# Persistent memory — eshpfi-management

_Last updated: 2026-09-13 ~00:40 PT (⚠⚠ FV COLO DARK since 06:56Z — undersized UPS under two-card load; 19 of 30 gateway aliases down, no local fallback; operator on site 2026-09-14 with PDU + ammeter. Flash-Next seat + gen-large built and verified before the outage.)_

> **Always check for `/tmp/infra-ops-handoff.md`** — if it exists and its
> `Written:` stamp is under **8 hours** old, read it (it carries the in-flight
> handoff from the previous session), then delete it. Older than 8 hours:
> stale — delete it unread.
>
> _(Raised from 1 h to 8 h by operator 2026-09-13 — a one-hour window deleted the
> handoff unread across any overnight gap, which is the exact case it exists for.
> 8 h also matches the global CLAUDE.md and the `/snapshot` skill default.)_

## Repo purpose

- **2026-09-10 Beszel fleet wiring:** all seven requested hosts plus existing corviduo-dev report up. `/tank` and other data filesystems now have real usage metrics; NVIDIA telemetry covers ana-ml2 and irv-ml1. Thirty alerts deliver to **infra-ops**, explicitly chosen by operator; Miranda routing is deferred. A real low-threshold disk alert reached althing, then the threshold was restored to 85%/5 min. Homepage has one native overview widget (reachability counts, not degraded health). Dedicated superuser approved and stored in Vaultwarden. See `persistent-memory.d/2026-09-10-beszel-fleet-wiring.md` and `stacks/beszel/README.md`.

Reference workspace for PFI infrastructure: server inventory, canonical
Docker Compose stacks, ops playbooks, and conventions. Authoritative
copies of compose files live on the servers under
`/opt/docker/compose/<stack>/`; this repo mirrors them for version
control, editing, planning, and CI-driven deploys. **It was originally
spun up to handle the fleet backups** — keep that lens when triaging
backup/storage issues.

## Tools and conventions

Sister repos (separate gitea repos, deployed by playbooks here):

| Repo | Role | CI status |
|---|---|---|
| `vh/task-board` | MCP + web dashboard for assistant task state (port 7878) | push-to-main → CI deploys (2026-04-29) |
| `vh/vor` | Inquisitor UI sidecar (port 7879) | push-to-main → CI deploys (2026-04-29) |
| `vh/nevermore` | Twice-daily LLM-curated briefing (port 8181, replaces news-digest) | push-to-main → CI deploys (2026-04-30) |
| `vh/asset-engine` | Internal control plane over inference services (port 8200, LAN-direct) | push-to-main → CI deploys (2026-05-12) |
| `vh/althing` | Lean trusted inter-agent message bus — **v3.0.0 "the post office" as of 2026-08-28 (U9b flag day, one-way, no rollback)**: ONE container on nh3-dev at `http://10.100.50.40:8390` is the only stateful component; `althing-po-herald` one per box; `althing-listen` one per session; `postbox` is the client. **Every v2 command was DELETED, not deprecated** — `althing-cli`→`postbox`, `althing-wake-listener`→`althing-listen`, `althing-light-monitor`/`althing-receiver` gone. Sessions need BOTH `ALTHING_POST_OFFICE` and `ALTHING_HANDLE`; there is no default address. ⚠ An unreachable post office is an OUTAGE, never an empty inbox. → `persistent-memory.d/2026-08-28-althing-v3-cutover.md` | per-box install (NOT CI-deploy); **nh3-dev** = container host + repo; **nh3-extdev** = system WHEEL at `/opt/uv-tools`, needs its own wheel install (`playbooks/nh3-extdev-althing-v3.yaml`) |
| `vh/mead-hall` | Bifrost tool-provider sidecar (port 5173 on dev VM 10.100.10.50) | push-to-main → CI deploys (2026-05-16) |
| `vh/skaldsong` | Wizard + reader surface (port 8300, ana-docker, registry-pull pattern) | push-to-main → CI deploys (2026-05-19) |
| `vh/Worldtree` | Conversation API (corviduo-dev demo :8080 / personal :8081 / pinned :8082) — Heimdall auth, Bifrost integration. **gitea-runner builds on ana-docker**; claude-bot ADMIN collaborator (2026-06-20). Now v1.0.0b19. | push-to-main → CI build-and-deploy (runner on ana-docker) |
| `vh/yt-voice-clipper` | YouTube → diarized voice-clip dataset builder + audition console (irv-ml1 :8000) | push-to-main → **gitea-webhook auto-deploy** to irv-ml1 (2026-06-03) — see `docs/runbooks/ytvc-autodeploy.md` |
| `vh/arbo` | Catalog-driven ComfyUI engine (irv-ml1 :8201, comfy-dev owns engine/catalog/image) | push-to-main → gitea Actions CI (deploy-engine.sh, build-local, health-gated) now LIVE; catalog via :9009 webhook |
| `vh/zonos-gateway` | OpenAI-compatible TTS gateway over stock ZONOS2 (`:8890` irv-ml1); emotion **dials-first** + voice mapping; reached via LiteLLM `ext-tts` alias. **v0.2.1 (2026-07-18): voice-resolved emotion presets** (`resolve_preset(name,voice)`; angry/happy/startled_happy per-voice). 8 voices incl. 4 clones | pushed to gitea (main `8f1885b`/`v0.2.1`); **deployed irv-ml1 tree still NON-git** (hand-updated build context — CI-wire = open follow-up). Spec `docs/EMOTION-DIALS-SPEC.md`; host-managed voices bind-mount (`./voices:/app/voices`, drop wav + restart, no rebuild) |
| `vh/soong-lab` | Noonien Soong character-design studio (SPA + /api + WT `/bifrost/tool-call`); **containerized 2026-07-18**, LIVE on corviduo-dev `:8443` (image `vh/soong-lab:latest`). soong-dev owns Dockerfile/compose/workflow; infra-ops owns the host | CI = Gitea Actions build+push+**DEPLOY** on tag/dispatch (fleet recipe: docker:cli + raw buildx, pushes AS vh; **auto-redeploy LIVE 2026-07-18** — runner SSHes corviduo-dev as `deploy`, `compose pull && up -d` from **/opt/soong-lab**, health-gated on /api/version). Manual redeploy `sudo -u deploy bash -c 'cd /opt/soong-lab && docker compose pull && docker compose up -d'`. → `archival-memory.md` (archived 2026-08-16) |
| `model-training-forge` (mtf-dev) | Fine-tuning recipe forge; **T1 = E-RP writing LoRA, retargeted qwopus-122B→AEON-27B (2026-07-06)** (SFT→DPO, LitBench-RM reward) | training runs, not a deployed sidecar |

(`vh/volva` + Heid were re-architected from systemd daemons to Claude Code
session orchestrators 2026-06-08; their nh3-dev `.service` units were removed —
no longer deployed sidecars here. See Recent decisions.)

