docs(AGENTS): fleet topology + llama.cpp laws for the PQ2_0 zealot era

- topology row + zealot section: PrismML vendor build, Bonsai-2 PQ2_0,
  q8_0 KV, measured ledger; runbook index active/retired rows
- new durable laws: custom-tensor-type runtime door (vendor build is forced),
  mmproj outside the engine fit projection, first-request warmup artifact,
  thinking-budget probe traps + load-bearing reasoning_effort fold

AGENTS.md

@@ -27,7 +27,7 @@ merge absorbed the former `vllmScripts` repo, history preserved).
 |--------|---------|---------|-------|------|
 | Dankserver | `frank-archon` | **radiance** (StillDeadcode, **pin 1.0.13** digest-pinned `stilldeadcode/radiance@sha256:4c3f484564…`, plain `podman run`, root systemd unit) | **Publisher payload** (swap 2026-10-15, MODEL_FILE-only): `StillDeadcode/qwen3.8-next-flash-fp8-iq4r-moe` → **`qwen3.8-next-flash-fp8-iq4r-moe.rad`** (121,969,901,568 B = **116,322 MiB**, Hub-LFS sha256 `0af5e962…` verified at fetch by `scripts/download-frank-archon-iq4r.sh`) — the OFFICIAL Qwen3.8-Flash-Next bf16 checkpoint rad-converted by the engine's author: routed experts as **4-bit codes into a 16-level non-uniform codebook** (128-pt Walsh–Hadamard rotation, GPTQ on a 10M-token calibration + 3 coordinate-descent sweeps, E4M3 scales per 64 weights), **10 whole bf16-protected experts/layer**, int8 W8A8 trunk/lm_head, fp8 HC mixtures, fp8 fixed-scale PLE n-gram (47.7 GiB), **MTP head + bf16 vision tower in the container**; card KL 0.0789 mean all-tokens / top-1 92.85 % vs BF16. **Base-alignment (censored) model** — the abliterated family lives on frank-dragoon. Shape carried VERBATIM from the 2026-10-13 record: **3 × 200,000 fp8-KV** (`--max-num-seqs 3`), **25,600 MiB slab + 5,120 MiB pool = 30,720 of the 30,969 MiB claimable**, 44 GiB/rank pinned host pool, PLE pinned to RAM, MTP depth 2, MNBT 2048, Froggeric v22.5 template. Decode/prefill ladders in the runbook are the **w4dn8 record** (the bar to beat; new-payload probes pending). Rollback: revert the swap commit (w4dn8 `.rad` on disk); the fp8-native `.rad` + the raw Swift1.5 BF16 source were DELETED from disk 2026-10-15 (Frank's one-time rm, volume for this payload; rebuildable via the fetch/build scripts). | 7770 |
 | Dankclaw | `frank-dragoon` | **halogen-flash-server 0.16.2** (`ghcr.io/peonist-ai/halogen-flash-server@sha256:0c61bf84…`, digest-pinned, plain `podman run` via `scripts/serve-frank-dragoon.sh`, root systemd unit) | Quat3rnion/halogen-swift1.5-qwen3.8-flash-next-v2-abliterated **`qwen38-flash-next-v2-swift15-abliterated.hgn`** (66,687,678,464 B = compacted r2 of the card's 89,775,868,032 B original, ~62 GiB resident; the v2 `.hgn` format loads in NO other engine), **+ REQUIRED `qwen38-flash-next-ngram.hgn`** (51,200,246,144 B, peonist @5cc17cea — read beside the checkpoint) **+ vision tower** (897,916,416 B, @e053f488 — **vision ON**, `image_url` parts only, ≥256 