lobes-cli 0.89.0__tar.gz → 0.89.2__tar.gz

lobes_cli-0.89.2/.agex/data/pr/323/await.json

@@ -0,0 +1,13 @@
+{
+  "schema": 1,
+  "state": "done",
+  "pr": 323,
+  "exit_code": 0,
+  "outcome": "clean",
+  "started_at": "2026-10-18T11:10:48.671094+00:00",
+  "finished_at": "2026-10-18T11:11:56.545796+00:00",
+  "briefing": "# `agex pr read` \u2014 PR #323\n\n- State: **OPEN** (fix: qwen3_coder_thinking passes reasoning=False on vLLM >= 0.30 (strict+thinking repeat) \u2014 #310 d5)\n- Branch: `fix/plugin-reasoning-gate` \u2192 `main`\n- URL: https://github.com/agentculture/lobes-cli/pull/323\n\n## CI checks\n\n| Check | Status | Conclusion |\n|---|---|---|\n| test | in_progress | \u2014 |\n| test | in_progress | \u2014 |\n| lint | completed | success |\n| secrets-scan | completed | success |\n| site-build | completed | success |\n| version-check | completed | success |\n| GitGuardian Security Checks | completed | success |\n\n\n## SonarCloud\n\n_Project not on SonarCloud (skipped)._\n\n## Comments\n\n- **[top-level]** `OriNachum`: /agentic_review\n- **[top-level]** `qodo-code-review[bot]`: <!-- qodo:billing-blocked -->\n\n**\u24d8 Qodo reviews are paused because the subscription is no longer active.** Ask your workspace admin to reactivate the subscription to resume reviews. [Manage...\n- **[top-level]** `qodo-code-review[bot]`: <!-- qodo:billing-blocked -->\n\n**\u24d8 Qodo reviews are paused because the subscription is no longer active.** Ask your workspace admin to reactivate the subscription to resume reviews. [Manage...\n\n\n## Qodo review\n\n_No Qodo review found._\n\n## Readiness\n\nAll required reviewers have posted feedback.\n\n---\n**Next step:** PR #323 is ready \u2014 wait for human merge.\n",
+  "gate_status": null,
+  "threads_unresolved": 0,
+  "ci_state": "ok"
+}

{lobes_cli-0.89.0 → lobes_cli-0.89.2}/.agex/data/pr/events.jsonl

@@ -239,3 +239,9 @@
 {"ts":"2026-10-07T00:01:53.296199+00:00","type":"pr_webhook_posted","pr":293,"event":"pr_opened"}
 {"ts":"2026-10-07T00:02:56.775080+00:00","type":"readiness_arrived","pr":293,"waited_secs":60}
 {"ts":"2026-10-07T00:03:01.419898+00:00","type":"pr_read","pr":293,"comment_count":3,"threads_unresolved":0,"ci_state":"ok"}
+{"ts":"2026-10-18T11:10:46.372078+00:00","type":"pr_opened","pr":323,"title":"fix: qwen3_coder_thinking passes reasoning=False on vLLM >= 0.30 (strict+thinking repeat) \u2014 #310 d5"}
+{"ts":"2026-10-18T11:10:47.900887+00:00","type":"pr_review_triggered","pr":323,"command":"/agentic_review"}
+{"ts":"2026-10-18T11:10:48.656513+00:00","type":"pr_webhook_posted","pr":323,"event":"pr_opened"}
+{"ts":"2026-10-18T11:10:48.671713+00:00","type":"pr_await_detach_spawned","pr":323,"max_wait":1800}
+{"ts":"2026-10-18T11:11:52.039761+00:00","type":"readiness_arrived","pr":323,"waited_secs":60}
+{"ts":"2026-10-18T11:11:56.546218+00:00","type":"pr_await_detach_done","pr":323,"outcome":"clean","exit_code":0}

