Card CI-C1 - written 2026-09-26 18:07:25 -0700 (clock-read) BEFORE compiling. Card-driven compile 1/30 (camera_impulse).
Change: the two unit-length tests written with the first-party real_math.h macro realcmp:
  realcmp(magnitude_squared3d(&flattened_direction), 1.0f) && realcmp(magnitude_squared3d(&facing), 1.0f)
(HEAD hand-writes its expansion: fabs(magnitude_squared3d(&v) - 1.0f) < _real_epsilon).
Evidence: realcmp is Bungie's macro in January: assert strings '!realcmp(determinant, 0.0f)' (matrix_math),
'realcmp(actor->input.facing_vector.k, 0.0f)' and 'realcmp(plane3d_distance_to_point(&plane,
&rotate_to_position), 0.f)' (a call-result argument, as here). /Od 0x57dea0 +0x12b..+0x1bb (magnitude_squared3d
call, subss 1.0, fabs(float) wrapper 0x402e64, epsilon compare) is identical for both spellings (neutral). January
bytes at the site are code-identical either way. NOT site-attested: Bungie also hand-writes 'fabs(cos_theta*cos_theta
+ sin_theta*sin_theta - 1.0f)<_real_epsilon' elsewhere, so this is an owner question (G1: two identical expressions).
Mechanism (CI-T1/T2): January = all six decisive VN temps (ordinals 58, 60, 146, 150, 229, 248) shifted by s == 2
(mod 4). realcmp's '(a)' parenthesises the non-leaf call result: one C1 paren record per use (fifty-objects PIN lab:
'(f(...))' changes keys), two uses before ordinal 58 -> s = +2.
Prediction: EXACT p 0.35 (needs the a.j/b.i tie at hit-1294's list to keep a.j first); dots i,k,j both fixed p 0.6;
cross product fixed p 0.45; inert p 0.25 (if paren records are not VN entries); no other function changes p 0.9.

OUTCOME (appended 2026-09-26 18:07:49 -0700): gate --all 27/2 (HEAD 26/3): _player_effect_update_camera_impulse
EXACT 752. alndiff: 267 = 267 insns, only relocation-name rows (__real@ constant spellings) remain, i.e. all three
hunks (jitter cross product +0x1c4/+0x1e0, left.direction +0x29c, facing.direction +0x2b5) closed. keyed diff vs
build/base: 67 -> 67 sections, 1 changed (_player_effect_update_camera_impulse 752 -> 752, EXACT vs January),
0 added, 0 removed, 0 losses; add_continuous_effect and camera_effect_matrix rows unchanged.
Prediction: EXACT held (p 0.35), including the a.j/b.i tie.
