Low nonces-per-thread geometry sweep
NPT values 1, 2, 4 or 8 at TPB 256/512 can outperform the retained T512-N32 baseline by at least 1%.
Diagnostic work, exact algebra and local capabilities are not treated as end-to-end mining advantage.
What was tested?
NPT values 1, 2, 4 or 8 at TPB 256/512 can outperform the retained T512-N32 baseline by at least 1%.
Why the test is meaningful
Lower NPT can reduce per-thread serial work but increases launch/scheduling overhead; equal total hashes are required.
candidates=TPB{256,512}×NPT{1,2,4,8}R=throughput_candidate/throughput_T512-N32gate R≥1.01How it was tested
Compile ten exact binaries and test four unseen headers, two repetitions and 80 equal-work runs.
What happened
T512-N4 was best at 1.003573×, positive on 4/4 headers but below the 1.01 gate on every dataset. All binaries verified B32 with zero discrepancies/spills.
Exactness and statistical controls
The preregistered experiment closes negative. The consistent 0.36% direction is retained only as a discovery signal for an independent replication.
What the result means
No registered low-NPT geometry clears the primary gate; N4 receives a separate replication, not retrospective promotion.
Limitations
- The 1% gate intentionally rejects small discovery effects.
- Only integer registered geometries were tested.
- No energy inference is made.
Evidence trail
Keep equal hash counts, publish all geometries and separate this failed gate from the successor replication.
Canonical variants
CANONICAL-EXP-144PREREGISTERED-CAMPAIGNSEALED-AUDITSource: internally audited canonical reports. Local filesystem structure, private headers and operational identifiers are excluded from publication.