chore: adopt pre-existing working-tree state before trunk merge

Uncommitted machine/session edits (live config, models, docs drift) were
blocking the merge of origin/main. Committed as-is to preserve them; the
merge that follows resolves overlapping prose files toward upstream
canonical, with local variants recoverable from this commit. Pre-merge
copies also in .AGENTTRASH/pre-merge-2026-10-18/.

agent/models.yml

@@ -16,25 +16,25 @@ providers:
     auth: none
     models:
       - id: frank-archon
-        # Model id renamed from frank-primary on 2026-10-08 (official
+        # Model id renamed from frank-archon on 2026-10-08 (official
         # unit-name scheme; the systemd service, scripts and runbooks keep
-        # the frank-primary name — service names are unchanged).
+        # the frank-archon name — service names are unchanged).
         # Server-side compaction routes to compaction.remoteEndpoint in
         # config.yml (192.168.0.143:7771 — frank-zealot's 3090); the
         # compactionModel candidate keeps the wire model id valid there.
         compactionModel: llamacpp (dankclaw-3090)/frank-zealot
-        name: "frank-archon (dankserver)"
+        name: "frank-archon"
         # Dankserver: BOTH Radeon AI PRO R9700s (gfx1201, TP=2 over the
         # 2026-09-20 board swap's root ports, wht6 wire). Engine:
         # radiance (StillDeadcode, Apache-2.0) from the digest-pinned
         # stilldeadcode/radiance image (index digest sha256:4c3f4845..., 287
         # MiB, kernels prebuilt for gfx1201), plain `podman run` through
-        # scripts/serve-frank-primary.sh, root unit (User=dank, dank's
+        # scripts/serve-frank-archon.sh, root unit (User=dank, dank's
         # rootless podman storage). Serving the OFFICIAL
         # Qwen/Qwen3.8-Flash-Next-FP8 at NATIVE modelopt block-FP8
         # precision — a rad-convert no-recipe pure copy into
         # qwen4exp-fp8-native.rad (172.76 GiB measured by plan-only), built by
-        # scripts/build-frank-primary-rad.sh against a staging dir whose
+        # scripts/build-frank-archon-rad.sh against a staging dir whose
         # config.json drops quantization_config so the shipped qwen4exp
         # plugin's (qwen4exp, "") selection key matches; per-weight
         # encodings are untouched. Experts stream from a 40 GiB/rank
@@ -44,7 +44,7 @@ providers:
         # froggeric's fixed qwen3.8 v22.5
         # (scripts/qwen38-froggeric-v22.5.jinja) via
         # --override-chat-template. Rollback: git tag
-        # frank-primary-freetoken-pr587-20261005 (FreeToken, one card).
+        # frank-archon-freetoken-pr587-20261005 (FreeToken, one card).
         contextWindow: 200000
         # maxTokens must survive the harness compaction gap AND the chars/4
         # token-estimate under-run on code-heavy prompts. The clamp is
@@ -96,8 +96,8 @@ providers:
     auth: none
     models:
       - id: frank-dragoon
-        # Model id renamed from frank-large on 2026-10-08 (official
-        # unit-name scheme; the service/unit/scripts keep the frank-large
+        # Model id renamed from frank-dragoon on 2026-10-08 (official
+        # unit-name scheme; the service/unit/scripts keep the frank-dragoon
         # name — service names are unchanged).
         name: "frank-dragoon (dankclaw-gpu)"
         # Server-side compaction runs on the compaction.remoteEndpoint in
@@ -107,37 +107,45 @@ providers:
         compactionModel: llamacpp (dankclaw-3090)/frank-zealot
         # halogen-flash-server 0.16.2 (ghcr.io/peonist-ai/
         # halogen-flash-server@sha256:0c61bf84, digest-pinned) serving the
-        # Quat3rnion ABLITERATED Swift1.5 finetune of peonist's
-        # Qwen3.8-Flash-Next v2 checkpoint (qwen38-flash-next-v2-swift15-
-        # abliterated.hgn, ~62 GiB resident, + REQUIRED n-gram table +
-        # vision tower) on the Strix Halo iGPU: TWO slots at
-        # HALOGEN_CTX=262144 each with the MTP head ALWAYS ON
+        # Quat3rnion NATIVE v2 ABLITERATED conversion of orcarouter's pinned
+        # BF16 Qwen3.8-Flash-Next (qwen38-flash-next-v2-abliterated.hgn,
+        # ~62 GiB resident, + REQUIRED n-gram table + vision tower, both
+        # hardlinked from the swift15 dir) on the Strix Halo iGPU: TWO slots
+        # over ONE 401,408-position KV pool (= 2 x ctx, raised 2026-10-18) at
+        # HALOGEN_CTX=200704 (the ceiling lowered from the card's 262,144 on
+        # 2026-10-17; ~10.8 GiB reserved where 200,704 cost 5.5, 262,144 cost
+        # 7.2 and 524,288 measured 14.6), with the MTP head ALWAYS ON
         # (HALOGEN_SPEC_ADAPT=0 — the default adaptive policy halves
-        # decode). Card: 53-57 t/s greedy, ~1,640 t/s prefill at 34k, ~72%
-        # fewer thinking tokens than the censored 27B. ABLITERATED base
-        # (refusals should not return) and VISION ON (OpenAI image_url
+        # decode). Measured 2026-10-17: 50 t/s greedy decode, 1,604 t/s
+        # cold prefill at 59k, NIAH 18/18. ABLITERATED base (no refusals:
+        # 0/5 on the hostile census) and VISION ON (OpenAI image_url
         # parts; >=256 px images or it answers "video"). Runbook:
-        # serverAppConfigs-docs/LLM/frank-large-flash-server-v2-swift15.md.
-        # Per-SLOT window the harness may fill — deliberately BELOW the
-        # engine ceiling (HALOGEN_CTX=262144): the clamp below leaves a
-        # 45k gap that absorbs the estimate error.
-        contextWindow: 217088
+        # serverAppConfigs-docs/LLM/frank-dragoon-flash-server-v2-native.md.
+        # Window the harness may fill — deliberately BELOW the engine
+        # ceiling (HALOGEN_CTX=200704, lowered from 262,144 on 2026-10-17):
+        # the clamp below leaves the 45k gap that absorbs the estimate error.
+        # Both slots SHARE one 401,408-position KV pool — exactly two full
+        # windows — so TWO maximum-length conversations can be resident at the
+        # same time (at the old pool == ctx 200,704 only two of <=100k could).
+        # The per-request ceiling is unchanged, so this window is unchanged.
+        contextWindow: 155648
         # The clamp is max_tokens = min(maxTokens, contextWindow − est −
         # 4096); the chars/4 est UNDER-counts the rendered prompt by ~3 %
         # on code/agent traffic (measured 2026-10-07: est 152,938 ->
-        # server 157,388). Worst-case request = ctx − 4096 + 3 % est
-        # error ≈ 219k, inside the 262,144 engine ceiling with ~43k
-        # margin (robust to a 10 % under-count). Compaction fires at 85 %
-        # of the window = 184,524 est ≈ 190k measured (Frank's 180k
-        # target) and rides on frank-zealot: input ≈ threshold −
-        # keepRecent(20k) ≈ 170k, inside its 200,704 slot.
-        # HALOGEN_MAX_TOKENS_CAP in frank-large.env must stay >= this.
+        # server 157,388). Worst-case request = window − 4096 + 3 % est
+        # error ≈ 156.2k, inside the 200,704 engine ceiling with ~44k
+        # margin (robust to a ~25 % under-count). Compaction fires at 85 %
+        # of the window = 132,300 est ≈ 136k measured — the 180k target of
+        # the 262,144 era is unreachable now (the window itself is 155,648)
+        # — and rides on frank-zealot: input ≈ threshold − keepRecent(20k)
+        # ≈ 112k, inside its 200,704 slot.
+        # HALOGEN_MAX_TOKENS_CAP in frank-dragoon.env must stay >= this.
         maxTokens: 45056
         reasoning: true
         # VISION ON as of 2026-10-06: the v2 vision tower loads beside the
         # checkpoint (HALOGEN_VISION_TOWER; red-PNG probe verified on the
         # card). image_url content parts only — the legacy top-level
-        # images field is the frank-secondary trap, do not repeat it here.