{lobes_cli-0.89.0 → lobes_cli-0.89.2}/.devague/frames/vllm-0-31-0-fleet.json

@@ -4,7 +4,7 @@
   "schema_version": 6,
   "status": "exported",
   "created": "2026-10-17T16:54:43Z",
-  "updated": "2026-10-17T18:55:03Z",
+  "updated": "2026-10-17T19:47:33Z",
   "claims": [
     {
       "id": "c1",
@@ -1381,6 +1381,28 @@
       ],
       "origin": "llm",
       "status": "proposed"
+    },
+    {
+      "id": "l6",
+      "code": "control-absent",
+      "what": "t6's dry run read the LIVE Spark: doctor's health_reachable and gateway_version_match did a GET /health on localhost:8001 because the copy's port pointed there; read-only, but those two findings describe the live box, not the copy",
+      "skipped_check": "point the copy's gateway port at a closed port before running doctor",
+      "refs": [
+        "t6"
+      ],
+      "origin": "llm",
+      "status": "proposed"
+    },
+    {
+      "id": "l7",
+      "code": "assumption-for-measurement",
+      "what": "t6's runbook claims a container that docker restarts on its own picks up the refreshed wrapper through its file bind mount ('prepare now, recreate later'); reasoned from docker behaviour, not tested; the merged compose chain was checked by YAML parse + simulation only, not docker compose config",
+      "skipped_check": "docker compose config on the merged chain; a real restart",
+      "refs": [
+        "t6"
+      ],
+      "origin": "llm",
+      "status": "proposed"
     }
   ],
   "obligations": []

{lobes_cli-0.89.0 → lobes_cli-0.89.2}/CHANGELOG.md

@@ -4,6 +4,19 @@ All notable changes to this project are documented here. The format is based on
 [Keep a Changelog](https://keepachangelog.com/en/1.1.0/), and this project
 adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [0.89.2] - 2026-10-18
+
+### Fixed
+
+- `qwen3_coder_thinking` plugin: on vLLM >= 0.30.0, which starts the structured-output grammar only after reasoning ends, the plugin now passes vLLM's own `reasoning` flag through (`False` on every engine measured) instead of deriving it per request, and logs which mode it chose at import. The thinking-prefixed grammar it built before masked EOS there, so a strict tool call with thinking repeated until `max_tokens` (measured on 0.30.0, 0.31.0 and 0.32.0rc1). Older engines keep the per-request derivation; an unreadable `vllm.__version__` refuses to boot (#310 deviation d5).
+- `lobes assess --strict-tools`: a strict leg that hits `max_tokens` (`finish_reason=length`) is no longer counted as a clean call, so the matrix now catches the repeat-until-`max_tokens` failure.
+
+## [0.89.1] - 2026-10-17
+
+### Added
+
+- Runbook docs/runbooks/lane-safety-r1-rollout.md: move an existing hand-kept ~/.lobes from the 0.86 scaffold to the 0.89.0 (lane-safety + #310 R1) scaffold without re-scaffolding — override-only compose merge (with the compose merge-semantics table), doctor --refresh-logwrap / --refresh-plugin / --fix, verification and rollback; dry-run record docs/evidence/2026-10-17-dryrun-r1-rollout-spark-copy.txt (#310 t6)
+
 ## [0.89.0] - 2026-10-17
 
 ### Added

{lobes_cli-0.89.0 → lobes_cli-0.89.2}/CLAUDE.md

@@ -145,7 +145,13 @@ tool-parser plugin (`lobes/vllm_plugins/`, loaded via vLLM's own
 `--tool-parser-plugin` file-path surface, cortex/main lane only — mirrors
 the `preserve_thinking` #93 scoping) — it derives the grammar's `reasoning`
 flag from the request's own `enable_thinking` instead of the hardcoded
-`False`. A separate, default-off gateway knob (`GATEWAY_FORCE_STRICT_TOOLS`)
+`False`. **That derivation is engine-dependent (#310 deviation d5,
+2026-10-18):** vLLM 0.30.0 and later start the grammar only after reasoning
+ends, so a thinking-prefixed grammar masks EOS and a strict call repeats until
+`max_tokens`; on those engines the plugin passes vLLM's own `reasoning` flag
+through (`False` on every engine measured) (measured on 0.30.0, 0.31.0 and 0.32.0rc1;
+`docs/evidence/2026-10-18-accept-vllm-031-spark-cortex.txt`). The boundary is
+the measured one, chosen by the plugin from `vllm.__version__`. A separate, default-off gateway knob (`GATEWAY_FORCE_STRICT_TOOLS`)
 opts existing callers into strict schemas without a client-side change, with
 a retry-without-strict fallback on a grammar-compile failure. See
 `docs/qwen3.6-27b-text-nvfp4-mtp.md`, `docs/openai-api.md`, and

