Frank 846f81ddb8 docs(vault): the work log keeps incidents and laws; the numbers move to the page that owns them ...

log.md drops 466 lines of per-swap ledger prose (1,169 lines out, 703 back) and
keeps the causal stories: why the host went down, why a knob resets at boot, why
a stale drop-in served a different model for 24 minutes, why the RAM was the
fault. Date sections are unchanged (65 before, 65 after) -- this is compression
inside sections, not deletion of days.

Every number the compression would have orphaned was ported to its owner first,
and a census of the deleted lines against the whole vault is what found them:

- frank-zealot: the two boots that priced non-KV (20,190/KV 12,800 => 7,390
  without the projector; 22,972/KV 14,113 => 8,859 with it, so the vision tower
  costs ~1.4 GiB for a 918 MB file), the 1,153 MiB-free 3-slot boot that made
  the seam price non-KV at 9 GiB, and the f16 wall's own figures (fit 33,198 vs
  23,558 free on a 25,088.00 MiB KV request).
- frank-archon: the pool-cut-is-headroom-not-arena law (2.75 GiB/rank bought
  zero slots; d(reserve) = d(arena) = +248 slots; 14.87 GiB / 6,496 slots
  against a 398,658-token pool), the MNBT compile-RANGE vs SEQS+1 law, and the
  2 x 196,919 prefix-cache proof (393,837 tokens resident above the pool's
  386,904; 94.1/102.3/117.6 t/s, acceptance 0.7284/0.7525).
- frank-dragoon: the KL arm of the 2026-10-17 A/B (mean 0.155 / median 0.057,
  top-1 agreement 85.1 %) joins the PPL and NIAH figures already on the era row.
- Payload and Disk Reclaim: the 121 GB layer census lesson -- a huge image layer
  can be the engine image, not a payload.

Also fixes a wikilink the rewrite had hard-wrapped across two lines, which the
link checker reads as a target that does not exist.

New seam: 'one page per live LLM endpoint'. The live set is derived from
scripts/ (an endpoint has both an env and a unit), and any <endpoint>-*.md page
in the vault fails it -- the enforcement of the one-page-per-endpoint law, so a
re-created era page cannot pass tests.

2026-10-19 01:05:32 -07:00

..

Pipeline Parallelism

docs(vault): one page per endpoint, forever -- 45 era pages fold into three

2026-10-19 00:49:03 -07:00

Dankclaw Improvements.md

docs(vault): one page per endpoint, forever -- 45 era pages fold into three

2026-10-19 00:49:03 -07:00

Dankclaw System Specs.md

docs(vault): one page per endpoint, forever -- 45 era pages fold into three

2026-10-19 00:49:03 -07:00

dankclaw-iommu-translation-off.md

docs(vault): one page per endpoint, forever -- 45 era pages fold into three

2026-10-19 00:49:03 -07:00

Dankserver System Specs.md

docs(vault): one page per endpoint, forever -- 45 era pages fold into three

2026-10-19 00:49:03 -07:00

dankserver-frontier-moe-feasibility.md

docs(vault): one page per endpoint, forever -- 45 era pages fold into three

2026-10-19 00:49:03 -07:00

dankserver-vram-growth.md

docs(vault): one page per endpoint, forever -- 45 era pages fold into three

2026-10-19 00:49:03 -07:00

dealign-ornith-crack.md

llm: retire the trio service names — fleet services renamed to frank-archon/frank-dragoon/frank-zealot

2026-10-10 17:03:14 -07:00

deployment.md

docs(frank-tts): what makes a locally built voice image stale, and what the warm text must survive

2026-10-18 18:07:19 -07:00

DFlash DSpark Speculative Decoding.md

docs(vault): one page per endpoint, forever -- 45 era pages fold into three

2026-10-19 00:49:03 -07:00

Dual-R9700-Cooperation.md

docs(vault): one page per endpoint, forever -- 45 era pages fold into three

2026-10-19 00:49:03 -07:00

FastFlowLM NPU Inference.md

docs(vault): one page per endpoint, forever -- 45 era pages fold into three

2026-10-19 00:49:03 -07:00

frank-archon.md

docs(vault): the work log keeps incidents and laws; the numbers move to the page that owns them

2026-10-19 01:05:32 -07:00

frank-dragoon.md

docs(vault): the work log keeps incidents and laws; the numbers move to the page that owns them

2026-10-19 01:05:32 -07:00

frank-medium-batch-ubatch.md

docs(vault): one page per endpoint, forever -- 45 era pages fold into three

2026-10-19 00:49:03 -07:00

frank-medium-llamacpp-dflash2.md

docs(vault): one page per endpoint, forever -- 45 era pages fold into three

2026-10-19 00:49:03 -07:00

frank-medium-model-swap.md

test(frank-medium): A/B DFlash2 vs MTP on FP8 — DFlash2 wins on decode

2026-09-20 22:20:53 -07:00

frank-medium-radiance-mxfp4.md

llm: retire the trio service names — fleet services renamed to frank-archon/frank-dragoon/frank-zealot

2026-10-10 17:03:14 -07:00

frank-medium-rocmfpx.md

docs(vault): one page per endpoint, forever -- 45 era pages fold into three

2026-10-19 00:49:03 -07:00

frank-npu.md

docs(vault): one page per endpoint, forever -- 45 era pages fold into three

2026-10-19 00:49:03 -07:00

frank-redo-2026-08-17.md

docs(vault): one page per endpoint, forever -- 45 era pages fold into three

2026-10-19 00:49:03 -07:00

frank-zealot.md

docs(vault): the work log keeps incidents and laws; the numbers move to the page that owns them

2026-10-19 01:05:32 -07:00

FreeToken Builds.md

llm: retire the trio service names — fleet services renamed to frank-archon/frank-dragoon/frank-zealot