+        # images field is the frank-zealot trap, do not repeat it here.
         input: [text, image]
         compat:
           # flash-server emits thinking as `reasoning_content` — same
@@ -155,7 +163,7 @@ providers:
             max: xhigh
           # The card's thinking recipe, on the wire. The flash-server
           # engine HAS a server-side sampler (HALOGEN_TEMPERATURE/TOP_P/
-          # TOP_K = 1.0 / 0.95 / 20 in frank-large.env) — these match it,
+          # TOP_K = 1.0 / 0.95 / 20 in frank-dragoon.env) — these match it,
           # so per-request values are consistent, not load-bearing.
           extraBody:
             temperature: 1.0
@@ -167,35 +175,47 @@ providers:
     auth: none
     models:
       - id: frank-zealot
-        # Model id renamed from frank-secondary on 2026-10-08 (official
+        # Model id renamed from frank-zealot on 2026-10-08 (official
         # unit-name scheme; the service/unit/scripts keep the
-        # frank-secondary name — service names are unchanged).
+        # frank-zealot name — service names are unchanged).
         name: "frank-zealot (dankclaw-3090)"
         # Server-side compaction routes to compaction.remoteEndpoint in
         # config.yml (192.168.0.143:7771 — this endpoint); the
         # compactionModel candidate keeps the wire model id valid there.
         compactionModel: llamacpp (dankclaw-3090)/frank-zealot
-        # Dankclaw's RTX 3090: llama.cpp (master, CUDA sm_86 build) serving
-        # HauhauCS/Qwen3.5-9B-Uncensored-HauhauCS-Aggressive at Q6_K
-        # (7,359,259,008 B) — qwen35 dense hybrid (32 layers: 8 full-attention
-        # + 24 Gated DeltaNet, 4 KV heads x 256 dim, native 262,144), TWO
-        # conversations (--ctx-size 401408 / --parallel 2) at Frank's 200k cap
-        # each. Exact KV: f16 K + f16 V. Vision via the publisher's OWN
-        # BF16 mmproj (921,704,448 B) — the model is natively multimodal.
-        # **MTP IS OFF** because the file carries no nextn/mtp tensors —
-        # SPEC_TYPE=none strips both spec flags. Priced from the MiMo Q6_K
-        # boots at this geometry (21,276/2,849 at ubatch 1024; 21,846/2,279 at
-        # 2048). 2026-10-07 (late) swap from Gemma4-26B-A4B Q4_K_M (1x200,704).
-        # Runbook:
-        # serverAppConfigs-docs/LLM/frank-secondary-llama-qwen35-9b-q6k-2x200704.md.
-        # contextWindow is the PER-SLOT window (Frank's cap): a longer
-        # prompt needs compaction.
-        contextWindow: 200704
+        # Dankclaw's RTX 3090: llama.cpp PrismML fork build (vendor 10777,
+        # commit e3de71522 — the ONLY loader for this payload's custom tensor
+        # type 142 + prism.hadamard.* metadata, and the minimum build that
+        # pairs MTP with a quantized V cache) serving
+        # prism-ml/Ternary-Bonsai-2-9B-gguf at PQ2_0 (2,675,410,720 B) WITH the
+        # MTP head in the file — qwen35 hybrid, 32 layers (8 full-attention +
+        # 24 Gated DeltaNet), 4 KV heads x 256 dim, native 262,144 — in THREE
+        # conversations (--ctx-size 786432 / --parallel 3), each slot at the
+        # model's TRAINED window.