px probes), **MTP spec decode ALWAYS-ON (`HALOGEN_SPEC_ADAPT=0` — the adaptive default disables the drafter per request and HALVES decode)**, MTP head INSIDE the checkpoint (`HALOGEN_MTP_HEAD` unset; no sidecar fetched), **2 slots × 262,144 ctx** (KV pool 524,288 positions ≈ 14.0 GiB at ~28 KiB/position; the 27B slots>1-disables-drafting law does NOT carry over — batching and speculation coexist, the card's quickstart ships 2 with the head), harness window 217,088 = ceiling − 45k estimate headroom, **server-side sampler 1.0/0.95/20** (card recipe; models.yml extraBody matches), coupled cap/timeout 65536/7200, prompt cache ON by default (warm 100k turn ≈ 2 s vs ≈ 88 s cold)|7770|
-| Dankclaw | `frank-zealot` | llama.cpp master (0.5.0-dev, commit `5fc4f3c`), CUDA sm_86 build (root systemd unit, relinked) | HauhauCS/Qwen3.5-9B-Uncensored-HauhauCS-Aggressive **Q6_K** (7,359,259,008 B; dense `qwen35` hybrid: 32 layers, 8 full-attention + 24 Gated DeltaNet, 4 KV heads × 256 dim, vocab 248,320, native 262,144, **no MTP tensors in the file**) + the publisher's own **BF16 mmproj** (921,704,448 B) for **vision**, on the RTX 3090, **TWO 200,704-token slots** (`--ctx-size 401408 --parallel 2`), **exact f16 K + f16 V** under flash attention (32 KiB/token over the 8 full layers = 12.25 GiB pool), batch 4096 / **ubatch 2048** (measured CUDA optimum), `--log-verbosity 4`, **MTP OFF** (forced: no `nextn` tensors exist). Priced from the MiMo Q6_K boots (same geometry, same card, same shape): 21,276/2,849 at ubatch 1024, 21,846/2,279 at 2048. Cutover 2026-10-07 (late) from Gemma4-26B-A4B Q4_K_M 1×200,704. All three endpoints route harness **compaction** here. | 7771 |
+| Dankclaw | `frank-zealot` | **llama.cpp PrismML fork build** (vendor release `0.2.0-dev` build **10754**, commit `2459f68b5`; root systemd unit, relinked — the ONLY loader for this payload) | `prism-ml/Ternary-Bonsai-2-27B-gguf` **PQ2_0** (7,206,168,928 B, sha256 `3907dc16…`; `qwen35` hybrid: **64 blocks, 16 full-attention + 48 Gated DeltaNet**, 4 KV heads × 256 dim, vocab 248,320, native 262,144, **no MTP head in the file**; `general.file_type 141` = custom tensor type **142** + `prism.hadamard.*` dequant metadata, which is what locks the runtime) + the publisher's own **Q8_0 mmproj** (629,246,976 B; Qwen3-VL tower, 27 blocks, 768 px, `qwen3vl_merger`) for **vision**, on the RTX 3090, **TWO 200,704-token slots** (`--ctx-size 401408 --parallel 2`), **q8_0 K + q8_0 V** under flash attention (34 KiB/token over the 16 full layers = 13,328 MiB pool; **f16 KV needs 25,088 MiB and cannot boot this card**), batch 4096 / **ubatch 2048** (measured CUDA optimum), `--log-verbosity 4`, **MTP OFF** (no head in the file). Boot 22,664 used / 1,461 free; 76.6 t/s single-stream decode, 691.7 t/s prefill at 166k. Cutover 2026-10-17 from Qwen3.5-9B-Uncensored Q6_K (same shape; payload + master CUDA build on disk as rollback). |
 