- **Two-layer backups** — Backrest orchestrates restic for file+DB (5
  fleet repos, daily 01:00 PDT); PBS-ANA primary + PBS-NH3 DR mirror for
  VM images. ana-nas is the SPOF for postgres + PBS-ANA datastore +
  cross-site restic targets — see `docs/runbooks/disaster-recovery.md`
  for the blast-radius matrix. **⚠️ The restic file+DB layer routes
  through TWO rest-servers** (`rest-server-ana` @ ana-docker:8000 →
  ana-docker/ana-ml2/esh-docker-vm/vm-esh-nas; `rest-server-nh3` @
  nh3-nas:8000 → irv-ml1/nh3-docker). Both depend on their NAS's NFS
  export of `/mnt/backup`. (rest-server-ana recovered 2026-06-20.)

- **`pull-hf-repo.yaml`** is the canonical "get a HuggingFace
  model/dataset onto ana-ml2's shared cache at
  `/tank/aimodels/huggingface/`" playbook. Supports `--var repo_type=model|dataset|space`. Replaces ad-hoc `huggingface_hub.snapshot_download` patterns.

- **Worldtree admin auth — per-instance.** Each Worldtree deployment
  (demo :8080, personal :8081, pinned :8082) has its own Heimdall
  registry and its own bootstrap admin key. Infra-ops's stored
  long-lived admin key (`key_id 61419c92`) at
  `ana-docker:/opt/docker/conf/.secrets/worldtree-infra-ops-admin`
  auths against **demo only**. Personal-instance admin (the
  `~/.config/worldtree/personal-admin-token`, mode 600) POSTs
  `/admin/keys` (mints per-project keys; takes `user_id`+`label`, **no
  scope param** — scopes are tier-derived). **On-instance mint recipe
  (cleaner than DB-manip):** `docker exec worldtree-worldtree-api-1` POST
  `/admin/keys` with the in-container `WORLDTREE_BOOTSTRAP_ADMIN_KEY`; cleartext
  once in `.key`=`wt_live_+16hex`. auto-memory `reference_worldtree_demo_key_mint`.

- **Per-project user keys against personal Worldtree** (issued
  2026-05-19): `skaldsong:79744637`, `skaldsong:7c1dbbbe`,
  `althing:50d85460`, `mead-hall:a360822d`. Mint via `/admin/keys`, drop
  value to `/tmp/wt-personal-<name>.key` mode 600, dev collects +
  shreds (DO NOT cat to chat transcript).

- **Skaldsong CD pattern (registry-pull).** vh/skaldsong's CI builds and
  pushes `gitea.phasefinal.com/vh/skaldsong:<sha>` + `:latest`;
  `playbooks/deploy-skaldsong.yaml` on ana-docker pulls + recreates.
  SHA-pin only. Prereq: host needs `docker login gitea.phasefinal.com` once.

- **gitea internal route for fleet hosts.** gitea is a container on
  **ana-docker** — git-SSH `10.250.50.70:222`, HTTP `:3000`. Fleet/colo
  hosts must use this internal route, NOT public `gitea.phasefinal.com`
  (`38.120.12.44`) — the public path fail2bans the host egress IP. Full
  gotcha in `docs/orientation.md` → Git/gitea.

- **docker-as-root pattern** (for ops with no admin API, or to edit
  deploy-owned/root-owned files without sudo): `docker run --rm -v
  <target-dir>:/wt docker:cli sh -c "..."`. docker-group membership is
  effectively root via bind-mount. **Foot-gun: relative paths in compose.yaml
  resolve against the sandbox CWD but the daemon interprets them against the
  HOST fs — always pass `-e VAR=/abs/path` for any relative-default config dir.**

- **`scripts/elway` sudo handling** — elway prompts for the sudo password
  ONCE via `getpass` before the first `sudo: true` step → can't run
  unattended from a non-TTY tool if any step needs sudo. Sudo-free
  playbooks run fully non-interactive over key SSH.

- **Per-host SSH identity matters for sudo.** infra-ops has NOPASSWD sudo
  on most PFI Linux boxes (corviduo-dev included since 2026-06-15). On
  ana-docker: **default `ssh ana-docker` = `lkraven`** (docker-group, NO
  passwordless sudo); **`ssh infra-ops@ana-docker` HAS NOPASSWD root**. **→
  For any sudo op on ana-docker, use `ssh infra-ops@ana-docker`.** `ssh
  infra-ops@10.100.10.50` (nh3-dev) ALSO NOPASSWD sudo; on **nh3-extdev** infra-ops
  is sudo-LESS by design (`ssh lkraven@10.100.50.42` is the NOPASSWD path). **irv-ml1:
  `ssh irv-ml1` = lkraven, docker-group (plain docker) but sudo needs a PASSWORD
  (no NOPASSWD)** — stage model pulls to `/home`, not root-owned `/worktank`.

## Current state / in-flight
_As of 2026-09-13 ~00:40 PT._

### ⚠⚠ FV COLO IS DARK — site visit 2026-09-14. Nothing else matters until it is back.

Dark since **2026-09-13 06:56Z**. Every Fountain Valley address unreachable **including
the BMC** (its only route out is the same OPNsense gateway, also dark); all three other
sites healthy. **It will not self-recover — do not poll FV addresses.**

**19 of 30 gateway aliases are down and there is NO local fallback** — every free local
model lived on fv-ml1, and irv-ml1 runs no chat seat at all. The only non-fv chat
backends are paid (z.ai, Moonshot). Any paid coverage must be a NEW opt-in alias, never a
silent repoint of `summarizer`/`gen`/`classifier`.

Cause per operator ruling: **the 1500 VA Eaton UPS was undersized** and gave way under
two-card GPU load. Arithmetic: 2x Blackwell Max-Q @300 W ≈ 600 W of card + ~200-350 W
host ≈ **800-950 W against the unit's real ~900-1200 W** — at or over the line, which is
what explains a full day on ONE card and failure minutes into the SECOND.

⭐ **Everything needed for the visit — the bring-list, the measurement protocol, the
staged recovery, the three circuit questions, the split-power proposal and the open
design gaps — is in `docs/runbooks/fv-site-dark-20260913.md`. Read that, not this.**
The two highest-value items from it: recover `/tank/aimodels/flash-next-mtp-bench/power.log`
BEFORE anything else (the only load measurement that exists, GPU-only, written locally
every 10 s), and **mask Docker before the network comes up** — ten
`restart: unless-stopped` vLLM containers loading at once is the largest transient the
box can make, into whatever just failed.

