docs: frank-large cache pin verified warm (16x prefill) + depth law

Probe on the live endpoint (2026-10-06): 17.3k cold prefill 27.0 s ->
warm turn resumes at cache_n 16,384, prefill 1.7 s. The boot WARNING
'smaller than ONE full-context entry' sizes against the native 262,144
window, not our 200,700 slot — documented as expected in the checklist.
Tuning notes record the depth law (decode 5.5 t/s @ 46k -> 3.3-3.8 @ 80k
vs 30-50 shallow; KV reads ~8 GB/s effective; no supported flag touches
decode) and the AGENTS.md unit paragraph regains the unit-facts
parenthetical the reflow dropped.

AGENTS.md

@@ -323,11 +323,16 @@ audio encoder, and left again with the Gemma4 retirement on 2026-10-01
 The **halogen engine** (`ghcr.io/peonist-ai/halogen@sha256:dc0a39a0…`,
 digest-pinned, published tag 0.1.4) as a podman container driven by
 `scripts/serve-frank-large.sh` under a **root systemd unit**
-(`scripts/frank-large.service`, `User=dank`, `Restart=always`, `ExecStop=podman
+(`scripts/frank-large.service`, `User=dank`, `Restart=always`,
-stop -t 120 frank-large-halogen`, the script `exec podman wait` so the restart
+`ExecStop=podman
-policy is real) on port 7770 from the Strix Halo iGPU. Engine stdout goes to
+stop -t 120 frank-large-halogen`, the script runs `podman run` FOREGROUND —
-`/home/dank/logs/frank_large.log` — **the boot ledger (KV pool, drafter state,
+the container's exit code is the service's exit so `Restart=always` stays
-any clamp) lives there, not in the journal**. Serving
+real, and the engine ledger lands on disk) on port 7770 from the Strix Halo
+iGPU. Engine stdout goes to `/home/dank/logs/frank_large.log` — **the boot
+ledger (KV pool, prompt cache, drafter state, any clamp) lives there, not in
+the journal** (the 2026-10-05 `-d` + `podman wait` shape captured only the
+container ID — fixed 2026-10-06).
+Serving
 `peonist-ai/halogen-qwen3.8-27b`: `qwen3.8-27b-p1w4d-d2.hgn` (35,865,565,184 B
 ≈ 33.4 GiB, sha256 `274c3fc7…`, 1352 tensors; p1 weights / w4 activations /
 d2 KV) + the flat `tokenizer/` directory (5 files). Cutover 2026-10-05 from the ROCmFPX llama.cpp fork
@@ -361,6 +366,19 @@ temperature is greedy, so the thinking recipe exists only in `models.yml`
 Thinking lands in **`reasoning_content`** (harness field unchanged); the same
 Qwen3.8 template rejects `reasoning_effort` outside `xhigh|medium|low`, so the
 `models.yml` fold stays mandatory.
+`HALOGEN_CACHE_MB=16384` pins the prompt cache (2026-10-06): auto is OFF
+on this box — the floor is 2 full-context entries (36.9 GB) and the boot
+ledger had 35.7 GB free — so every deep request re-prefilled its whole
+79-81k context (84.9-179.8 s per request for ~500-650 new tokens per
+turn) until the pin. Probe-verified: a 17.3k cold prefill of 27.0 s
+resumes in 1.7 s on the next turn (cache_n 16,384). Decode-at-depth is
+structural: measured 5.5 t/s at 46k → 3.3-3.8 at 80k vs 30-50 shallow
+(the KV pool reads at ~8 GB/s effective vs ~180-240 for the weights);
+`docs/FLAGS.md` documents no decode-affecting lever and 0.1.4 is the
+latest engine. The boot WARNING "smaller than ONE full-context entry
+(18.4 GB at kMaxCtx=262144)" is a false positive for our slot: the
+deepest possible entry is 200,700 × ~70 KB ≈ 14 GB, inside the pin, and
+the 262k entry it fears cannot exist (over-slot is a 400).
 
