal on the POSITIVE (d &gt; ε) and NEGATIVE (d &lt; −ε) arms;
+    /// the IN_PLANE arm (|d| ≤ ε) visits the positive child and emits
+    /// NOTHING. Plain nodes group IN_PLANE with NEGATIVE.
     /// </summary>
     public static void BuildDrawPortalsOnly(
         WalkBspNode? root, int pass, Vector3 viewpointInBuilding,