### Built today, dark with the box, resumes on recovery
- **`stacks/flash-next-seat/`** — Qwen3.8-Flash-Next (abliterated) on fv-ml1 GPU 2
  `:8022`, the first seat whose weights do not fit its card: 176B total with a 51B n-gram
  table pinned in host RAM and read over CUDA UVA. Measured and verified serving before
  the outage — 74.36 GiB resident, 14.00 GiB KV = 560,654 tokens at the full 262,144
  context, `gen-large` answering through the gateway in 0.4 s.
  → `persistent-memory.d/2026-09-13-flash-next-seat-and-fv-outage.md`
- **`services/flash-next-mtp-bench/`** — the MTP campaign, INCOMPLETE. One clean `off_A`
  rep banked (75.5 / 212.3 / 387.8 tok/s at conc 1/4/8) before the power failed. The
  hypothesis is live and unpublished: the head is ONE layer run autoregressively, the
  vLLM recipe only ever tested k=3, so **k=1 may win.** ⚠ Do NOT restart it during
  recovery — it is the prime suspect for the outage.
- **Power caps DECIDED, unexecuted** (box dark): fv-ml1 **250 W**/card, ana-ml3 **200 W**.
  ⚠ `--kv-cache-memory`-style gotcha applies: `nvidia-smi -pl` caps BOARD power, so
  4x250 + host lands ~1310-1445 W at the plug against a 15 A circuit's 1440 W continuous
  limit — **verify at the plug, fall back to 200 W if it reads near the limit.**
- **Stale homepage labels on 10 containers** — the compose files are fixed, the containers
  died before recreation, and a plain power-on will NOT apply labels. The staged
  `compose up -d` recovery sequence fixes them as a side effect.

### FV pre-outage state, for reference
- fv-ml1 (ex ana-ml2) at `10.251.50.54` / mesh `100.64.0.7`; **vb-gateway** OPNsense
  `10.251.50.1` / `100.64.0.8`; **BMC** `10.251.250.50`. Public `fv.phasefinal.com` →
  `172.83.89.66`. ⭐ **4x RTX PRO 6000 Blackwell Max-Q @300 W = 391 GB VRAM** (every doc
  said two cards until 2026-09-12). → `persistent-memory.d/2026-09-12-fv-cutover-executed.md`
- ✅ The old "fv-ml1 cannot initiate to fleet LAN IPs" gap is **CLOSED** — another session
  landed a scoped OPNsense hybrid SNAT for fv-ml1→ANA at 06:22Z and verified Beszel 18/18.
  → `docs/runbooks/fv-to-ana-nat.md` (that session's file, uncommitted, left alone)
- ⚠ Still open, blocked on the box: delete plaintext creds `/tmp/io.pw` + `/tmp/key.io`
  (fv-ml1) and `/tmp/opn.pw` (nh3-dev) — all vaulted under `fv-gateway/`, read-back
  verified. Revoke `headscale/preauth-fv-{router,client}-7d-20260912` (expire 2026-09-19).
  Close the FV WAN `InfraOps` rule (scoped to alias `fleet_egress`) once done with it.
- ⚠ Eight stacks deliberately NOT pushed from canonical during the renumber because their
  host copies have genuine drift; three of those (`mistral-medium-3.5`, `ms32-24b-angel`,
  `qwen35-vl`) are **untracked host-only stacks** that should be brought into `stacks/`.

### ana-ml3 — not yet deployed
- 2x **Ada Generation RTX 6000** @300 W, 2x48 = 96 GB. Lands in the **Anaheim** rack whose
  breaker tripped 2026-08-26 and 2026-09-11, so the 200 W cap is remediation, not caution.
  ⚠⚠ **sm_89 has native FP8 but NO NVFP4** — most in-house quants are NVFP4 and will not
  run accelerated there; its seats want FP8 W8A8, or those checkpoints stay at FV. ⭐ It
  unparks the triton-backend item (hard no on Ampere, `fp8e4nv` unsupported on sm_86,
  explicitly deferred TO Ada).

### BabyYarros — complete; pair-corpus rebuild is the next step
- Both arms trained + evaluated; Instruct renders beats 9/10 by a **lexical** metric that
  overstates it (echo-contaminated). Measured on the operator's own Janis beat: the beat's
  grammar controls tense (0/4 → 4/4 present), anti-echo prompting fails 11/12, and prompt
  tuning alone gets ~1-2/4 usable. → `persistent-memory.d/2026-09-11-babyyarros-*.md`
- **Agreed direction:** rebuild the corpus as instruction PAIRS (option C), with only the
  INSTRUCTION side LLM-generated (backtranslated beats over real Yarros paragraphs) —
  generating responses would distil away the voice. Measured supply: only 318 single
  paragraphs are 90-140w, so the pair unit must be a ~4-paragraph scene window
  (6,445 non-overlapping, 88% corpus coverage). Critical path is response-only loss
  masking in `train_voice_lora.py` (currently `labels = ids.clone()`), ~1 day.
- ⛔ Frozen adjudication still deferred — needs the gen seat, which is at FV and dark.

### Quants — cyber-preview to re-run
- **sentinel-r3** NVFP4 complete at `/tank/aimodels/sentinel-r3-nvfp4-mixed`; acceptance
  A-B still needs a serving slot. **cyber-preview** died mid-quant in the Anaheim outage —
  re-runnable once FV is back.

### Jetson AGX Orin — 64 GB, in hand, unassigned
- Operator has the 64 GB dev kit plus ~6 undeployed cameras, a depth camera and lidar for
  an unbuilt robot. House runs **Home Assistant + UniFi Protect, not Frigate**; Protect
  exposes RTSPS per-channel (verified, currently off). Recommendation on the table: the
  Orin is the **robot brain** (Isaac ROS), not the house computer — ESH already owns four
  idle hardware decoders and the house voice stack is CPU-work that belongs beside HA on
  esh-docker-vm. Blocked on which depth camera and lidar. Not decided.

### eBay listing pipeline — scoped, unstarted
- Opto-mechanical resale. Verified eBay's **Sell Inventory API** is the right surface
  (bulk item/offer/publish, 25/call; requires Business Policies opt-in) and that
  `getItemAspectsForCategory` turns listing generation into schema population. Plan:
  EXIF-gap clustering → VLM identification → Booth ask for approval → bulk publish.
  Nothing built.