{lobes_cli-0.89.0 → lobes_cli-0.89.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: lobes-cli
-Version: 0.89.0
+Version: 0.89.2
 Summary: lobes — run, assess, and switch the local vLLM model.
 Project-URL: Homepage, https://github.com/agentculture/lobes-cli
 Project-URL: Issues, https://github.com/agentculture/lobes-cli/issues

lobes_cli-0.89.2/docs/evidence/2026-10-17-dryrun-r1-rollout-spark-copy.txt

@@ -0,0 +1,264 @@
+DRY RUN — R1 (lobes-cli 0.89.0) lane-safety scaffold onto a hand-kept ~/.lobes
+==============================================================================
+
+Date:      2026-10-17
+Procedure: docs/runbooks/lane-safety-r1-rollout.md (steps 2-8; step 9 not run)
+Plan:      #310 vllm-0-31-0-fleet, task t6
+Target:    a sanitised COPY of the DGX Spark's ~/.lobes, in session scratch
+           space. Secret values in the copy's .env were replaced before the
+           run. No live box was touched: no ssh, no `docker compose`, no
+           container created, recreated, stopped or updated.
+CLI:       lobes 0.89.0, run from source (`uv run lobes ... --compose-dir <copy>`)
+           on the Spark host, so card detection read the real GB10.
+
+This record quotes structure only: finding ids, compose lines (which hold
+${VAR} references, not values) and .env key NAMES. No .env value is quoted.
+
+--- The copy's compose chain (`lobes fleet files --json`) -----------------------
+
+  -f docker-compose.yml -f docker-compose.audio.yml
+  -f docker-compose.shape.yml -f docker-compose.override.yml
+
+  docker-compose.yml is a full copy of an older packaged template plus hand
+  edits; the shape file parks three services in an inert profile; the override
+  carries the Hebrew audio deltas, gateway passthroughs and a backfilled comfyui.
+
+--- Step 3: doctor BEFORE (finding ids) ------------------------------------------
+
+  [ok]   docker_available, compose_present, env_coherence
+  [warn] scaffold_files            (cf-tunnel.env.example missing)
+  [warn] profile_staleness         (20 required keys missing from .env)
+  [warn] gateway_passthrough       (3 keys; not lane safety, out of scope here)
+  [warn] logwrap_stale             (deployed wrapper differs from the package)
+  [warn] compile_cache_unmounted   (8 vLLM services: vllm-embed-deep, vllm-middle,
+                                    vllm-minor, vllm-multimodal-coder, vllm-muse,
+                                    vllm-primary, vllm-rerank, vllm-worker)
+  [warn] drop_caches_grant_missing (host finding: this Spark has no passwordless
+                                    page-cache clear)
+  [ok]   watcher_inactive          (not installed; optional)
+  [warn] plugin_stale              (deployed tool-parser plugin differs)
+  [ok]   health_reachable
+  [FAIL] gateway_version_match     (deployed gateway 0.86.0.dev642 vs CLI 0.89.0)
+
+  Note: health_reachable and gateway_version_match read the LIVE Spark
+  gateway's /health on localhost (the copy's port), read-only. They describe
+  the live box, not the copy.
+
+--- Step 3: per-service lane-safety state, BEFORE (parsed, not compose config) ---
+
+  Services the shape file parks (vllm-embed, vllm-hand, vllm-multimodal) are
+  omitted. "wrapped" = entrypoint is mg-logwrap; caches = /cache/vllm and
+  /cache/flashinfer binds.
+
+  service                 restart         wrapped  caches  start_period
+  vllm-primary            unless-stopped  yes      no      600s
+  vllm-rerank             unless-stopped  yes      no      600s
+  vllm-embed-deep (prof)  unless-stopped  yes      no      600s
+  vllm-minor (prof)       unless-stopped  yes      no      600s
