PMU throughput sweep — 2026-05-24T21-46-27

Arcs per anchor: 5000 · density: 3 · seed: 42 · threads: 10 (Rayon work-stealing)
Run: node scripts/pmu/pmu-throughput-sweep.mjs --widths 12,24,48,96,144 --depths 2,5 --arcs 5000
WDprec walks/sec mean nsσ nsCVSE(10K) p50 nsp99 ns cells litL1D fit
122f64 20.38M 86.3 422.4 4.89 4.2 83 208 42 / 144 ✓ fits
122f32 21.24M 102.5 128.8 1.26 1.3 83 292 42 / 144 ✓ fits
125f64 11.31M 547.5 3830.8 7.00 38.3 458 1042 42 / 144 ✓ fits
125f32 11.84M 569.4 12109.2 21.27 121.1 458 1084 42 / 144 ✓ fits
242f64 22.77M 123.5 667.0 5.40 6.7 125 250 92 / 576 ✓ fits
242f32 28.08M 130.9 2330.0 17.80 23.3 125 250 92 / 576 ✓ fits
245f64 6.85M 1152.1 7084.5 6.15 70.8 1000 1541 92 / 576 ✓ fits
245f32 7.31M 1154.0 4185.4 3.63 41.9 1083 1917 92 / 576 ✓ fits
482f64 26.34M 218.1 4164.5 19.09 41.6 208 292 187 / 2304 ✓ fits
482f32 27.42M 211.8 2236.6 10.56 22.4 208 333 187 / 2304 ✓ fits
485f64 3.83M 2459.7 15005.7 6.10 150.1 2209 2792 187 / 2304 ✓ fits
485f32 3.46M 2762.9 10622.9 3.84 106.2 2541 4750 187 / 2304 ✓ fits
962f64 11.55M 600.0 11075.7 18.46 110.8 500 709 382 / 9216 ✓ fits
962f32 14.09M 492.1 9092.7 18.48 90.9 416 542 382 / 9216 ✓ fits
965f64 1.19M 8114.8 40543.5 5.00 405.4 6750 15125 382 / 9216 ✓ fits
965f32 1.19M 7889.0 67205.2 8.52 672.1 6625 12208 382 / 9216 ✓ fits
1442f64 10.64M 746.6 5329.8 7.14 53.3 667 834 571 / 20736 ✗ exceeds
1442f32 10.87M 744.8 7209.9 9.68 72.1 666 1125 571 / 20736 ✓ fits
1445f64 763.6K 12895.1 36031.6 2.79 360.3 11417 19542 571 / 20736 ✗ exceeds
1445f32 724.8K 13653.2 56925.0 4.17 569.3 11875 22041 571 / 20736 ✓ fits

How to read σ: per-walk wall-clock σ across N samples. Long right tail (max = OS jitter + GC + context switches) dominates the raw σ; the load-bearing column for σ-separation claims is SE(10K) (standard error of the mean at N=10,000 walks: σ/√N). Two action fingerprints separated by Δmean > 3.4 × SE(10K) clear the Skybridge 3.4σ threshold from one 10K-walk sample.

Receipts written to .thetacog/pmu/throughput/sweep-w<W>-d<D>-<prec>.json.