feat(R1-P0/P1): CSequence research base + verbatim AnimSequenceNode
P0 — research + pins: full CSequence-family verbatim extraction (1756
lines, per-function raw pseudo-C + cleaned flow, decomp line anchors),
ACE cross-reference (9 ranked divergences; headline: retail frame_number
is x87 long double — ACE's float is the worst case, our double the best
available; ACE's frame-boundary epsilon is an ACE fabrication, NOT
retail), current-sequencer map, and the R1 gap map (20 gaps, 13 keeps,
P0-P6 port order). Pinned the one decomp ambiguity (leftover-time carry
after advance_to_next_animation — ACE reading adopted; cdb confirmation
protocol recorded, non-blocking).
P1 — AnimSequenceNode verbatim (gap G1/G2/G16/G18):
- direction-aware BARE-INT boundary pair (get_starting_frame 0x00525c80 /
get_ending_frame 0x00525cb0): reverse starts at high+1, ends at low —
NO epsilon;
- multiply_framerate (0x00525be0) swaps low/high on negative factor;
- set_animation_id (0x00525d60) retail clamp order (high<0 -> num-1;
low>=num -> num-1; high>=num -> num-1; low>high -> high=low);
- ctors with retail defaults (30f/-1/-1; AnimData copy + clamp);
- get_pos_frame null out-of-range (retail; ACE returns identity),
floor double overload; get_part_frame same discipline;
- NO per-node IsLooping/Velocity/Omega — loop membership is list
structure, physics accumulators live on the sequence (G16).
22 conformance tests (clamp table, boundary mirror table, swap
round-trip, bounds/floor semantics).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
1371c2a14c