+  vllm-middle (prof)      unless-stopped  yes      no      600s
+  vllm-multimodal-coder   unless-stopped  yes      no      600s
+  vllm-muse (prof)        unless-stopped  yes      no      600s
+  vllm-worker (prof)      unless-stopped  yes      no      600s
+  gateway                 unless-stopped  no       -       20s
+  stt                     unless-stopped  no       -       600s
+  chatterbox              unless-stopped  no       -       300s
+  realtime                unless-stopped  no       -       30s
+  comfyui (innereye)      unless-stopped  yes      -       90s
+  bluetts (override only) unless-stopped  no       -       120s
+
+  Restart policy: every service already has `restart: unless-stopped`; R1
+  adds none. bluetts is in no packaged compose file, so it stays unwrapped.
+
+--- Step 2: dated backups ---------------------------------------------------------
+
+  .env, docker-compose.yml, docker-compose.audio.yml, docker-compose.shape.yml,
+  docker-compose.override.yml, mg-logwrap.sh, qwen3_thinking_tool_parser.py
+  copied to <name>.bak-<stamp>.
+
+--- Step 4: docker-compose.override.yml diff (the only compose file changed) -----
+
+  Totals: 5 hunks, 139 lines added, 2 lines changed, 0 removed.
+  docker-compose.yml, docker-compose.audio.yml and docker-compose.shape.yml:
+  byte-identical to their backups (cmp).
+
+  Hunk 1, existing `stt:` block (merged into it, not a second key):
+  +      - MG_LOG_DIR=/logs/model-gear
+  +      - MG_LOG_NAME=stt
+  +    volumes:
+  +      - ${MODEL_GEAR_LOG_DIR:-./logs}:/logs/model-gear
+  +      - ./mg-logwrap.sh:/usr/local/bin/mg-logwrap:ro
+  +    entrypoint: ["bash", "/usr/local/bin/mg-logwrap"]
+  +    command: ["/opt/nvidia/nvidia_entrypoint.sh", "python3", "/app/listen_server_whisper.py"]
+  +    healthcheck:
+  +      start_period: 780s
+
+  Hunk 2, existing `chatterbox:` block: the same log env and binds, then
+  +    entrypoint: ["bash", "/usr/local/bin/mg-logwrap"]
+  +    command: ["python3.12", "-m", "lobes.realtime.chatterbox_multilingual_server"]
+       healthcheck:
+  -      start_period: 300s
+  +      start_period: 480s
+
+  Hunk 3, existing `realtime:` block (no env_file there, so the boot knobs are
+  passed explicitly):
+  +      - MG_LOG_DIR=/logs/model-gear
+  +      - MG_LOG_NAME=realtime
+  +      - LOBES_BOOT_DELAY_S=${LOBES_BOOT_DELAY_S:-120}
+  +      - LOBES_BOOT_ORDER=${LOBES_BOOT_ORDER:-}
+  +      - LOBES_BOOT_STAGGER_S=${LOBES_BOOT_STAGGER_S:-20}
+  +    volumes: (the same two binds)
+  +    entrypoint: ["bash", "/usr/local/bin/mg-logwrap"]
+  +    command: ["python", "-m", "lobes.realtime"]
+  +    healthcheck:
+  +      start_period: 210s
+
+  Hunk 4, existing `gateway:` block:
+  +      - MG_LOG_DIR=/logs/model-gear
+  +      - MG_LOG_NAME=gateway
+  +      - LOBES_BOOT_DELAY_S=${LOBES_BOOT_DELAY_S:-120}
+  +      - LOBES_BOOT_ORDER=${LOBES_BOOT_ORDER:-}
+  +      - LOBES_BOOT_STAGGER_S=${LOBES_BOOT_STAGGER_S:-20}
+  +      - LOBES_THERMAL_ZONES=${LOBES_THERMAL_ZONES:-}
+  +      - LOBES_THERMAL_SHED_C=${LOBES_THERMAL_SHED_C:-}
+  +      - LOBES_THERMAL_RESUME_C=${LOBES_THERMAL_RESUME_C:-}
+       volumes:
+         - ${LOBES_MESH_DIR:-./mesh}:/home/gateway/mesh
+  +      - ${MODEL_GEAR_LOG_DIR:-./logs}:/logs/model-gear
+  +      - ./mg-logwrap.sh:/usr/local/bin/mg-logwrap:ro
+  +    entrypoint: ["bash", "/usr/local/bin/mg-logwrap"]
+  +    command: ["python", "-m", "lobes.gateway"]
+  +    healthcheck:
+  +      start_period: 200s
+
+  Hunk 5, existing `comfyui:` block, then 8 new service blocks:
+  -      start_period: 90s
+  +      start_period: 270s
+  +  vllm-primary:
+  +    environment:
+  +      - VLLM_CACHE_ROOT=${LOBES_VLLM_CACHE_ROOT:-/cache/vllm}
+  +      - FLASHINFER_WORKSPACE_BASE=${LOBES_FLASHINFER_WORKSPACE_BASE:-/cache/flashinfer}
+  +    volumes:
+  +      - ${LOBES_COMPILE_CACHE_DIR:-./compile-cache}/primary:/cache/vllm
+  +      - ${LOBES_COMPILE_CACHE_DIR:-./compile-cache}/primary-flashinfer:/cache/flashinfer
+  +    healthcheck:
+  +      start_period: 780s
+  (the same block for vllm-rerank, vllm-embed-deep, vllm-minor, vllm-middle,
+   vllm-multimodal-coder, vllm-muse, vllm-worker; lane names rerank,
+   embed-deep, minor, middle, multimodal-coder, muse, worker)
+
+  The image commands were respelled from the copy's own Dockerfiles:
+  Dockerfile.gateway and Dockerfile.realtime (ENTRYPOINT, inherited by
+  Dockerfile.realtime.local), Dockerfile.chatterbox-ml.local (ENTRYPOINT) and
+  Dockerfile.whisper-stt (CMD, identical to the packaged file). The stt
+  chain's /opt/nvidia/nvidia_entrypoint.sh is the packaged audio-he overlay's
+  spelling; the base image's ENTRYPOINT was NOT inspected (no docker call).
+
+--- Step 4: merge check (YAML parse with a duplicate-key check + a merge -------
+--- simulation of environment-by-key, volumes-by-target, healthcheck-by-key) ----
+
+  No duplicate keys. Effective state AFTER, same services as above:
+
+  service                 wrapped  logs  caches  VLLM_CACHE_ROOT  start_period
+  vllm-primary            yes      yes   yes     yes              780s
+  vllm-rerank             yes      yes   yes     yes              780s
+  vllm-embed-deep         yes      yes   yes     yes              780s
+  vllm-minor              yes      yes   yes     yes              780s
+  vllm-middle             yes      yes   yes     yes              780s
+  vllm-multimodal-coder   yes      yes   yes     yes              780s
+  vllm-muse               yes      yes   yes     yes              780s
+  vllm-worker             yes      yes   yes     yes              780s
+  gateway                 yes      yes   -       -                200s
+  stt                     yes      yes   -       -                780s
+  chatterbox              yes      yes   -       -                480s
+  realtime                yes      yes   -       -                210s
+  comfyui                 yes      yes   -       -                270s
+  bluetts                 no       no    -       -                120s
+
+--- Step 5: lobes doctor --refresh-logwrap [--apply] ------------------------------
+
+  plan:    would copy mg-logwrap.sh -> mg-logwrap.sh.bak
+           would write the packaged mg-logwrap.sh
+  applied: wrote packaged mg-logwrap.sh
+
+--- Step 6: lobes doctor --refresh-plugin [--apply] -------------------------------
+
+  plan:    deployed sha256 22a030552d03..., packaged sha256 36a95d103704...
+           would copy qwen3_thinking_tool_parser.py -> qwen3_thinking_tool_parser.py.bak
+           would write the packaged qwen3_thinking_tool_parser.py
+  applied: wrote packaged qwen3_thinking_tool_parser.py
+
+--- Step 7: lobes doctor --fix [--apply] (run AFTER the override) -----------------
+
+  Files written:  cf-tunnel.env.example
+  Dirs created:   compile-cache/ plus 16 lane dirs: embed-deep, middle, minor,
+                  multimodal-coder, muse, primary, rerank, worker, each with its
+                  -flashinfer sibling
+  .env keys appended (20, names only; no existing line changed):
+    lane safety (3):  LOBES_THERMAL_ZONES, LOBES_THERMAL_SHED_C,
+                      LOBES_THERMAL_RESUME_C
+    audio (17):       AUDIO_URL, BARGE_IN_MODEL, BARGE_IN_WINDOW_MS,
+                      CHATTERBOX_PORT, DEFAULT_AEC_MODE, DEFAULT_SYSTEM_PROMPT,