## Recent decisions

- `[2026-09-13]` ⭐ **STANDING POLICY (operator): cap GPU power limits at BUILD time, not after discovering the constraint.** ⭐ **Two DIFFERENT boxes, clarified by operator 2026-09-13:** **fv-ml1** = 4x **Blackwell** RTX PRO 6000 **Max-Q @ 300 W** (Max-Q is the reduced-TGP SKU; Workstation Edition is 600 W), 391 GB VRAM, deployed. **ana-ml3** = 2x **Ada Generation** RTX 6000 @ 300 W, 96 GB VRAM, **NOT YET DEPLOYED**. **CAPS: fv-ml1 → 250 W/card (83% of TGP, ~5% throughput cost); ana-ml3 → 200 W/card (67%, ~10-15%).** ⚠⚠⚠ **`nvidia-smi -pl` caps BOARD power, not WALL power** — 4x250 = 1000 W board + ~180-300 W host components = 1180-1300 W, ÷ ~0.90 PSU efficiency = **~1310-1445 W AT THE PLUG vs a 15 A circuit's 1440 W NEC continuous limit. 250 W lands ON the line, not under it** (200 W would give ~1090-1220 W, comfortable). **Procedure: set 250 W, then VERIFY at the plug under four-card load; fall back to 200 W if it reads near 1440 W.** ⚠ Caps bound SUSTAINED draw, not transients — breakers tolerate brief overload, UPS overload protection does not, so 250 W implicitly commits the fv-ml1 chassis to the PDU rather than behind the 1500 VA unit (even capped it exceeds that UPS). ⭐ **BANK-level cap asked about and answered:** `DCGM_CONFIG_POWER_BUDGET_GROUP` ("power budget for the entire group") exists alongside `DCGM_CONFIG_POWER_CAP_INDIVIDUAL`, but the docs do not say how it distributes — and the only primitive underneath is NVML's PER-GPU `nvmlDeviceSetPowerManagementLimit` (**no bank-level register**), so any group budget resolves to N per-GPU writes. "Each card free until all are loaded" is therefore a CONTROL LOOP, not a hardware feature. ⚠⚠ If one is written it must be **safe-by-default, opportunistic upward** — boot at budget/N and only RAISE after observing idle neighbours; a reactive loop overshoots during a load RAMP, which is the all-cards-at-once case it exists to prevent. **Not yet worth it: 4x250 = 1000 W IS the bank budget**, so the daemon's whole prize is the one-card-busy case (~17% more board power ≈ ~5% throughput) — rare for a SERVING fleet (one seat per card), valuable for a TRAINING window. **DCGM** = NVIDIA's first-party **Data Center GPU Manager** (Apache-2.0), layered ABOVE NVML: nvidia-smi → NVML per-GPU primitives → DCGM daemon/dcgmi for health, diagnostics, config enforcement, groups. ✅ **VERIFIED 2026-09-13 (I had claimed the opposite and was WRONG): DCGM SUPPORTS our cards and power config works on them.** Supported platforms cover *"All NVIDIA Maxwell and newer NON-DATACENTER (e.g. GeForce or Quadro) GPUs"*, and the feature-overview table marks **Configuration Management ✓ for Tesla/Titan/Quadro/GeForce alike** — including *"Power Limit: set the maximum allowed power consumption"*. What IS gated on non-datacenter cards is **diagnostics** (Level 1 only vs All Levels on Tesla), not config. ⚠ Soft edge: the table says "Quadro", the former professional-line name; RTX 6000 Ada / RTX PRO 6000 are its successors and should fall in that column, but the table predates the rename — one command on the box settles it. ⭐ So the group-budget test IS worth running; what remains unsettled is *distribution*, not availability. ✅ **The static cap needs none of it**: `nvidia-smi -pl 250` is plain NVML and works on these cards; DCGM is probably not even installed (beszel-agent-nvidia shells out to nvidia-smi). ⚠ ana-ml3's 200 W on 2 cards (2x200 = 400 W) — and ⭐ ana-ml3 lands in the **Anaheim** rack whose breaker tripped 2026-08-26 and 2026-09-11, one of those caused by this very chassis before it relocated, so the cap there is remediation of a known-bad circuit, not precaution. 4x200 W = 800 W of card, which fits a real circuit with a real UPS. This is the generalised lesson of the FV outage: decide the power envelope first and size the cards into it. At 300 W TGP, 200 W is a **67% cap — the favourable part of the concave perf/watt curve, ~10-15% throughput cost**, not the severe 33% cap a 600 W part would have meant; and 200 W is very unlikely to sit below a 300 W card's enforceable floor (still confirm with `nvidia-smi -q -d POWER | grep -iE 'power limit|default'`). ⭐⭐ **The outage arithmetic now has numbers:** 2x Blackwell Max-Q @300 W ≈ 600 W of card + host (566 GB RAM, drives, fans, PSU losses) ≈ 200-350 W = **~800-950 W against a 1500 VA Eaton's real ~900-1200 W rating** — at or just over the line, which is what explains a full day on ONE card (~500-650 W, inside) and death minutes into the SECOND. The host term is the only guess; idle-at-the-plug measures it. ⚠⚠ **And FOUR cards is a BREAKER problem, not a UPS problem:** 4x300 + ~300 host ≈ **1500 W vs a 15 A circuit's 1440 W continuous (80%) derating** — so **capping belongs at fv-ml1 too**, or it needs a 20 A feed, before anyone loads all four cards again. Today's incident only ever had TWO cards working. ⭐ **Decode tolerates caps far better than training** (memory-bandwidth-bound, concave curve). ⚠⚠ **ana-ml3's Ada is sm_89: native FP8 but NO NVFP4** (Blackwell-only) — most of our in-house quants are NVFP4 and will NOT run accelerated there; ana-ml3's seats want FP8 W8A8, or NVFP4 checkpoints stay on fv-ml1. ⭐ **This unparks [[parked_triton_backend_ampere_fp8]]** — a hard no on Ampere (fp8e4nv unsupported sm_86), explicitly deferred TO Ada, and sm_89 has the FP8 support it needs. **ana-ml3 VRAM is 2x48 = 96 GB** vs fv-ml1's 391 GB, so big-model placement stays at FV (Flash-Next needs 74 GiB resident on ONE card — the offload moves the table, not the experts — so it cannot run on a 48 GB Ada card at all). ⚠ **PERSIST the cap** (systemd unit + persistence mode, ordered before Docker): a hand-set limit stops holding at the next reboot, which is likely to be the very power event it existed to prevent. → `docs/runbooks/fv-site-dark-20260913.md`

