P32 warp-broadcast epoch cancellation
One global epoch load per warp, broadcast with shuffle, can meet cancellation latency while retaining at least 99.95% of P32 throughput.
Diagnostic work, exact algebra and local capabilities are not treated as end-to-end mining advantage.
What was tested?
One global epoch load per warp, broadcast with shuffle, can meet cancellation latency while retaining at least 99.95% of P32 throughput.
Why the test is meaningful
Warp broadcast divides global control loads across 32 lanes and avoids a block barrier, creating a lower-overhead cancellation primitive.
work=33,554,432 hashes per launchepoch_lane=global_load(lane 0); epoch=shfl(epoch_lane)promote iff response≤15 ms and retention≥0.9995How it was tested
Implement one epoch read per warp for each P32 launch; run full 2^32 exactness, 24 paired benchmarks and four independently timestamped cancellations.
What happened
All responses passed at 9.94–14.84 ms and exactness had zero discrepancies. Retention was 0.999309 with CI95 0.999092–0.999526, favorable on 0/4 headers.
Exactness and statistical controls
B32, compiled factors, 44 registers and zero spills passed. Functional cancellation passed, while the preregistered throughput gate did not.
What the result means
The mechanism is exact and responsive, but its measured cost prevents promotion as the default. Independent replication is needed to characterize the penalty.
Limitations
- A functional primitive is not an end-to-end strategy.
- The point estimate misses the strict retention gate.
- No ASIC behavior is inferred.
Evidence trail
Use the sealed P32 control and warp-epoch binaries, full-domain audit, balanced ordering and the unchanged 0.9995 gate.
Canonical variants
CANONICAL-EXP-165PREREGISTERED-CAMPAIGNSEALED-AUDITSource: internally audited canonical reports. Local filesystem structure, private headers and operational identifiers are excluded from publication.