+                      DEFAULT_TURN_DETECTION, DEFAULT_VOICE, PARAKEET_MODEL,
+                      PARAKEET_PORT, REALTIME_PORT, TTS_CONCURRENCY, TTS_SPEED,
+                      TTS_VOICE_CONCURRENCY, VAD_MAX_TURN_MS,
+                      VAD_PREFIX_PADDING_MS, VAD_THRESHOLD
+  Each appended value was compared with the `${KEY:-default}` the copy's
+  compose files use for that key:
+    - 16 audio keys: equal to the compose default, so no effective change.
+    - AUDIO_URL: differs from the base file's empty default, but equals the
+      audio overlay's default, which wins later in the chain. No change.
+    - the 3 thermal keys: differ from the empty compose default. With the
+      hunk-4 passthrough, the recreated gateway would arm thermal shedding.
+      The runbook says so in step 7.
+  No LOBES_BOOT_*, LOBES_START_*, LOBES_MEM_GATE* or cache key was appended:
+  they are not required keys, and the wrapper and compose carry their defaults.
+
+  The --fix plan run BEFORE step 4 listed only compile-cache/ itself, because
+  the per-lane directories come from the binds the compose files declare. Order
+  matters: override first, then --fix --apply.
+
+--- doctor AFTER (finding ids) ----------------------------------------------------
+
+  [ok]   docker_available, compose_present, env_coherence
+  [ok]   scaffold_files            (all 15 present)
+  [ok]   profile_staleness         (every required key present)
+  [warn] gateway_passthrough       (unchanged, same 3 keys; out of scope)
+  [ok]   logwrap_stale             (matches the package)
+  [ok]   compile_cache_unmounted   (all 8 vLLM services mount both caches)
+  [ok]   compile_cache_size        (0 bytes)
+  [warn] drop_caches_grant_missing (host; step 8 needs the operator)
+  [ok]   watcher_inactive          (not installed; optional)
+  [ok]   plugin_stale              (matches the package)
+  [ok]   health_reachable
+  [FAIL] gateway_version_match     (the LIVE gateway; clears only after step 9)
+
+  lane_state_writable is absent: the copy has no logs/ directory.
+
+--- Step 9 (not run): what it would do ---------------------------------------------
+
+  `lobes up gateway --build` (dry run) on the copy:
+    would run: docker compose -f docker-compose.yml -f docker-compose.audio.yml
+    -f docker-compose.shape.yml -f docker-compose.override.yml
+    up -d --no-deps --build gateway; first: clear the page cache (when granted).
+  The audio overlay is in the gateway's chain.
+  `lobes doctor --repin-version` (dry run) planned to set MODEL_GEAR_VERSION=0.89.0.
+
+--- Rollback rehearsal --------------------------------------------------------------
+
+  On a second copy of the post-procedure tree: restored the 7 dated backups,
+  removed mg-logwrap.sh.bak, qwen3_thinking_tool_parser.py.bak,
+  cf-tunnel.env.example, compile-cache/ and the dated backups themselves.
+  `diff -r` against the pristine copy: identical.
+
+--- Not dry-run here ------------------------------------------------------------------
+
+  - `docker compose config` on the merged chain (forbidden for this task); the
+    merge was checked by parsing and simulation only.
+  - The respelled image commands (no `docker image inspect`).
+  - Any recreate, the boot hold, the start counter, the memory gate, the cache
+    write and the gateway's thermal shed on a live container.
+  - The machine-CLI grant and the watcher install (host changes).
+  - The claim that a self-restarted (not recreated) container picks up the new
+    wrapper through its file bind mount: reasoned, not tested.
+  - Thor, Orin and spark2 specifics (other overrides, docker-compose.gpu.yml,
+    no repo checkout on the box).