 Fetch: `scripts/download-frank-large-halogen.sh` into
 `/home/dank/models/halogen-qwen3.8-27b/` (`-f` + byte count + LFS sha256 +

serverAppConfigs-docs/LLM/frank-large-halogen27b-dflash2.md

@@ -54,6 +54,15 @@ bearing numbers:
   set + host baseline on the 121.2 GiB box. The engine sizes the pool itself
   and **prints a boot ledger naming any clamp** — the post-deploy check reads
   that ledger, not a formula.
+- **`HALOGEN_CACHE_MB=16384` (pinned 2026-10-06)** — auto is OFF on this
+  box: the floor is 2 x a full-context entry (36.9 GB at 262k; one entry
+  is ~13.7 GB at our 200,700-slot x ~70 KB/token) and the ledger had
+  35.7 GB free, so the engine refused and every deep agent request
+  re-prefilled its whole 79-81k context — 84.9-179.8 s measured
+  (serve_api lines, 2026-10-06) for ~500-650 new tokens per turn. The
+  README prices the cache at ~20x time-to-first-token on a warm turn;
+  entries evict LRU by bytes; the reserve (HALOGEN_CACHE_RESERVE_MB,
+  default 8192) is re-checked against live MemAvailable on every store.
 - **No server-side sampler:** an absent temperature is GREEDY. The card's
   thinking recipe (temp 1.0 / top_p 0.95) lives in `models.yml` `extraBody`
   — on the wire only. The template defaults `reasoning_effort` to **xhigh**
@@ -94,10 +103,14 @@ bearing numbers:
   those links dangle inside the container mount.
 - **Unit:** `scripts/frank-large.service` (root, `User=dank`, relinked into
   `/etc/systemd/system/`), `ExecStart` = serve script, `ExecStop` = `podman
-  stop -t 120 frank-large-halogen`, script `exec podman wait` so
+  stop -t 120 frank-large-halogen`, script runs `podman run` FOREGROUND
-  `Restart=always` is real. Engine stdout goes to
+  (2026-10-06: the 2026-10-05 `-d` + `exec podman wait` shape captured
-  `/home/dank/logs/frank_large.log` (the boot ledger lives there, not in the
+  only the container ID — the engine ledger existed nowhere on disk and
-  journal).
+  the post-deploy grep below read a one-line hex file; foreground run
+  returns the container's exit code so `Restart=always` is real and the
+  ledger lands in the log). Engine stdout goes to
+  `/home/dank/logs/frank_large.log` (the boot ledger lives there, not in
+  the journal).
 - **Cutover:** `deploy_llm_large` in `scripts/deploy-remote.sh` sweeps the
   legacy bare `llama-server --model /home/dank/models/kingjones-*` process and
   the legacy `frank-large` container name by NAME (never `pkill -f
@@ -109,8 +122,15 @@ bearing numbers:
 
 ## Verification (post-deploy checklist)
 
-1. `grep -iE 'clamp|kv|pool|drafter|dflash' /home/dank/logs/frank_large.log`
+1. `grep -iE 'clamp|kv|pool|drafter|dflash|cache' /home/dank/logs/frank_large.log`
-   — ledger shows the pool at the asked shape, DFlash2 active, NO clamp line.
+   — ledger shows the pool at the asked shape, DFlash2 active, the
+   prompt cache line naming the pinned budget, NO clamp line. EXPECT the
+   WARNING "smaller than ONE full-context entry (18.4 GB at
+   kMaxCtx=262144)": it sizes against the NATIVE window, not our
+   200,700 slot — the deepest possible entry (~14 GB at ~70 KB/token)
+   fits the 16 GB pin, and the 262k entry it fears cannot exist (over
+   slot is a 400). Probe-proven 2026-10-06: the warm turn hit with
+   cache_n=16,384.
 2. `curl -s http://192.168.0.143:7770/health` -> 200.
 3. Real completion with `reasoning_effort` + temp/top_p from the wire recipe;
    `reasoning_content` populated; decode rate from the timings.
@@ -120,6 +140,38 @@ bearing numbers:
    incl. a tool call (V1TOOLS).
 6. Prompt > 200,700 tokens returns the 400 naming the limit (contract test).
 
