Axis position shift during helical boring - Fusion 360 Personal + Mesa 7i96s + C

Hi everyone,
 
I'm experiencing cumulative axis position shifts (mostly Y, some X) during helical boring operations generated by Fusion 360 Personal Use. The shift happens between passes, not during cutting. I've done extensive testing to isolate the cause and I'm stuck — hoping someone with trajectory planner experience can help.
HARDWARE:
- 3-axis mill, 1204 ballscrews
- Mesa 7i96s ethernet card
- Teknic ClearPath CPM-SDSK-2311S-RQN servos (step/dir mode)
- 75VDC PSU
- Raspberry Pi 5 running Debian Bookworm
- LinuxCNC Master (2.9)

THE PROBLEM:
When running a Fusion 360 helical bore operation (3 passes: rough, semi-finish, finish), the machine shifts position during the helical arc cutting itself. The first boring pass starts correctly, but position drifts during the operation. The shift is cumulative across passes — each subsequent pass is worse. I initially thought the shift was only happening between passes during repositioning, but after extensive testing I've confirmed it also occurs during the actual G3 helical arc execution.

WHAT I'VE RULED OUT:
1. NOT mechanical: Dial indicator shows <0.01mm backlash. Coupling and ballscrew are tight.
2. NOT motor faults: ClearPath LEDs stay green, no faults. (Note: HLFB is wired but not yet configured in HAL — this is on my to-do list.)
3. NOT lost steps during linear moves: I wrote a test program doing 20 cycles of G91 G1 Y50 F1000 / G1 Y-50 and Y returns perfectly. Zero shift.
4. NOT electrical noise / VFD: Shift occurs with VFD off, spindle off, no tool, air cutting only.
5. NOT step timing: Tested DIRSETUP/DIRHOLD at 2000, 5000, and 10000ns. 5000 was slightly better but didn't solve it.

WHAT I'VE FOUND:
The problem appears related to how the trajectory planner handles the G-code Fusion 360 Personal generates. Key factors:
A) Fusion 360 Personal Use converts ALL G0 rapids to G1 moves. This means repositioning between bore passes (retract, XY travel, approach) are G1 moves subject to G64 path blending. The machine blends through the repositioning corners instead of making exact stops.
The bore operation uses helical arcs (G3 in XY with simultaneous Z) — this is the exact pattern described in GitHub Issue #426 (Arc Blending Bug).
C) I tested ARC_BLEND tuning parameters:
ARC_BLEND_ENABLE = 1
ARC_BLEND_FALLBACK_ENABLE = 1
ARC_BLEND_OPTIMIZATION_DEPTH = 50
ARC_BLEND_GAP_CYCLES = 4
ARC_BLEND_RAMP_FREQ = 100
D) ARC_BLEND_ENABLE = 0 did NOT fix the second file either.

G-CODE STRUCTURE:
The problematic repositioning sequence between bore passes (02.ngc):
N635 G1 X-77.844 Y12.6 (exit move)
N645 Z5. (retract - G1, should be G0)
N650 Z15. (retract more - G1, should be G0)
N655 Z4. (approach - G1, should be G0)
N660 Z2. (approach to cut depth - G1)
N665 X-80.544 Y13.2 (XY reposition AT cut depth - G1, gets blended!)
N670 G3 X-81.144 Y12.6 ... (arc cutting begins)
My INI Snippet:
[EMCMOT]
EMCMOT = motmod
SERVO_PERIOD = 1000000
ARC_BLEND_ENABLE = 1
ARC_BLEND_FALLBACK_ENABLE = 1
ARC_BLEND_OPTIMIZATION_DEPTH = 50
ARC_BLEND_GAP_CYCLES = 4
ARC_BLEND_RAMP_FREQ = 100
 
[JOINT_0] (X)
MAX_VELOCITY = 50
MAX_ACCELERATION = 150
STEPGEN_MAXVEL = 62.5
STEPGEN_MAXACCEL = 250.0
DIRSETUP = 5000
DIRHOLD = 5000
STEP_SCALE = 200.100
 
[JOINT_1] (Y)
MAX_VELOCITY = 50
MAX_ACCELERATION = 150
STEPGEN_MAXVEL = 62.5
STEPGEN_MAXACCEL = 250.00
DIRSETUP = 5000
DIRHOLD = 5000
STEP_SCALE = 200.460
 
G-code preamble: G64 P0.005 Q0.005

QUESTIONS:
1. Is there a known interaction between Fusion 360 Personal's G1-only rapids and the trajectory planner's arc blending that could cause multi-millimeter position shifts?
2. Is there a recommended set of ARC_BLEND parameters for helical boring operations where G1 repositioning is mixed with G3 arcs?
3. Would modifying the linuxcnc.cps post processor to insert G61 before repositioning moves and G64 before cutting moves be the correct approach?
4. Should I be looking at something else entirely?
I can attach the G-code files if helpful. Any guidance appreciated — I've been chasing this for a while.
 
Thanks!