lobes_cli-0.89.2/docs/evidence/2026-10-18-accept-vllm-031-spark-cortex.txt

@@ -0,0 +1,71 @@
+ACCEPTANCE — DGX Spark cortex on vLLM 0.31.0 (v0.31.0-ubuntu2404), with the d5 plugin fix
+Issue #310 · plan vllm-0-31-0-fleet task t12 · 2026-10-18 · operator: lobes (Claude)
+Continues docs/evidence/2026-10-17-accept-vllm-031-spark-cortex.txt (partial), whose three findings are closed:
+  (1) governor GPU loss -> explicit GPU devices (d2), governor re-attached and verified below;
+  (2) strict+thinking regression -> ROOT CAUSE is the lobes plugin, not the engine (d5), fixed and verified below;
+  (3) hot acpitz zones -> thermal shed raised to 98/94 C (c55), active on the re-imaged 0.89.1 gateway below.
+Status: ACCEPTED on the official pin. Deviations d3 (split pin) and d4 (eugr exception) are withdrawn (d6).
+
+STRICT + THINKING — ROOT CAUSE (d5)
+  Request: tools=[finish(summary: string), strict: true], tool_choice auto, temperature 0, max_tokens 2048,
+  thinking on (template default). Reproduced on three engines, 3/3 runs each, through cortex-spark:
+    v0.30.0-ubuntu2404 and v0.31.0-ubuntu2404 (2026-10-17/18, partial transcript): ~78 calls, finish=length.
+    eugr GB10 nightly (vLLM 0.32.0rc1.dev301, ghcr.io/spark-arena/dgx-vllm-eugr-nightly@sha256:cc47adc1...):
+      run 1: calls=78 finish=length completion_tokens=2048 first_args={"summary": "hello"}
+      run 2: calls=78 finish=length completion_tokens=2048
+      run 3: calls=77 finish=length completion_tokens=2048
+    Same image, DSpark speculative config removed (isolation): 78 / 78 / 78 calls, finish=length.
+      -> DSpark ruled out.
+  Offline (CPU, xgrammar matcher, Qwen3.8 tokenizer), the plugin's thinking-prefixed grammar
+  (tag "" ... "</think>", then "\n\n", then triggered_tags <tool_call>):
+    fed the full text (thinking + </think> + one call): EOS allowed after the call.
+    fed only the text AFTER </think> (one call):         EOS NOT allowed; another <tool_call> allowed.
+    Same on all four combinations of {nightly grammar, new grammar} x {nightly xgrammar, eugr xgrammar}.
+  Reading: vLLM >= 0.30 starts the structured-output grammar only after reasoning ends, so the matcher sits
+  inside the plugin's thinking tag, accepts each tool call as thinking text, and masks EOS until max_tokens.
+  (The 2026-10-18 handoff's "grammar identical across engines" was wrong: the new engines add
+  excludes [<think>, </think>] to the thinking block's any_text. It does not change the result; lapse l11.)
+  Live test of the fix: the deployed plugin copy hand-patched to pass reasoning=False (backup
+  qwen3_thinking_tool_parser.py.bak-*-reasonfalse), eugr image:
+    DSpark off: 1 / 1 / 1 calls, finish=tool_calls, 45 tokens each; reasoning kept (72 chars)
+    DSpark on:  1 / 1 / 1 calls, finish=tool_calls, 65 tokens each
+    enforcement control, summary maxLength=3, prompt demands a 43-char string:
+      strict=True  -> {"summary": "the"}                                          (grammar enforced)
+      strict=False -> {"summary": "the quick brown fox jumps over the lazy dog"}  (unconstrained)
+    (an enum control did not discriminate: the model obeyed the enum with and without strict)
+
+ENGINE COMPARISON (cortex, DSpark n=7, 262144 window, util 0.58; lobes benchmark 1000 in / 1000 out)
+  eugr 0.32.0rc1.dev301:  decode 17.2 / 17.2 tok/s, prefill 887 tok in 1.12 s   (1 run;  hottest zone 87.1 C)
+  official 0.31.0:        decode 15.9 / 15.9 tok/s, prefill 887 tok in 0.83 s   (2 runs, identical; 85.2 C)
+  spark2 nightly baseline (t11): 17.0 / 16.9 tok/s.
+  0.31.0 = ~92% of eugr, ~94% of the nightly baseline. The 2026-10-17 0.31.0 run measured 14.0 / 14.4; that does
+  not reproduce today and the difference is not isolated (that run's zones idled near 88 C; today's peaked 85 C).
+  Operator decision (d6): official 0.31.0 on the Sparks too; eugr stays a measured, unadopted candidate.
+
+ACCEPTANCE ON 0.31.0 (VLLM_NIGHTLY_IMAGE=vllm/vllm-openai@sha256:3fb023cc..., plugin patched per d5)
+  boot: `lobes up cortex --apply` 04:2xZ, healthy 04:25:32Z, engine (v0.31.0), KV cache 754,621 tokens.
+  strict+thinking replay: 1 / 1 / 1 calls, finish=tool_calls, 65 tokens each.
+  strict, thinking off: 1 call, 26 tokens.  non-strict + thinking: 1 call, reasoning 204 chars.
+  maxLength=3 control: strict -> "the"; non-strict -> the full 43-char string.
+  lobes assess --tools: 17*23=391 PASS, train 145 min PASS, reasoning field present, tool_choice:auto PASS.
+  lobes assess --strict-tools: strict_thinking PASS, strict_nothink PASS, matrix PASS.
+  lobes assess --preserve-thinking: +47 prompt tokens -> preserve_thinking LIVE.
+  vision colour controls (solid PNGs, 64 and 224 px, red/blue/green), gateway cortex-spark and lane direct: 12/12.
+
+GOVERNOR (d2)
+  vllm-primary devices: /dev/nvidia0 /dev/nvidiactl /dev/nvidia-modeset /dev/nvidia-uvm /dev/nvidia-uvm-tools.
+  `lobes governor run --container model-gear-vllm-primary --apply --bands "92:12,95:8,98:4" --pause-c 101
+   --resume-c 95` (lobes 0.89.1): first tick 04:53:48Z cpus=16 tz=62.4C; container NanoCpus=16000000000 after the
+  docker update; a fresh `nvidia-smi` exec inside the lane still reports NVIDIA GB10; /health 200.
+  It runs as a session background process, not a durable unit (follow-up).
+
+GATEWAY
+  `lobes doctor --repin-version --apply`: MODEL_GEAR_VERSION 0.86.0.dev642 -> 0.89.1.
+  `lobes up gateway --build --apply`: chain kept docker-compose.audio.yml; healthy; lobes-cli 0.89.1 in-container;
+  LOBES_THERMAL_SHED_C=98, LOBES_THERMAL_RESUME_C=94; AUDIO_URL set; /capabilities cortex ready=true.
+  First mesh reannounce reported "Connection reset by peer"; afterwards cortex-spark, cortex-spark2 and plain
+  cortex all answered 200 through this gateway.
+
+END STATE (04:5xZ): Spark cortex on official v0.31.0-ubuntu2404 with DSpark, explicit GPU devices, governor
+  attached, hand-patched plugin (reasoning=False; the packaged engine-aware fix lands in R2, t14), gateway 0.89.1
+  with the 98/94 C shed. innereye down by design. spark2 untouched (nightly) until t13.

