CARD T1  (written 2026-09-26 01:35:03 -0700, before the trace)  unit source/math/periodic_functions
fn @periodic_function_build_variable_period_x_table@4 (parked instruction-scheduling; reopen criterion = "a decoded
VC7 DAG scheduler rule for this ready-node choice"). NEW evidence class available since the park: the C2 list
scheduler record layout decoded live in fable5 decal-sched (+0x2c primary key descending, +0x36 original list
index ascending, +0x3a unit class, one load per cycle, readiness = insertion cycle).
Observation-only trace of PRODUCTION source (canonical body; residual = J `inc edx; fld [ebp-4]` at +0x6f/+0x70,
ours `fld [ebp-4]; inc edx`).
Pre-registered predictions (production):
 (a) readiness: `fld [ebp-4]` (d945fc) and `inc edx` (42) are both in the ready list in the SAME cycle (the one
     after `fmul [.25]`) - i.e. the order is a sort-key decision, not readiness;
 (b) key: fld's +0x2c > inc's +0x2c (fld heads the long cos/fmul/fadd chain; inc feeds only the [ebp-8] store),
     so fld sorts first; +0x36 is irrelevant here;
 (c) consequence for January: to issue inc first, January's inc must have the higher +0x2c at that cycle, OR the
     fld is not ready yet (readiness), OR the two share a unit class with an already-issued op (fld refused).
     I expect (c-readiness) is the only plausible one: a height swap would need a much longer inc chain.
Decisive: the trace table rows at offsets 0x5d..0x80 (insert cycle, key2c, w36, unit, select cycle).
