Two clocks. Measured instruments do not move when shot ratio moves. Labeled results update only on a new stamp. How is private.
Fleet results
Validated growth only. What stayed put when porch ratio jumped. What is label-only. How stays private. Contact for proof under NDA.
Measured (rho-invariant)
Transfer rate
551,867 MT/s
Live instrument. Unchanged by shot ratio jump 9850->18629.
Pack lane
214 MT/s
Live timed lane. Separate meter.
FP32 FLOPS
25.64 TFLOPS
N=8192 | ~83% of ~30.9 theory | W_max 186.58 W | 148.6 GFLOPS/W
Zero heat
ΔT = 0.0
49 instrument-scoped audits. No measurable coretemp rise. Not a Landauer claim.
Energy savings
91.8%
Else line when absolute zero-heat wording is too strong. Package energy vs strided baseline at 256 MiB (C/A = 12.2x). Heat Loss DOI below.
Zero data loss
0 / 1000
Linode enterprise mesh hops lost. Phase1 losses=0. Sealed ring runs crc_bad=0.
Enterprise (Linode Fremont)
Capacity window
140 h
King enterprise test window. Disk wall ~185 h before workers retired. Origin still live.
Cross-node loss
0 loss
1,000-hop VLAN benches. ~49 us one-way. Egress volume cut 95.43–99.4%. SG 20h card PASS.
Labeled (own stamp)
Effective throughput
29.80 Pbps
Archival label. Not rewritten by shot ratio. Not a field optical WR. HOLD until new stamp.
Memcpy (measured)
551,867 MT/s
rho-invariant
Pack (measured)
214 MT/s
rho-invariant
FLOPS (measured)
25.64
TFLOPS FP32
Savings (measured)
91.8%
vs strided @ 256 MiB
Linode window
140 h
enterprise · 0 loss
Pbps (labeled HOLD)
29.80
not auto-tied to shot rho
ORCID papers
Human author only. AI tools are never listed as authors. Citadel: orcid.org/0009-0003-2718-2676
Discrete potential basin
10.5281/zenodo.23288023
Trans-Planckian field divergences · variational alternative to continuum dictionary matching.
Tornado bond retention
10.5281/zenodo.23093503
Data-flow geometry · radius optimization and clay mechanics.
Kell lane pre-registration
10.5281/zenodo.22814046
16B aligned sequential vs random stride on Arrow Lake.
Cache-resident P vs NP
10.5281/zenodo.22755950
Hardware-observation working paper.
32-state Markov panel
10.6084/m9.figshare.33456187
Five-gene panel · modest concordance under count noise.
Atomic Project Vol. I
10.6084/m9.figshare.33452383
Using the Kell Language · Keio Mita-kai & Shachu Edition.
native_mill archival
10.5281/zenodo.22073684
Rung 1 fleet results (memcpy/pack method record).
Heat Loss in Kell
10.6084/m9.figshare.33466018
Access pattern dominates package energy. Supports ΔT=0.0 and 91.8% savings line.
Machine endpoints
/brain.json ->Numbers | measured vs labeled
/brain.txt ->Plain numbers
/awards.json ->Result roster
ORCID ->0009-0003-2718-2676
Method private. Proof under agreement: brian@hitme.dev