INDEPENDENT KELL K3/K4 VALIDATION
date=2026-08-29
reviewer=Varadarajan Venkatesan
uid=FLEET-D345-RAJ-20260829T205300Z
start=D3.D4.D5
he_has_a_part=Q1_PASS + Q4_FAIL
q_ledger=tpg_wake/RAJ_Q_UID_ALL_NOW.tsv
environment=AMD Ryzen 5 5600G · 6 vCPU · 23 GiB RAM · Linux Mint 21.2
kernel=5.15.0-187-generic
host=Oracle VirtualBox VM

SCOPE
  reviewed=Rung 1 method · LIBRARY_PROXY script · homepage · certificate
           history · Kell for AI Agents page · both public best_numbers.json locations
  independent_files=validation.md · raw_runs.txt · machine_details.txt

VERDICTS
  LIBRARY_PROXY=PASS
    detail=the reviewer ran bash library_proxy_bench.sh 256 five times;
           all five completed successfully
    measures=256 MiB dd write to a temporary file and read to /dev/null
    boundary=not native_mill · not cache-resident memcpy

  native_mill_memcpy_259093_MT_s=UNCERTAIN
    reason=no public runnable native_mill reproducer in the supplied pack

  native_mill_pack_360800_MT_s=UNCERTAIN
    reason=no public runnable native_mill reproducer in the supplied pack

  public_best_receipt_consistency=FAIL
    reason=the two public best_numbers.json sources disagreed on memcpy,
           runner, timestamps, and latest-run data; one downloaded root
           receipt contained blank GB/s fields

  K3=UNCERTAIN
    reason=no sufficiently defined K3/KELL3 test in the reviewed materials

  KELL4_pure_3_of_3=UNCERTAIN
    reason=certificate history recorded the result, but no public runnable
           KELL4 test or three raw runs were supplied

  90_percent_pure_Kell4=UNCERTAIN
    reason=no public reproducer for the certificate-history values

  gotokell_kt_2_0_23_programs_23_of_23=UNCERTAIN
    reason=no public runtime, .kl suite, or local RAM output was supplied

OVERALL
  green=independent LIBRARY_PROXY reproduction
  unresolved=native_mill · K3 · KELL4 · local RAM claims
  failed=public receipt consistency at review time
  no_inference=missing values and unavailable claims were not inferred

PUBLIC CLAIM
  independent Linux reproduction of the public LIBRARY_PROXY benchmark
  is PASS. Native_mill and K3/KELL4 claims remain claim-scoped and UNCERTAIN
  until a public runnable method and matching receipt are available.

WORD=honest · independent · public_method