live: heartbeat sync 2026-10-18T16:47:59-07:00 (Refs #11)

agent/memories/sharpshooter/wt-20261010-164545-6d1fc1f-1mfj0dsoc238q/architecture.md

@@ -0,0 +1,9 @@
+## LLM Runtime KV Cache (frank-archon)
+
+**KV cache is unquantized (fp8 native)** — radiance engine 1.0.13 on frank-primary (frank-archon) runs with fp8 KV cache pool of 3.46 GiB. A prior quantized state was reversed; the unquantized configuration is the current production state.
+
+**Per-slot context ceiling** — `frank-archon` and `frank-dragoon` share a 262,144 position per-slot ceiling (`--ctx-size 401408 --parallel 2`), distinct from the 27B-era 200,192 limit. The harness clamps `max_tokens` to `min(45056, contextWindow − est − 4096)` and the 400-zone law under-counts; the 400 constant is the floor.
+
+**MTP head placement** — `frank-dragoon` halogen-flash-server 0.16.2 embeds the MTP head inside the checkpoint (`HALOGEN_MTP_HEAD` unset; no sidecar fetched). `frank-zealot` llama.cpp has no MTP tensors in the Q6_K GGUF file.
+
+**Per-slot vs pool context** — `frank-zealot` exposes `200704` as the per-slot window (configured `--ctx-size 401408 --parallel 2`), not the 262,144 native model size or the 409,600 aggregate pool. The harness must compact before llama.cpp accepts longer prompts.

agent/memories/sharpshooter/wt-20261010-164545-6d1fc1f-1mfj0dsoc238q/product.md

@@ -0,0 +1,15 @@
+## Service Naming Convention (ACTIVE)
+
+**Three-tier fleet naming** — the LLM fleet uses a consistent three-part alias scheme across all user-facing interfaces (models.yml, config.yml, homepage cards, OpenWebUI grants, Hermes routing):
+
+- `frank-archon` — primary inference endpoint (radiance runtime, `frank-primary` host)
+- `frank-dragoon` — large model endpoint (halogen-flash-server, `frank-large` host)
+- `frank-zealot` — secondary model endpoint (llama.cpp, `frank-secondary` host)
+
+**Internal/external bifurcation** — infrastructure components (systemd units, compose files, scripts, env files) use the internal names (`frank-primary`, `frank-large`, `frank-secondary`); the wire-facing and operator-facing interfaces use the alias names (`frank-archon`, `frank-dragoon`, `frank-zealot`).
+
+**Resolution rules** — the default model resolves as a `defaultProvider` + `defaultModel` pair (the provider must serve the model); a subagent override uses a bare model `id`, which resolves only when unique across providers. Two ports on one host need two providers — the file cannot express "same host, different port".
+
+**When renaming** — update the provider block, model `id`, `contextWindow`, `maxTokens`, `input` in `models.yml`, then fix every `modelRoles` entry that named the old endpoint — a role pointing at a provider the file does not define fails at call time.
+
+**Model id alignment** — as of 2026-10-09 the StarCraft unit names are the official model ids and also the ids on the wire. Verify a rename on `/v1/models`, never on a chat response: radiance echoes the **requested** model name, so a 200 proves nothing about the served id (llama.cpp and halogen ignore it).

agent/memories/sharpshooter/wt-20261010-164545-6d1fc1f-1mfj0dsoc238q/state.json

@@ -0,0 +1,10 @@
+{
+	"v": 1,
+	"lastConsolidatedAt": 1791684585241,
+	"lastResult": {
+		"at": 1791676692246,
+		"sessions": 1,
+		"deltas": 3,
+		"model": "frank-zealot"
+	}
+}

agent/memories/sharpshooter/wt-20261010-164545-6d1fc1f-1mfj0dsoc238q/style.md

@@ -0,0 +1,7 @@
+## Naming & Terminology
+
+**Service aliases** — the three LLM endpoints are referenced as `frank-archon`, `frank-dragoon`, `frank-zealot` in all user-facing documentation, configuration files, and API responses.
+
+**Infrastructure references** — systemd units, Podman compose files, shell scripts, and environment files use the internal names `frank-primary`, `frank-large`, `frank-secondary`.
+
+**Cross-reference clarity** — when a document spans both layers, explicitly note the mapping (e.g., "`frank-archon` (radiance, `frank-primary`)" or "`frank-dragoon` (halogen, `frank-large`)").