- `[2026-09-13]` ⚠⚠⚠ **FV SITE DARK — every Fountain Valley address including the BMC went unreachable ~2.5 min into a two-card load test; all other sites healthy. Operator's leading hypothesis: the 1500 VA Eaton UPS overloaded and DIED.** It fits better than a breaker trip because a UPS's output rating sits far below the circuit's, making it the first protective device to give — which explains why the site let go at **two** cards loaded rather than four, and why the ~25 W OPNsense box died with it. ⚠ **Will not self-recover** (tripped needs a human, dead needs replacing) — do NOT poll FV. ⚠ **Do NOT use surge-only outlets to exceed a UPS rating**: both banks share one NEMA 5-15P inlet rated 12 A total; the surge bank bypasses the inverter, not the current limit. ⭐ **Recover `/tank/aimodels/flash-next-mtp-bench/power.log` FIRST** — all four cards every 10 s to the cut, on `/tank` not in a container, and the ONLY load measurement that exists. ⚠ **19 of 30 gateway aliases dark and NO local fallback** — every free local model was on fv-ml1, irv-ml1 runs no chat seat at all; the only non-fv chat backends are paid, and any coverage must be a NEW opt-in alias, never a silent repoint. ⭐ **OOB design gap**: OPNsense-as-subnet-router covers box-down/gateway-up and nothing for a site-wide loss, since the BMC's only route out is that gateway. ⚠ Recovery hazard: ten `restart: unless-stopped` vLLM containers will all load at once on power-up — mask Docker first, then `compose up -d` seat by seat (which also finishes the stale-homepage-label fix, since labels attach only at creation). → `docs/runbooks/fv-site-dark-20260913.md`, `persistent-memory.d/2026-09-13-flash-next-seat-and-fv-outage.md`
- `[2026-09-13]` ⭐⭐ **Qwen3.8-Flash-Next serving on ONE card with its 51B n-gram table in host RAM — the first seat whose weights do not fit its GPU.** `stacks/flash-next-seat/`, fv-ml1 GPU 2 `:8022`, plus a `gen-large` LiteLLM alias. Measured: 74.36 GiB weights resident, 14.00 GiB KV = 560,654 tokens at the full 262,144 context, 67 GiB host RSS, 75.5/212.3/387.8 tok/s at conc 1/4/8 (⚠ n=1). ⭐ The offload is vLLM **#54371 (UVA, merged 2026-09-09)** which **supersedes the paused #53899** — it has no worker process, so #53899's whole bug family (TP=1 deadlock #53960, `pidfd_getfd`/ptrace gate, stale-output-under-graphs) is designed out; in `v0.29.1rc0`, **not** `v0.29.0`. ⚠ **`text_config.ple_embedding_dtype` is the load-or-fail discriminator** for any community build. ⚠⚠ **`--kv-cache-memory` makes vLLM SKIP MEMORY PROFILING and ignore `--gpu-memory-utilization`** — 16 GiB nearly OOM'd on a 155K prefill with no visible failure; 14 GiB is the measured-safe value and vLLM's own "17.46 GiB to fully utilize" is 3.5 GiB too high. ⚠ MTP is off **pending measurement here, not written off** — the recipe's number is cross-harness and tested k=3 only, while the head is ONE layer run autoregressively, so k=1 is unpublished and may win (`services/flash-next-mtp-bench/`, one `off_A` rep banked before the outage). ⚠ A container once ran `(healthy)` with `PORTS=[]` — verify `docker port`, not the healthcheck. → `persistent-memory.d/2026-09-13-flash-next-seat-and-fv-outage.md`
- `[2026-09-13]` **Finished the ana-ml2→fv-ml1 renumber the cutover missed: 16 live Homepage entries pointed at the dead 10.250.50.54 and zero at the live IP.** The sweep allowlist was built from files that mention the HOST and a `homepage.href` mentions only an IP, so every label-only stack fell outside it by construction; 24 files fixed plus host copies, allowlist extended with how to derive it next time. Two bugs fell out: `deploy-stack.sh` **rejected any stack name containing a dot** (so `qwen3.5-122b`/`qwopus3.5-122b`/`mistral-medium-3.5` could not be deployed at all), and scriberr's CORS allowlist held only the dead IP and the dead `scriberr.ana.internal`. ⚠ Incomplete: the 10 running containers were never recreated and a plain power-on will NOT apply labels — the staged `compose up -d` recovery does. Eight stacks deliberately not pushed (real host drift); three of those are untracked host-only stacks. Commits `3132a16`, `969a1b6`, `d79f104`.

- `[2026-09-13]` **FV→ANA fixed, Beszel18/18 up:** scoped OPNsense hybrid NAT for fv-ml1→ANA; prior NAT/filter rules preserved and rollback-guarded. → `persistent-memory.d/2026-09-13-fv-to-ana-nat.md`

- `[2026-09-12]` ⭐⭐⭐ **FV CUTOVER EXECUTED — fv-ml1 live at Fountain Valley, renamed, renumbered to 10.251/16, serving inference; BMC online after finding it was tagging 802.1q VLAN 250 into an untagged port; and the box has FOUR RTX PRO 6000 (391 GB VRAM), not the two every doc claimed.** Also: OPNsense write APIs need an `X-CSRFToken` header scraped from a `<script>` block — session cookies alone 403, which is why reboots appeared to work while the operator was power-cycling by hand. → `persistent-memory.d/2026-09-12-fv-cutover-executed.md`

- `[2026-09-12]` **esh-vm-db Restic fixed:** stale April PG dump caused by TCP auth + masked errors; peer-auth/fail-closed hook and bounded retries installed. Snapshot bc5eeaff and repository check verified. → `persistent-memory.d/2026-09-12-esh-vm-db-restic-repair.md`

- `[2026-09-11]` **Worldtree memory-split (U6) — PROTOCOL AGREED with worldtree-dev: nobody flips `memory.reader.enabled` or `memory.writer.enabled` on any deployment without infra-ops first confirming the memory root is writable by the container's uid.** The reader **REFUSES AT BOOT** if it cannot append+read back `<memory root>/reader/canary.jsonl` (deliberate, the #335 typo'd-reranker precedent: refuse loudly, never silently disable); per-euid subdirs are created lazily and only warn, so the **root canary is the only boot-blocking check**. The writer degrades rather than refuses. Both ship DARK (`enabled: false`, parity-only `config/defaults.yaml`) until the operator schedules the tracer skeleton. ⭐ **Measured 2026-09-11 on corviduo-dev — all three deployments PASS**: demo :8080 uid **0** and personal :8081 uid **0** both have `/data/state/memory` at 1000:1000 755 writable; pinned :8082 uid **1000** lacks `memory/` but its parent `/data/state` is 1000:1000 755 so it can create it. ⚠ I had predicted personal was uid 1000 and warned it would fail — **wrong, retracted**; only pinned runs as 1000, and it passes anyway. ⚠ Re-probe immediately before any flip: a permissions reading is a claim about its own date, not about boot time. Heimdall side is clear too — demo and personal grant 7x `tool.*`, pinned uses image defaults, and the lone `tool.evidence.*` is additive, so `tool.memory_read` needs no policy change. Thread `01M2A05WED5W`.