+        # KV is q8_0 K + q8_0 V (17 KiB/token over the 8 full layers =
+        # 13,056 MiB pool, ledger-exact); f16 KV would bill 26,112 MiB and
+        # cannot boot this card. Boot ledger 17,876 used / 6,249 free. Vision
+        # through the publisher's Q8_0 mmproj (624,229,664 B — Q8_0 is the
+        # publisher's requirement for concurrent image requests), whose
+        # worst-case lands outside the engine's fit projection. **MTP OFF —
+        # measured, not assumed**: the draft context bills its own f16 KV over
+        # the WHOLE pool (3,072 MiB) and at ubatch 2048 that squeeze halves
+        # decode (167.2 → 109.2 t/s); SPEC_TYPE=draft-mtp + UBATCH_SIZE=1024 is
+        # the 213.5 t/s arm if the card ever has room. 2026-10-18 swap from
+        # Ternary-Bonsai-2-27B PQ2_0 at 2 × 200,704. Measured: 167.2 t/s
+        # single-stream greedy decode, 280.2 t/s across three concurrent
+        # streams, 2,877 t/s prefill at 106,418. Base-alignment: censored base
+        # lineage; unlike the 27B there is NO CRACK re-export matching this
+        # file. Runbook:
+        # serverAppConfigs-docs/LLM/frank-zealot-llama-prism-bonsai9-mtp.md.
+        # contextWindow is the PER-SLOT window (= the native 262,144): a longer
+        # prompt needs compaction — llama.cpp answers a 400, never truncates.
+        contextWindow: 262144
         # Prompt + budget must fit inside the slot window; thinking counts
         # against the budget. Compaction fires at 85 % of the window =
-        # 170,598 est; worst-case prompt + budget ≈ 170,598 × 1.03 +
-        # 16,384 ≈ 192,100 ≤ 200,704 — a longer prompt is a 400 naming
-        # the limit, never a truncation.
+        # 222,822 est; worst-case prompt + budget ≈ 222,822 × 1.03 +
+        # 16,384 ≈ 245,891 ≤ 262,144. This endpoint is the fleet's
+        # compaction hub: the widest window in the fleet.
         maxTokens: 16384
         reasoning: true
         input: [text, image]
@@ -204,29 +224,33 @@ providers:
           # `reasoning_content` — the same field the MiMo/Ornith eras used;
           # harness reads unchanged.
           reasoningContentField: "reasoning_content"
-          # The Qwen3.5 chat template has NO reasoning_effort variable (unlike
-          # the Qwen3.8-template endpoints, where an unmapped level is a hard
-          # Jinja raise). llama.cpp passes chat_template_kwargs through to
-          # Jinja, and Jinja ignores variables a template never reads — so the
-          # map is inert here, kept so role-level overrides resolve identically
-          # across the fleet.
+          # Bonsai-2's template DOES read reasoning_effort (the Qwen3.5-9B
+          # template did not — the map used to be inert here): it accepts ONLY
+          # xhigh (default), medium, low and calls raise_exception on anything
+          # else, so an unmapped harness level is a hard HTTP 500 on the role's
+          # first request, naming Jinja and not the model. The fold below is
+          # therefore load-bearing, identical to the frank-archon/dragoon law.
+          # llama.cpp passes chat_template_kwargs through to Jinja.
           reasoningEffortMap:
             minimal: low
             low: low
             medium: medium
             high: xhigh
             max: xhigh
-          # The Qwen3.5 publisher recipe (model card, thinking mode):
-          # temperature 0.6 / top_p 0.95 / top_k 20 / min_p 0. Non-thinking
-          # mode is 0.7/0.8/20/0 — the harness runs thinking mode.
+          # The payload's OWN sampling metadata (general.sampling.* in the
+          # GGUF): temperature 1.0 / top_p 0.95 / top_k 20 — thinking-mode
+          # recipe, the same trio the other Qwen3.8-family endpoints carry.
+          # (The old 0.6 was the Qwen3.5-9B publisher's card, not this file.)
           # extraBody is merged last and wins over the harness default.
           extraBody:
-            temperature: 0.6
+            temperature: 1.0
             top_p: 0.95
             top_k: 20
             min_p: 0.0
             presence_penalty: 0.0
             chat_template_kwargs:
               enable_thinking: true
-            # (no preserve_thinking: that kwarg belongs to the Qwen3.8
-            # templates; nothing in this template reads it)
+              # The template DOES read preserve_thinking: when false it drops
+              # prior assistants' <think> blocks from history. Keep them —
+              # multi-turn tool traffic on this endpoint is the majority.
+              preserve_thinking: true