{lobes_cli-0.89.0 → lobes_cli-0.89.2}/docs/operating-a-deployment.md

@@ -124,6 +124,13 @@ restores the earlier behaviour (table at the end of this section). The Thor is
 code-only during this delivery, so every Thor line here is
 **DECLARED/UNVALIDATED (#108)**.
 
+To bring an existing box (a 0.86 scaffold, often hand-edited) up to these
+protections without re-scaffolding it, follow
+[`runbooks/lane-safety-r1-rollout.md`](runbooks/lane-safety-r1-rollout.md). It
+says which piece goes in `docker-compose.override.yml`, which `lobes doctor`
+lane writes which file, how to prepare now and recreate later, and how to roll
+back.
+
 **Boot hold (#308).** After a host reboot, dockerd restarts containers by
 restart policy and ignores compose `depends_on` (assumption c32, unverified on
 DGX OS and JetPack). Before, every lane raced for unified memory at once, which

{lobes_cli-0.89.0 → lobes_cli-0.89.2}/docs/qwen3.6-27b-text-nvfp4-mtp.md

@@ -237,7 +237,24 @@ heavier co-resident agents need the headroom.
    rebuild (the pin is unchanged) and no served-model change — this
    constrains *around* the template drift, it does not fix the drift itself
    (out of scope here; filing the `reasoning=False` hardcode upstream is a
-   tracked follow-up). A companion, independent gateway knob
+   tracked follow-up).
+
+   **The derivation is engine-dependent (#310 deviation d5, 2026-10-18).**
+   It is right only on an engine that applies the grammar from the first
+   generated token, which the nightlies did. vLLM 0.30.0 and later start the
+   grammar only **after** reasoning ends. A thinking-prefixed grammar then
+   sits inside its thinking tag: every tool call is accepted as thinking
+   text, EOS stays masked, and a strict request repeats the call until
+   `max_tokens` (about 78 calls in 2048 tokens, on 0.30.0, 0.31.0 and
+   0.32.0rc1, with and without DSpark). On those engines the plugin passes
+   vLLM's own `reasoning` flag through (`False` on every engine measured),
+   chosen from `vllm.__version__` at import and logged; an unreadable version
+   refuses to boot. A `0.29.x` dev build is classified as pre-0.30. Measured on the DGX Spark:
+   one call with reasoning kept, and a `maxLength=3` control still enforced
+   under strict (`docs/evidence/2026-10-18-accept-vllm-031-spark-cortex.txt`).
+   The 0.30.0 boundary is the measured one, not a proven upstream change
+   point: 0.27-0.29 are unmeasured and keep the derivation, and the vLLM
+   change behind it is not isolated. A companion, independent gateway knob
    (`GATEWAY_FORCE_STRICT_TOOLS`) that injects `strict: true` for existing
    callers who don't set it themselves is documented in
    [`docs/openai-api.md`](openai-api.md) and