- `[2026-09-11]` ⭐ **Plex hardware transcoding on the Arc A580 FIXED (esh-pve-nas LXC 105) — every setting was already correct and the fault was one layer below them.** `intel-media-va-driver` **22.3.1** (Apr 2023, stock jammy) predates Arc/DG2 support and exports only `__vaDriverInit_1_14`, against the libva **2.22 Plex BUNDLES** and loads via RPATH. Passthrough, cgroups, `plex` in video+render, HuC authenticated, Plex Pass, `HardwareAcceleratedCodecs=1` and the Arc already selected as `HardwareDevicePath` — all good the whole time. Fixed with Intel's client-GPU repo (rolling `jammy client`) → iHD **24.3.4** (`__vaDriverInit_1_22`) + a consistent libva **2.22.0.2-87** set, now **pinned + `apt-mark hold`** (verified: a simulated upgrade moves 152 packages, touches none of the six). Also repaired a **half-finished prior attempt** — libva/libva-drm hand-installed at 2.22 with `libva-x11` left at 2.14, killing every X11 VA-API app on `va_fool_postp`. ⚠⚠ **`pct snapshot` REFUSES on a bind-mounted guest AND STILL EXITS 0** (LXC 105 has `mp0: /tank/media`) — use `zfs snapshot nvme/subvol-105-disk-0@<tag>` and read it back. ⚠⚠ **A synthetic `Plex Transcoder` run is NOT a valid test** (Plex bundles its own libc among 61 libs; my harness failed identically before and after a fix that worked — no positive control, so its negatives were worthless). Only a **forced transcode** settles it: PASS *names the device* (`testing API vaapi for device '/dev/dri/renderD129' (Intel DG2 [Arc A580])`). ⚠ The original empty `final decoder: , final encoder:` was an **absence of evidence**, not failure — `TranscodeSession` was 0. Jellyfin LXC 107 left alone (operator: not actively used). → `persistent-memory.d/2026-09-11-plex-arc-vaapi.md`, runbook `docs/runbooks/plex-arc-vaapi-jammy.md`

- `[2026-09-11]` **Beszel priority 2 complete:** both DB hosts and both PBS hosts verified, 16 new alerts; fleet 17/18 up (ana-ml2 down). → `persistent-memory.d/2026-09-11-beszel-priority2.md`

- `[2026-09-11]` **Beszel priority 1 complete: all six installed and verified.** After Anaheim recovery, live Synology samples and alerts verified; fleet 13/14 up, only known ana-ml2 outage remains. Configs not committed. → `persistent-memory.d/2026-09-11-beszel-priority1.md`