 Retired endpoints (`frank-npu`, and every prior runtime) are in
 the [runbook index](#runbook-index). The full swap history lives in
@@ -429,69 +429,89 @@ Rollback: revert the swap commit and redeploy — the 27B payload
 halogen 0.1.4 image digest all remain on disk; the ROCmFPX door behind
 that (`scripts/build-rocmfpx.sh`, pin `36e9acd`) is untouched.
 
-### frank-zealot — llama.cpp on the RTX 3090 (Dankclaw, ACTIVE)
+### frank-zealot — llama.cpp (PrismML fork) on the RTX 3090 (Dankclaw, ACTIVE)
 
-**llama.cpp master** (0.5.0-dev, commit `5fc4f3c`; CUDA sm_86 build
-`/home/dank/GitPackages/llama-master/build-cuda/bin/llama-server`,
-`GGML_CUDA_FA_ALL_QUANTS=ON`) as a **root systemd unit**
+**The PrismML fork build of llama.cpp** (vendor release, `--version` =
+`0.2.0-dev (build 10754, commit 2459f68b5)`;
+`/home/dank/GitPackages/llama-prism/llama-server`) as a **root systemd unit**
 (`scripts/frank-zealot.service`, `relink_unit`ed into
-`/etc/systemd/system/`), serving
-`HauhauCS/Qwen3.5-9B-Uncensored-HauhauCS-Aggressive` at **Q6_K**
-(7,359,259,008 B, sha256 `c0ba7beb…` computed locally — the Hub carries no
-LFS oid) **plus the publisher's own BF16 mmproj** (921,704,448 B), both
-`/home/dank/models/Qwen35-9B-HauhauCS-Q6K/`, on port 7771. Cutover
-2026-10-07 (late) from the Gemma4-26B-A4B Q4_K_M 1×200,704 shape (same day).
+`/etc/systemd/system/`), serving `prism-ml/Ternary-Bonsai-2-27B-gguf` at
+**PQ2_0** (7,206,168,928 B, sha256 `3907dc16…` = the Hub LFS oid) **plus the
+publisher's own Q8_0 mmproj** (629,246,976 B), both
+`/home/dank/models/Bonsai2-PQ2_0/`, on port 7771. Cutover 2026-10-17 from the
+Qwen3.5-9B-Uncensored Q6_K era (same shape).
 
-**Architecture: qwen35 dense hybrid** — read from the GGUF's own header, not
-the model card: 32 layers, `full_attention_interval = 4` ⇒ **8 full-attention
-layers (3, 7, 11, …, 31) + 24 Gated DeltaNet linear layers**, 16 attention
-heads / **4 KV heads** × 256 dim, `ssm` conv_kernel 4 / state_size 128 /
-group_count 16 / time_step_rank 32 / inner_size 4096, native 262,144, vocab
-248,320. Tensor types are Q6_K × 250 + F32 × 177 — standard quants only, so
-the stock master build loads it (arch `qwen35` is registered in this build's
-`src/llama-arch.cpp`). **KV accounting:** the pool scales with the 8 full
-layers only — 8 × 4 × 256 = 8,192 elements/token per cache, so f16 K + f16 V
-bills **32 KiB/token** = 12.25 GiB at the 401,408 pool; the 24 DeltaNet
-layers bill a fixed recurrent state per SLOT, not per token, which is exactly
-why two 200k slots fit where the 26B MoE fit one.
+**The runtime is part of the payload identity.** The file is
+`general.file_type 141` (PQ2_0, 2.13 bpw, group 128) with **402 tensors of the
+custom GGML type 142** plus `prism.hadamard.*` metadata (normalized
+Sylvester–Walsh transform, block 1024, 401 named weights, inverse on
+`token_embd`, `gdn_v_grouped`). Stock llama.cpp refuses type 142 at load, and a
+loader that ignored the Hadamard metadata would dequant wrongly — same lock class
+as the ROCmFPX era. Retargeting the payload without the fork ships an endpoint
+that cannot load, so the seam pins both paths.
 