+## Tuning notes (2026-10-06)
+
+Measured on real agent traffic (serve_api lines) the day after cutover:
+
+- **Decode is depth-bound, and the KV path is the reason.** Shallow
+  prompts (< 1k tokens) decode 30-50 t/s at commit 4-6.3/round; 46k runs
+  5.5, 63-67k runs 4.1-6.5, 79-81k runs 3.3-3.8. The round-time slope is
+  ~8 us/token of context: the KV pool reads at ~8 GB/s effective while
+  the weights stream ~180-240 GB/s (the engine's README puts batch-1
+  weight streaming at the 240 GB/s hardware wall). Commit also falls
+  from 4-6 to 2.5-3 accepted/round as the conversation grows.
+- **No supported flag touches decode.** `docs/FLAGS.md`: HALOGEN_W4A4 /
+  W4A4_EXCL are prefill-only ("decode is unaffected either way"); the
+  undocumented kernel levers are declared unsupported and their defaults
+  are the measured winners. 0.1.4 is the latest release (CHANGELOG
+  head; kernels unchanged since 0.1.0). The publisher's tg bench never
+  measured depth — its tg rows run over the ten SHORT eval prompt
+  shapes — so the README's "decode barely degrades at depth" claim was
+  never exercised at 46-80k.
+- **The lever that existed was the prompt cache, and it was OFF** (see
+  the shape bullet). The pin turns the 85-180 s re-prefill per deep
+  turn into a resume; decode itself is expected UNCHANGED by the pin —
+  the cache resumes prefill, it does not touch the round loop.
+  Probe-proven (two synthetic 17.3k turns, 2026-10-06): cold prefill
+  27.0 s (640 t/s, cache_n 0) -> warm turn cache_n 16,384, prompt_n 946
+  (the extension + one 2048-aligned chunk), prefill 1.7 s — 16x on the
+  pair, answer correct warm.
+- **Everything else stays put.** dflash2 over mtp (publisher 31.71 vs
+  26.88 mean; at depth both pay the same KV term and dflash2's higher
+  commit wins), KV_SLOTS=1 (batch XOR speculation), SLOT_CTX=200700
+  (Frank's window), W4A4 shipped (+9 % prefill, decode unharmed).
+
 ## Rollback
 
 Revert the frank-large swap commit and redeploy `dankclaw --service frank-large`:

serverAppConfigs-docs/LLM/index.md

@@ -81,6 +81,14 @@ frank-small, port 7771]
   No server-side sampler — the thinking recipe rides the wire in models.yml.
   Rollback: revert the swap commit; the ROCmFPX payload + fork build +
   `build-rocmfpx.sh` pin all remain on disk
+  Tuned 2026-10-06: `HALOGEN_CACHE_MB=16384` pins the prompt cache —
+  auto was OFF at the 36.9 GB two-entry floor, so every deep request
+  re-prefilled its 79-81k context (85-180 s each); a 17.3k cold prefill
+  of 27.0 s now resumes in 1.7 s on the warm turn (cache_n 16,384). The
+  serve script captures the engine ledger on disk (foreground `podman
+  run`; the `-d` era wrote only the container ID). Decode-at-depth is
+  structural: 3.3-3.8 t/s at 80k vs 30-50 shallow; no supported flag
+  touches it.
 - [[frank-small-llama-swift15-iq3s-mtp-131072]] — **ACTIVE since 2026-10-05
   (evening); 2 × 131,072 no-spec since 2026-10-06** (root systemd unit,
   port 7771): llama.cpp master (CUDA sm_86) serving ukisai/Swift-1.5-