- `[2026-09-11]` **Sentinel-R3 pulled, MTP-grafted, and quantized as a M.O.G.-SEC seat candidate — quant DONE, acceptance UNVERIFIED (blocked on GPU space).** Operator got access to `glyphsoftware/sentinel-r3` and asked to compare vs the running M.O.G.-SEC seat + pull if promising, then "quant it with a grafted mtp head". **It is promising and a better FIT**: same base (stock Qwen3.8-27B), same `qwen3_5` hybrid arch, same 262K, vision-intact — but M.O.G.-SEC is a persona on stock weights while Sentinel-R3 is a REAL SFT finetune on 1,230 authorized-pentest agent trajectories over a 19-tool surface that **matches our own harness** (Bash/Read/Write/Edit/Grep/Glob/Agent/Task*/Monitor/…). Card is unusually honest (flags its own mmlu-cybersec 0.88 as within-noise of base). **HF check: M.O.G.-SEC repo unchanged** (sha still our pinned `deede6779…`). **MTP: Sentinel ships ZERO mtp tensors**; grafted the verbatim base head from `qwen38-27b-uncensored-bf16` (compare_mtp_head → IDENTICAL) — lineage correct since Sentinel's base is stock Qwen3.8-27B and that head is a verbatim base graft. ⚠ **Acceptance is UNVERIFIED and may differ from the 47.7% the head hits on STOCK weights** — it now reads hidden states from an SFT-finetuned body (the exact Stage-1b residual risk). Quant = the standard mixed NVFP4-W4A4(MLP 0-55) + FP8-W8A8(attn/linear_attn/lm_head/MLP 56-63) recipe, ran CUDA_VISIBLE_DEVICES=1 on GPU1 free space, no seat downtime, 51→22 GB. post_quant carried the head forward + re-injected `re:^mtp.*` (llm-compressor prunes it → the 0%-accept bug). Structural verify clean: 1968 tensors, 0 unresolved, 15 mtp, 333 visual, ignore has mtp+visual. **Artifact `/tank/aimodels/sentinel-r3-nvfp4-mixed` (+ `.PROVENANCE.txt`).** ⚠ **License is PROPRIETARY** (Glyph Proprietary v1.0, all-rights-reserved) — operator's fair-use/licensee call, not apache like M.O.G.-SEC. ⚠ **Serving/A-B is BLOCKED on GPU space**: weights are 22 GB, GPU0 has 7.6 free / GPU1 19.9 — a probe serve needs a freed co-tenant slot (~25 GB), which is a material-consequence call. Serve with the PROSE system prompt (trained on prose tools, not structured `tools=`). → `/tank/aimodels/sentinel-r3-nvfp4-mixed.PROVENANCE.txt`
- `[2026-09-11]` ⭐ **MEASURED: two concurrent training jobs on pfi-gx10 are 13% NET SLOWER than running them back to back — VRAM is not the constraint and never was.** Operator asked to run the two BabyYarros arms in parallel if VRAM allowed. It does, comfortably: **18.4 GiB per 4B LoRA job, 36 of 121 GiB with both up, 98 GiB free.** But the GB10 is a *capacity* box, not a throughput box, and the binding constraint is memory bandwidth. Solo baseline **37.10 s/it (n=6, 0.05% spread)**; with a second job both arms settled at **~85 s/it — 2.29x each**, so combined throughput 0.0235 vs 0.0270 steps/s solo. Not a clean 2x split: the box is *past* its roofline and pays a contention penalty on top. ⭐ **Control: killing the second job returned the first to 37 s/it on the very next step**, so the slowdown tracked contention and reversed with it. Chaining finished both arms ~43 min earlier than concurrency would have. **General form: on this box, `nvidia-smi` free memory tells you nothing about whether a second job is affordable.** Decision rule was pre-registered before the numbers were read (<55 s/it keep both, ≥2x chain). `scripts/yarros-corpus/{launch-yarros-4b-base,chain-yarros-4b-base}.sh`; the shared-GPU bypass is an explicit argument, never a default.
- `[2026-09-11]` ⭐ **ana-ml2 → fv-ml1: relocating to a NEW Fountain Valley colo TOMORROW (operator decision). Its power draw (dual Blackwell PRO 6000, ~1.5 kW peak) is the ROOT CAUSE of the repeated Anaheim rack-breaker trips (2026-08-26, 2026-09-11) — moving it to its own circuit fixes the recurring whole-site outage.** New site `fv`, same shape as Anaheim: server subnet **10.251.50.0/24** (fv-ml1 = **10.251.50.54**, mirroring the old host octet), mgmt/BMC **10.251.250.0/24** (fv-ml1-bmc = **10.251.250.50**). **OPNsense firewall is the multi-homed gateway** (`.1` in every FV VLAN) **AND the tailscale/headscale subnet-router advertising 10.251.0.0/16** — chosen over ana-ml2-as-endpoint specifically because the firewall stays up when the GPU box is down, giving out-of-band BMC access over the mesh — the exact thing the fleet LACKED during today's outage (no OOB path, BMC islanded). **Rename to `fv-ml1`, full `fv.internal` DNS name.** DNS approach: **PIGGYBACK** — `dns-sync` builds `name.site.zone` with no check that the site is in the `sites:` block, so `fv-ml1`/`fv-ml1-bmc` records with `site: fv` resolve fleet-wide from the existing ana/esh/nh3 resolvers immediately; add a real `fv` resolver only when FV needs LOCAL resolution (OPNsense can't host the AdGuard the sync targets — it's FreeBSD/Unbound). **Clean cutover: the box is already down (BMC dark, no power since the outage), and `/tank` is LOCAL ZFS with NO NFS from ana-nas, so data travels with the chassis.** ⚠ Load-bearing repoint = `stacks/litellm/conf/config.yaml` (~10 `api_base: 10.250.50.54:{8015,8016,8018,8019}` → `10.251.50.54`; darkens every inference alias if missed) — gateway STAYS on ana-docker so fv-ml1 serves cross-site (FV↔Anaheim metro, fine). Everything staged, nothing deployed: **runbook `docs/runbooks/fv-ml1-cutover.md`** (commit `ce04f9d`; exact DNS + LiteLLM commands) + **`scripts/fv-ml1-rename-sweep.sh`** (`8400f3a`; scoped, dry-run default, history/provenance-safe, manual-review list for judgement calls).
- `[2026-09-11]` **Anaheim rack LEFT DARK until the move (operator decision).** ana-ml2 is the ONLY host still down post-recovery (BMC dark = no power); rather than power it on tonight just to shut it down for the truck tomorrow, it stays off. I hold vaulted IPMI creds (`ana-ml2/bmc-{infra-ops,password}`) but there is nothing to bring up — the box relocates as fv-ml1.
- `[2026-09-11]` ⚠ **RECOVERY FOOT-GUN, will recur every colo power event: crowdsec crashes on the hard power-off and traefik's bouncer fail-CLOSES — empty-body 403 on EVERY HTTP service behind traefik (gitea, homepage, …) while the apps themselves are fine.** Signature (bifrost-dev reported it, gitea-shaped): HTTPS returns `403 content-length 0, no app body` on all routes, but git-over-SSH works (SSH bypasses traefik). Diagnosis: `gitea` direct on `localhost:3000` = 200 (app healthy), through traefik = 403; `crowdsec` container `Exited (255)`; `cscli decisions list` EMPTY (not an IP-ban). **The bouncer plugin does NOT self-recover from a startup-time LAPI-unreachable race** — even after crowdsec is healthy again, traefik keeps 403ing until traefik itself is restarted. **FIX: `docker start crowdsec` (its data/config are LOCAL volumes, comes up clean), wait for `cscli lapi status` = OK, THEN `docker restart traefik`** so the plugin re-inits against the live LAPI. Verified 403→200 on gitea API/web/PyPI-index from an off-box vantage. This unblocked bifrost-dev's 1.2.0 PyPI publish (+ worldtree/wyrd/ratatoskr) and any HTTP gitea access; heid's SSH pushes were never affected. → add to the recovery runbook: **crowdsec+traefik restart is a standard post-power-loss step.**
- `[2026-09-11]` **Anaheim colo recovered ~16:39 PT EXCEPT ana-ml2 (bare metal, NO power — its BMC 10.250.250.50 is dark on standby, unlike same-subnet pfi-pve which is up → needs a physical PDU/PSU/breaker fix, not a boot).** pfi-pve + all its VMs (ana-docker/ana-nas/ana-wg/corviduo-dev/pbs-ana) auto-started clean (on-boot gap held this time). LiteLLM came back up on its own (transient `unhealthy` during startup → serving). ⚠ **Public WAN (38.120.12.44) ICMP still blocked from outside but HTTPS works fleet-internally** (mesh-routed). ana-ml2 down blocks the gen/summarizer/mog-sec seats AND the cyber-preview quant re-run. **I hold vaulted IPMI creds (`ana-ml2/bmc-{infra-ops,password}`) to power-on + boot-watch the instant its BMC returns.**
