test(physics): AP-157 measured — CylHeight half retired, sorting-sphere half proven collision-unreachable; AD-55 byte-decoded

Campaign S S1A, both outcomes the measure-first rule exists for.

AP-157's CylHeight half is RETIRED as a non-divergence: retail's own
cylsphere overload (CObjCell::find_cell_list @0x0052b9f0) copies
localtoglobal(low_pt) + radius per cylsphere, capped at 10, and never
reads height — retail collapses a cylsphere to a base-point sphere
exactly as acdream does.

The sorting-sphere half measured REAL against retail's registration set
— 1,812 of 3,343 evaluated Setups (54%) fail containment at 1 mm, worst
shortfall 18.135 m — and then PROVEN collision-unreachable: for this
branch the flood spheres and the collision-test geometry are the same
per-part Sphere list, so every omitted cell is one the entity's test
geometry cannot reach, and retail's wider sorting-sphere registrations
are narrow-phase rejects on retail too. Fix deferred to the next
bake-schema revision rather than performing Slice I3 surgery for zero
behavioural delta. The measurement test stays in the tree as the
permanent record (population cross-checked against the dispatch test's
independently-committed constants: 3,506 = 3,605 - 99).

AD-55 is byte-decoded and RESOLVED against our constant: the binary
loads qword [0x007c6b28] = pi/18 exactly and executes FCOS — retail's
Sledding flatness threshold is cos(10 deg) = 0.984808. Our 0.99999536f
is cos(0.17453 DEGREES): the radian literal misread as degrees, which
makes the object-friction arm unreachable on real terrain (nothing is
flatter than 0.175 deg). Evidence note carries the full instruction
listing and the polarity of the test ah,0x41 / jp idiom; the one-line
fix + conformance test is S5, queued behind the running implementation
slice for build-slot reasons.

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

docs/research/2026-08-07-ad55-sledding-constant-byte-decode.md

@@ -0,0 +1,66 @@
+# AD-55 resolved by byte-decode — the Sledding flatness threshold is cos(10°), and our constant is a unit slip
+
+**Date:** 2026-08-07 (overnight). **Method:** `reference_pe_byte_decode` — raw
+bytes from the PDB-paired v11.4186 binary (`check_exe_pdb.py` → MATCH,
+CodeView GUID `9e847e2f-777c-4bd9-886c-22256bb87f32`), not the decomp text.
+
+## The question AD-55 filed
+
+`PhysicsBody.calc_friction`'s Sledding near-flat branch compares
+`GroundNormal.Z > 0.99999536f` (≈0.175° from flat). The raw decomp of
+`CPhysicsObj::calc_friction` @0x0050ee70 instead shows
+`__fcos(0.17453292519943295)` — cos(10°) ≈ 0.984808 — compared against
+`contact_plane.N.z`. One of the two had to be a decode artifact.
+
+## The bytes @0x0050ef53 (verbatim from the binary)
+
+```
+d9 86 38 01 00 00    fld   dword [esi+0x138]      ; contact_plane.N.z
+dd 05 28 6b 7c 00    fld   qword [0x007c6b28]     ; the constant
+d9 ff                fcos
+de d9                fcompp                        ; cos(const) vs N.z
+df e0                fnstsw ax
+f6 c4 41             test  ah, 0x41
+7a 0a                jp    +0x0a                   ; skip the friction load
+8b 86 bc 00 00 00    mov   eax, [esi+0xbc]        ; this->friction
+89 44 24 04          mov   [esp+4], eax
+d9 44 24 04          fld   dword [esp+4]
+```
+
+`qword [0x007c6b28]` = **0.17453292519943295 = π/18 exactly** (verified by
+direct read at the mapped file offset). The binary genuinely executes `FCOS`
+at runtime — the compiler did not fold it — so the threshold retail compares
+against `N.z` is **cos(π/18 rad) = cos(10°) = 0.984807753...**.
+
+## The verdict
+
+- **The decomp was RIGHT. Our port is wrong.** `0.99999536f` is
+  `cos(0.17453292519943295°)` — the *radian* literal read as *degrees* and
+  run through a degree→radian cosine. A one-character-class unit slip that
+  survived because nothing gates slope feel numerically.
+- **Felt consequence:** the branch means "on ground flatter than the
+  threshold, use the object's own friction; on steeper ground (while slow —
+  the `arg3 < 6.25` speed² gate at 0x0050ef46 guards this), keep the 0.2
+  sliding friction." With our constant, "flat" requires < 0.175° — real
+  terrain triangles essentially never qualify, so the object-friction arm of
+  Sledding is unreachable in practice and slow movers keep sliding friction
+  on gentle slopes retail treats as flat. Ice-feel in exactly the S4/S5
+  slope-feel family.
+- **Fix shape (S5):** replace the constant with retail's semantics. Either
+  the folded `0.98480775f` with a comment carrying this evidence, or the
+  exact `MathF.Cos(MathF.PI / 18f)` computed once — prefer the folded
+  constant + comment, matching how AP-7's 0.25f landed. Polarity must be
+  ported from the `test ah,0x41; jp` idiom above, not assumed: the friction
+  load is SKIPPED when the jump is taken (cos(10°) > N.z, i.e. steeper than
+  10°, or unordered), and taken when N.z ≥ cos(10°). Verify our branch's
+  existing polarity against this before changing only the constant.
+- The speed² gates in the same function — 1.5625 (= 1.25²) at 0x0050ef24 and
+  6.25 (= 2.5²) at 0x0050ef46 — matched our port already at AP-7 and are
+  untouched.
+
+## Bookkeeping owed at the fix
+
+Retire AD-55 (the row's open question is now answered against our constant);
+conformance test pinning `0.98480775f` + the ported polarity; S4/S5's
+slope-feel session covers the felt change. Until the fix lands, the register
+row stands corrected by this note.