The response half of the ground model (binary @4aa6cf-4aab0b), after the
5.3.98 snap:
WALLS BLOCK BY FULL FRAME REJECTION, never slide or climb: any blocking
contact restores the start-of-frame position and zeroes the velocity.
THE CRASH HURTS: the accumulated collision Damage is dispatched at our own
mech, zone -1 -- through the same cylinder lottery as enemy fire -- which
is what makes wall-grinding self-limiting in the original.
A HARD HIT STAGGERS: past the binary's |v|^2 > 40 threshold (@0x4ab184),
both gait channels bind the bump clip (slot 0x20, the reason the clip
array outgrows the name table) and recover to Standing at its end. Both
channels together, per the donor's foot-slip finding: staggering only the
leg channel desyncs the two cycles permanently.
LIVE, one arena run, no fault -- and the run turned into an unplanned
full-system exercise: 8 wall contacts (three at knockdown force, iv2
745-932; one low-speed bump correctly blocked without stagger), and the
self-damage cascaded through vital zones to MECH KILL, pilot notification,
respawn HEAL-and-place, four times over. The death and respawn machinery
all held.
OPEN, stated rather than tuned away: THE PRICE IS TOO HOT. Wall crashes
land 14k-24k damage points against a game where a PPC is ~12 and an upper
torso carries ~124 -- a hard crash one-shots vital zones through the
lottery. BT411's economy audit hit the identical symptom ("tapped a wall
and died instantly") and traced part of it to compounded StaticBounce
reflections across multi-solid frames. The fix lives in the
Mech::ProcessCollision override (with the crushable-icon sentinel and the
gyro crunch), which is the NEXT increment -- the pricing stays as the
engine computes it until that override is reconstructed from the binary,
because guessing a divisor would be tuning, not archaeology. The [crash]
log line prints amount + iv2 under BT_MECH_LOG as the audit trail.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
CydandClaude Fable 5
4ace2571d9