Gotokell

Kell scoreboard · kernels

• e15 — done · ~5 day class (5.5d wall with ~1d outage)
• e16 — mapped · carpet + TOC · QED HOLD
• Kraken wire — R≈9850 class · many items · 64 B → 630,400 B rebuild
• SE 5M:1 — distribution target only · not a zip-have claim

Scoreboard = Atom16 / gilb / Kraken path on Forge·CB·Ellie. Phones are carry later.

FLOPS / PB/s · separate jar

GPU TFLOPS (e.g. ~25 FP32 on the card) are real when measured — and they are not Kell kernels. PB/s-style headlines are FLOPS-family theater for this porch. We do not score the mill that way. Dual-clock: CUDA jar ≠ Kraken tm/s ≠ gilb pack lane.

P vs NP Reformulation · Zenodo 22755950

Cache-resident discrete search · hardware observation · not Clay proof · Sep 2026 working paper.

Kell Architecture & Methods

Atom16 · memory transit · heat loss DOIs · retired oncology audit · rung 1 archive.

Arrow Lake GOP · Sep 12 2026

Historical silicon proof preserved · 100k Atom16 UEFI pre-boot · RTX 5070 framebuffer.

George Foreman Telemetry

Cite-only fleet stamp · door :1700 · sky hub witness · gen=1799 xp=1709.

World Sickbay

Conservative public extension · world cancer porch · separate from retired figshare headline.

Lab worksheets

Power · V60 · E1–E10 · full fold math — lab lane only · never lifted to porch headlines.

Same file, two boxes

630,400 bytes in. 64 bytes move. 4,000 pieces rebuild. This box 46.7 ns. California 130.1 ns. August mark 230 ns. The bars climb. That is the savings.

These nanoseconds are mill timing on Atom16. They are not packet loss, not dollar egress, and not door wait.

Names (do not mix jars)

Mill ns — Atom16 rebuild timing (the shot).
Door wait — how long :1700 takes to answer (milliseconds).
Packet loss — drops on a path (often zero while wait is still real).
Spend egress — bytes or dollars leaving the lab without a law.
Heat-loss DOIs stay their own meter. Latency is not egress.