2026-10-10 17:03:14 -07:00

freetoken-benchmark-sweep.md

docs(vault): one page per endpoint, forever -- 45 era pages fold into three

2026-10-19 00:49:03 -07:00

GGUF Compatibility.md

docs(vault): one page per endpoint, forever -- 45 era pages fold into three

2026-10-19 00:49:03 -07:00

Heterogeneous Split Triage.md

docs(vault): fleet-wide fact re-verification pass (67 pages)

2026-10-18 13:15:43 -07:00

Hybrid CPU GPU Processing Dankserver.md

docs(vault): one page per endpoint, forever -- 45 era pages fold into three

2026-10-19 00:49:03 -07:00

ikllama-ornith-dankserver.md

llm: retire the trio service names — fleet services renamed to frank-archon/frank-dragoon/frank-zealot

2026-10-10 17:03:14 -07:00

index.md

docs(vault): one page per endpoint, forever -- 45 era pages fold into three

2026-10-19 00:49:03 -07:00

kingjones-q8agent-rocmfpx.md

docs(vault): one page per endpoint, forever -- 45 era pages fold into three

2026-10-19 00:49:03 -07:00

ling-3.0-flash-trial.md

docs(vault): one page per endpoint, forever -- 45 era pages fold into three

2026-10-19 00:49:03 -07:00

ling-3.0-flash-vllm-plan.md

llm: retire the trio service names — fleet services renamed to frank-archon/frank-dragoon/frank-zealot

2026-10-10 17:03:14 -07:00

Llama Cpp Multi-Device Splits.md

docs(vault): fleet-wide fact re-verification pass (67 pages)

2026-10-18 13:15:43 -07:00

Llama Cpp Serving Sizing.md

docs(vault): one page per endpoint, forever -- 45 era pages fold into three

2026-10-19 00:49:03 -07:00

LLM Bench Windows and Measurement.md

llm: retire the trio service names — fleet services renamed to frank-archon/frank-dragoon/frank-zealot

2026-10-10 17:03:14 -07:00

LLM Performance Diagnosis.md

llm: retire the trio service names — fleet services renamed to frank-archon/frank-dragoon/frank-zealot

2026-10-10 17:03:14 -07:00

Model Swap and Rollback.md

docs(vault): one page per endpoint, forever -- 45 era pages fold into three

2026-10-19 00:49:03 -07:00

PCIe Link and Atomics.md

docs(vault): one page per endpoint, forever -- 45 era pages fold into three

2026-10-19 00:49:03 -07:00

Quantization by Hardware Architecture.md

docs: OKF v0.1 compliance pass + fact-check + 3 research briefs

2026-09-22 02:09:41 -07:00

Quantization Production.md

docs(vault): one page per endpoint, forever -- 45 era pages fold into three

2026-10-19 00:49:03 -07:00

qwen3.8-spec-matrix.md

docs(vault): one page per endpoint, forever -- 45 era pages fold into three

2026-10-19 00:49:03 -07:00

qwen3.8-strix-optimizations.md

docs(vault): one page per endpoint, forever -- 45 era pages fold into three

2026-10-19 00:49:03 -07:00

RAM CPU Offloading in VLLM.md

docs: OKF v0.1 compliance pass + fact-check + 3 research briefs

2026-09-22 02:09:41 -07:00

RDNA4 Container Serving.md

docs(vault): one page per endpoint, forever -- 45 era pages fold into three

2026-10-19 00:49:03 -07:00

RDNA4 MoE Expert Offload VRAM Budget.md

docs(vault): one page per endpoint, forever -- 45 era pages fold into three

2026-10-19 00:49:03 -07:00

Running Qwen3.6 27B FP8 on Dankclaw.md

llm: retire the trio service names — fleet services renamed to frank-archon/frank-dragoon/frank-zealot

2026-10-10 17:03:14 -07:00

Running Qwen3.6 35B A3B FP8 on Dankserver.md

llm: retire the trio service names — fleet services renamed to frank-archon/frank-dragoon/frank-zealot

2026-10-10 17:03:14 -07:00

Shared Memory Architecture Dankclaw.md

docs(vault): one page per endpoint, forever -- 45 era pages fold into three

2026-10-19 00:49:03 -07:00

swift15-campaign-worklog.md

llm: retire the trio service names — fleet services renamed to frank-archon/frank-dragoon/frank-zealot

2026-10-10 17:03:14 -07:00

Tcclaviger Fork Ops.md

llm: retire the trio service names — fleet services renamed to frank-archon/frank-dragoon/frank-zealot

2026-10-10 17:03:14 -07:00

ubatch-benchmark-large.md

docs(vault): one page per endpoint, forever -- 45 era pages fold into three

2026-10-19 00:49:03 -07:00

vLLM Boot Forensics and Config Interrogation.md

docs(vault): one page per endpoint, forever -- 45 era pages fold into three

2026-10-19 00:49:03 -07:00

VLLM Index.md

docs(vault): one page per endpoint, forever -- 45 era pages fold into three

2026-10-19 00:49:03 -07:00

VLLM Overview.md

llm: retire the trio service names — fleet services renamed to frank-archon/frank-dragoon/frank-zealot

2026-10-10 17:03:14 -07:00

vLLM Porting and Compile Triage.md

docs(vault): one page per endpoint, forever -- 45 era pages fold into three

2026-10-19 00:49:03 -07:00

vllm_research_findings.md

docs: OKF v0.1 compliance pass + fact-check + 3 research briefs

2026-09-22 02:09:41 -07:00

x4-pair-experiment.md

docs(vault): one page per endpoint, forever -- 45 era pages fold into three

2026-10-19 00:49:03 -07:00