-**The shape (2026-10-07):** `--ctx-size 401408 --parallel 2` = TWO
+**Architecture: qwen35 hybrid** — read from the GGUF's own header: **64
+blocks**, `full_attention_interval = 4` ⇒ **16 full-attention + 48 Gated
+DeltaNet layers**, 24 attention heads / **4 KV heads** × 256 dim, embedding
+5,120, ffn 17,408, `ssm` conv_kernel 4 / state_size 128 / group_count 16 /
+time_step_rank 48 / inner_size 6,144, native 262,144, vocab 248,320. **KV
+accounting:** the pool scales with the 16 full layers — 16 × 4 × 256 = 16,384
+elements/token per cache, so **q8_0 K + q8_0 V bills 34 KiB/token** = 13,328
+MiB at the 401,408 pool (ledger-confirmed to the MiB), while **f16 KV needs
+25,088 MiB and cannot boot on a 24,123 MiB card** (measured `cudaMalloc` OOM).
+The 48 DeltaNet layers bill a fixed recurrent state per SLOT — ~150 MiB each —
+so pool arithmetic alone under-prices the shape.
+
+**The shape (2026-10-17):** `--ctx-size 401408 --parallel 2` = TWO
 conversations of 200,704 (Frank's cap; the harness `contextWindow` quotes
-the per-slot number), `--flash-attn on`, **f16 K + f16 V** (exact KV),
-batch 4096 / **ubatch 2048** (the measured CUDA optimum: +4.6 % prefill,
-decode flat, 616 MiB — MiMo-era A/B on this card), `--log-verbosity 4`,
-`--mmproj` for vision, **speculative decoding OFF** (`SPEC_TYPE=none`).
+the per-slot number), `--flash-attn on`, **q8_0 K + q8_0 V**, batch 4096 /
+**ubatch 2048** (the measured CUDA optimum carried from the MiMo/Swift A/Bs on
+this card), `--log-verbosity 4`, `--mmproj` for vision, **speculative decoding
+OFF** (`SPEC_TYPE=none`). Boot: **22,664 MiB used / 1,461 free**; the
+projection (21,439) excludes the projector's 873.10 MiB worst-case, which
+lands AFTER the fit check — read settled `nvidia-smi`, never the projection.
+Measured: 76.6 t/s single-stream decode (the 9B ran 94.4), 1,258 t/s prefill
+at 35k, 691.7 t/s at 166k, 109.4 t/s aggregate across two concurrent streams.
+The **first request after boot is a warmup artifact** (2.6 t/s prefill / 37.7
+t/s decode measured) — never report it.
 
-**Why MTP is off**: it is not a choice — **this Q6_K file carries no
-`nextn`/`mtp` tensors** (verified from the tensor-name list), and MTP is a
-property of the file, not of the flags. The publisher card advertises "MTP
-support" for the base architecture; the quantized export dropped the head.
+**Why MTP is off**: the PQ2_0 export carries no MTP head (upstream ships a
+separate `…-CRACK-MTP-GGUF`, +451 MB, for that), and MTP is a property of the
+file, not of the flags.
 
-**Measured boots** (nvidia-smi, the referee): the deployed boot is
-**21,436 used / 2,689 free** (ubatch 2048) — the pricing boots were the
-MiMo-V2.6-Distill-Qwen-9B pair at this exact geometry (**21,276 / 2,849** at
-ubatch 1024, **21,846 / 2,279** at 2048), so the 415 MiB lighter weights
-bought back what the projection predicted. The engine ledger confirms the KV
-law to the MiB: `CUDA0 KV buffer size = 12544.00 MiB` (`K (f16): 6272.00`,
-`V (f16): 6272.00`), compute buffers 1,136 MiB on CUDA0 + 848 MiB host, the
-mmproj 879 MiB + 248 MiB. Decode is ubatch-invariant (Swift-era 3090 law). The
-Q6_K rung costs ~10 % decode vs Q5_K_M (weight bandwidth; MiMo-measured 94.4
-vs 104.8 t/s).
-
-**Vision**: the publisher's own BF16 mmproj via `--mmproj` (the model is
-natively multimodal; the Q6_K file carries no vision tensors). Probe traps
+**Vision**: the publisher's own Q8_0 mmproj via `--mmproj` (Qwen3-VL tower: 27
+blocks, 768 px, patch 16, merge 2, `qwen3vl_merger`, deepstack). Probe traps
 carry over: ≥256 px probe images or the model answers "video", and only the
-OpenAI `image_url` content part reaches the model.
+OpenAI `image_url` content part reaches the model. **New trap:** thinking
+tokens count against `max_tokens`, so a vision probe at 64 tokens returns
+EMPTY `content` with `finish_reason: "length"` — budget ≥512 or the endpoint
+looks blind while it is only mid-thought.
+
+**Base alignment**: Bonsai-2 is the censored BASE lineage (the 9B it replaced
+was uncensored). The CRACK re-export on disk
+(`/home/dank/models/Bonsai2-CRACK/`, header census proves byte-identical
+geometry) is a one-line `MODEL_FILE` swap if a refusal ever surfaces.
 
 **Compaction hub**: `models.yml` routes `compactionModel` for
 `frank-archon`, `frank-dragoon` and `frank-zealot` (the 3090 endpoint) here — 200,704 is the
 only window in the fleet that fits a worst-case compaction input (~170k
 tokens). With `--parallel 2` a compaction request no longer serializes behind
-a live generation (the Gemma4 era's `--parallel 1` law).
+a live generation (the Gemma4 era's `--parallel 1` law). Retrieval through a
+25,960-token prompt was verified exact after the swap.
+
+**Contract law learned at verbosity 4:** a raw `/completion` request prints
+`failed to parse reasoning content (… parse_reasoning=true)` whenever the
+model closes a bare prompt with `` — cosmetic
+(`probe_request.cpp:789`); the `/v1/chat/completions` path parses
+`reasoning_content` correctly, which is the only path consumers use. Do not
+lower `VERBOSITY` to silence it: the KV/buffer ledger is what VRAM forensics
+read.
 
 Runbook:
-`serverAppConfigs-docs/LLM/frank-zealot-llama-qwen35-9b-q6k-2x200704.md`.
-Rollback: revert the swap commit — the Gemma4 payload
+`serverAppConfigs-docs/LLM/frank-zealot-llama-prism-bonsai2-pq20.md`.
+Rollback: revert the swap commit — the Qwen3.5-9B Q6_K payload
+(`/home/dank/models/Qwen35-9B-HauhauCS-Q6K/`) with the master CUDA build
+(`~/GitPackages/llama-master/build-cuda/`), the Gemma4 payload
 (`/home/dank/models/Gemma4-26B-A4B/`), the MiMo Q6_K payload
 (`/home/dank/models/MiMo-Q6K/`), the Swift payload
 (`/home/dank/models/Swift15-UncIQ3XXS/`) and the Ornith Q8_0 payload
@@ -576,7 +596,8 @@ accept 0.7246).
 
 **The KV dtype law (measured 2026-10-11, applied at 4 slots 2026-10-13)**:
 `--kv-cache-dtype` takes ONLY `bf16|fp8` — one dtype for the whole cache, so
-llama.cpp's exact-K/quantized-V pair (what frank-zealot runs) has NO radiance
+llama.cpp's exact-K/quantized-V pair (the 9B-era frank-zealot ran it; the
+Bonsai-2 swap runs q8_0/q8_0) has NO radiance
 equivalent; the rung under bf16 is fp8 on both halves. Measured per rank from
 the two boot ledgers: `kv_attn` (13 layers, 4-token blocks) pages 26.00 → 52.00
 KiB, while `kv_qsa_bkey` (13 layers) stays 3.25 KiB at BOTH dtypes (the QSA
@@ -1021,6 +1042,31 @@ linked runbooks.
   `Unexpected reasoning effort high` until `high: xhigh` was added on
   2026-09-16. Halogen accepts the wider set, so the two entries legitimately
   differ.
+- **The vision projector is NOT billed in llama.cpp's own fit check.** The
+  engine prints `[mtmd] estimated worst-case memory usage of mmproj is
+  873.10 MiB` and then fits WITHOUT it: frank-zealot's Bonsai-2 boot projected
+  21,439 MiB and settled at 22,664 (2026-10-17). Price a 3090-class shape from
+  settled `nvidia-smi` after a real boot, never from
+  `model + context + compute` — the projector (and any drafter context) lands
+  after `/health` would already have answered. Same failure class as the
+  DSpark drafter OOM.
+- **The FIRST request after a boot is a warmup artifact.** Measured on
+  Bonsai-2/3090: 2.6 t/s prefill and 37.7 t/s decode on request #1, then
+  1,258 t/s prefill and 54.7 t/s (per stream, two concurrent) on the next — a
+  3× spread on the same shape. Benchmark the second request or later, or the
+  graph-capture cost reads as a model fault.
+- **Thinking tokens count against `max_tokens`, so a short-budget vision probe
+  reads as blindness.** The Bonsai-2/`image_url` probe at `max_tokens: 64`
+  returned `content: ""` with `finish_reason: "length"` and half a thought in
+  `reasoning_content` — the image HAD reached the model (322 prompt tokens vs
+  42 for the text alone). At 512 tokens the same image answers "Red". Budget
+  ≥512 for any thinking model you probe for vision.
+- **Quantized KV is not the only dtype law a fork build can hide.** Vendor
+  builds carry their own compile options: frank-zealot's PrismML prebuilt
+  served `--cache-type-v q8_0` under flash attention at the exact 34,816
+  B/token (the `GGML_CUDA_FA_ALL_QUANTS` door), so a build decision must be
+  proven by the KV buffer line in the boot ledger, not by the build flags you
+  remember passing.
 
 **Ops**
 
@@ -1092,7 +1138,7 @@ Active:
 | Service | Runbook |
 |---------|---------|
 | frank-dragoon (flash-server 0.16.2, abliterated Swift1.5 v2 .hgn + MTP always-on, 2×262,144, vision ON, Dankclaw) | `serverAppConfigs-docs/LLM/frank-dragoon-flash-server-v2-swift15.md` |
-| frank-zealot (llama.cpp CUDA, Qwen3.5-9B-Uncensored Q6_K, 2×200,704 exact f16 KV MTP-off, vision ON, Dankclaw) | `serverAppConfigs-docs/LLM/frank-zealot-llama-qwen35-9b-q6k-2x200704.md` |
+| frank-zealot (llama.cpp PrismML fork, Bonsai-2-27B PQ2_0, 2×200,704 q8_0 KV MTP-off, vision ON, Dankclaw) | `serverAppConfigs-docs/LLM/frank-zealot-llama-prism-bonsai2-pq20.md` |
 | frank-archon (radiance StillDeadcode, publisher Qwen3.8-Flash-Next iq4r .rad, 2×R9700 TP=2, Dankserver) | `serverAppConfigs-docs/LLM/frank-archon-radiance-qwen38-iq4r.md` |
 | Dankclaw kernel posture (`amd_iommu=off`) | `serverAppConfigs-docs/LLM/dankclaw-iommu-translation-off.md` |
 | Deployment flow + incidents | `serverAppConfigs-docs/LLM/deployment.md` |
@@ -1102,6 +1148,7 @@ Retired (rollback targets and history):
 | Runtime | Host | Window | Runbook |
 |---------|------|--------|---------|
 | radiance Swift1.5-Qwen3.8-Flash-Next GPTQ w4dn8, 3×200,000 fp8-KV slab 25600 (`frank-archon`) | Dankserver | 2026-10-08 → 10-15 (rollback target — `.rad` on disk; owns the slab-slope ladder, KV dtype law, KLD campaign, and every shape law the iq4r swap carried verbatim) | `frank-archon-radiance-swift15-w4dn8.md` |
+| llama.cpp CUDA (master) Qwen3.5-9B-Uncensored Q6_K, 2×200,704 exact f16 KV MTP-off (`frank-zealot`) | Dankclaw | 2026-10-07 → 10-17 (rollback target — payload + master CUDA build on disk; owns the 32 KiB/token 8-full-layer KV law and the 2→3-slot reslice law) | `frank-zealot-llama-qwen35-9b-q6k-2x200704.md` |
 | llama.cpp CUDA Gemma4-26B-A4B QAT Q4_K_M, 1×200,704 f16-K/q8_0-V MTP-off (`frank-zealot`) | Dankclaw | 2026-10-07 → 10-07 (same-day swap; rollback target — payload on disk; owns the `attention_k_eq_v` pair-billing KV law and the gemma4-assistant drafter-cost boots) | `frank-zealot-llama-gemma4-q4km-200704.md` |
 | ROCmFPX llama.cpp fork Qwen3.8-27B Uncensored Q6_0_ROCMFPX_AGENT + MTP (`frank-dragoon`) | Dankclaw | 2026-10-03 → 10-05 (rollback target, payload + fork build on disk) | `frank-dragoon-rocmfpx-q6agent-uncensored-mtp.md` |
 | tcclaviger vLLM fork Qwen3.8-Flash-Next-Uncensored FP8, TP=2, shape G (`frank-archon`) | Dankserver | 2026-09-29 → 10-05 (rollback target — tag `frank-archon-vllm-shapeG-20261003`; trio + env lore in the tag, image in rootless storage) | `dankserver-qwen38-fp8-expert-offload-tuning.md` |