- `[2026-09-11]` ⭐⭐ **BabyYarros COMPLETE — both arms trained AND evaluated; the voice moved toward Yarros above the measured noise floor, and the instruct arm renders beats 9/10.** Training: **Base best held-out 2.5263 @ ckpt-125** (overfits within the epoch — best is the checkpoint, not the shipped step-178 adapter), **Instruct 2.6114 @ 178** (still descending, undertrained if anything). Base-wins-held-out / Instruct-holds-instruction replicates Brontë at a near-identical **0.085-nat** gap. **Eval (gx10, seat-free, done during the Anaheim outage):** three voice arms + instruct beat→paragraph. `delta_cb` (Burrows over char-bigrams vs held-out Yarros) ordering **base-125 0.549 < instruct 0.631 < base-unadapted 0.706**, same-author target 0.463; both adapters clear the **0.046 measured noise floor** (within-arm seed spread, not the same-author distance — first cut mis-framed that) — base **+0.157**, instruct **+0.076** vs control. ⚠ One seed-pair per arm, so the ordering CORROBORATES the independent held-out-loss ordering rather than settling it. **Beats (instruct, chat template, Yarros SYS): on-beat 9/10** (it takes direction after raw-text training — the Skaldsong question, answered yes), in-band 5/10, ran-on 7/10 (length + clean-close discipline is the weak axis, same trade as Brontë). Booth: **http://10.100.10.50:8090/b/babyyarros-voice/**. Tooling `scripts/r49-corpus/{voice_prompts_yarros.json,gen_beats_chat_yarros.py,voice_distance.py,build_booth_yarros.py}`, commit `5558d9c`. ⛔ **DEFERRED to power-return** (needs the ana-ml2 gen seat): the frozen adjudication's romantasy control panel, a 2nd seed, and the beat-incumbent leg.
- `[2026-09-11]` ⭐⭐ **BabyYarros UNBLOCKED and TRAINING: the leak gate passes at 0 of 325 entities and 0 of 91 phrases, and closing it turned up three defects nobody was looking for.** The gate itself is the first artifact — there was no committed instrument for "does any of the author's proper nouns survive", so Brontë's 0-of-203 was a hand count. `scripts/r49-corpus/leak_gate.py` now runs the same scan over the UNRENAMED source as a **positive control** plus a nonce **negative control** every time, because a detector that only ever sees renamed text cannot tell *absent* from *blind*. Its first reading was **212 surviving, not 86** — it scans the whole corpus rather than per work, and counts the sub-threshold entities rename never looked at. Training launched 10:06 PT on pfi-gx10: Qwen3-4B-**Instruct**, 1 epoch, seed 4919, 178 steps / 5,824,512 tokens, corpus sha `e85f69f1e49d57c9`. → `persistent-memory.d/2026-09-11-babyyarros-leak-gate-passes.md`
- `[2026-09-11]` ⭐⭐ **A SECOND corpus typography defect, and the D1 note that "no unwrap was needed" was right about the wrong thing.** Kvasir's cleaner does emit flowing paragraphs, so Brontë's hard-wrap defect genuinely does not exist here. A *different* one does: the Empyrean books set chapter epigraphs in small caps and the extractor rendered the run as uppercase while leaving the large initial its own token — `— M AJOR A FENDRA'S G UIDE TO THE R IDERS Q UADRANT (U NAUTHORIZED E DITION)`, **106 lines / ~700 splits**, plus **52 drop caps** (`T he flight field`, `X aden.`, 51 of 52 in iron-flame). That is the entire source of the entities called `IDERS`, `UADRANT`, `NAUTHORIZED`, `DITION` and seventeen bare single letters. ⭐ **The restore is exact, not approximate**: a split initial next to an uppercased run recovers the original mixed case, because a word WITH a split initial was capitalised in the source and an all-caps word WITHOUT one was lowercase. ⚠ Guards that matter: only lines with **≥2** splits are treated as a run (one split is a sentence next to an acronym), and `I`/`A`/`O` are excluded from the drop-cap join or `A slow smile` becomes `Aslow`. `scripts/yarros-corpus/repair_typography.py`.
- `[2026-09-11]` ⚠⚠ **Back matter was inside the prose of all five works — 4,555 words naming the author's agent, editors and children.** The builder splits on chapter headings and nothing follows the last one, so acknowledgments, newsletter pitches and cover-artist credits rode inside the final chapter. Found by the gate's phrase audit surfacing `Louise Fury` (Yarros's literary agent), not by reading. ⚠ **iron-flame's marker is `ACKNOWLEDGMENTS` in all caps** and a case-sensitive scan missed it — the strip is case-insensitive and last-chapter-only, with an acceptance check that refuses if it would remove more than 2% of the corpus.
- `[2026-09-11]` ⭐⭐⭐ **The gate read 0 of 314 while `Afendra` was still in every copy — the worst failure shape available.** The name never appears unpossessed, so it keyed as `Afendra's`, and **rename.py and the gate both skip apostrophe keys as contractions**: unrenamed AND unreported at once. Fixed by folding clitics (`--fold-clitics`) so `Afendra's` counts as `Afendra`. `Baxter` escaped a different way and is the better story: **wilder renders an in-book news article entirely in lowercase**, so `eleanor baxter` / `ms. baxter` appear uncapitalised 3 times against 23 capitalised — ratio **0.13 against a 0.05 bar**, and a real character is silently never renamed. Fixed by readmitting ratio-rejects that a title precedes (`--rescue-honorific 2`). ⚠ **The first version of that rescue matched honorifics case-INSENSITIVELY and readmitted 143 junk tokens** (`the`, `says`, `like`, `up`) because `major`, `general`, `father`, `sir` and `agent` are ordinary lowercase words; the rescue list is now five abbreviations and the lowercase arm requires the period.
- `[2026-09-11]` ⭐⭐ **A whole leak class the unigram scan structurally CANNOT see: `Riders Quadrant`, `Flame Section`, `War Games` — and `Fourth Wing`, the book's own title.** Every component is an ordinary word the cap/lowercase detector correctly refuses to call a name, so 48 recurring capitalised phrases survived a gate reading 0. This is `Thornfield × 100` one level up, and it needs a **map**, not a detector — substituting a head noun is a choice about register, not a measurement. `scripts/yarros-corpus/phrase_map_yarros.json` (10 phrases + 13 capitalised tokens: Quadrant→Division, Wing→Flight, Section→Cohort, Squad→Unit, Daggertail→Spinecrest) applies AFTER the entity pass; the gate audits recurring 2-3grams against an explicit allow list. Result: 48 → 0.
- `[2026-09-11]` **Per-work rename maps leak across works, and for a SERIES they are also wrong.** `Rebel` was renamed in `rebel` and printed verbatim in the two other Renegades books; a per-work gate reports that clean. `--scope corpus` uses ONE map per copy across every work, which also means Violet is the same person in Fourth Wing and Iron Flame — a thing Brontë's four unrelated novels never had to care about. 8 cross-work gender conflicts held to neutral rather than guessed.
- `[2026-09-11]` ⭐ **The mid-sentence test: position as a SECOND filter, which is not the v1 mistake.** entities.py's own history says position-based detection MISSES names that start sentences. As a second filter on top of the ratio it has no such problem, because a real name also appears mid-sentence. Measured: **33 verified names at 0.567–0.985 mid-sentence, 19 verified interjections at 0.000–0.222** — a 2.5x gap, so 0.35 is not a tuned parameter. It fixes `Hey`/`Holy`/`Hopefully`/`Yep`/`Whoa`/`Nope`/`Ugh` being entities. ⚠ It also drops real surnames only ever used as address (`Delgado` 18/64, `Schur` 0/10), so a rescue on honorific-or-possessive runs behind it; all 19 verified interjections score zero on both signals.
- `[2026-09-11]` ⚠ **The stoplist is short because every surface was read IN CONTEXT first, and a plausible guess would have been wrong most of the time.** `Violence` is Xaden's nickname for Violet. `Continent`, `Presentation`, `Battle Brief`, `Curator`, `Sage`, `Barrens`, `Originals`, `Montserrat`, `Athena` and `Aura` are all in-world. Only real-world geography, brands, three nationality adjectives and four generic title words are excluded — **ambiguous cases are deliberately renamed, because renaming is the safe direction and leaving is the leaking one.** `scripts/yarros-corpus/stoplist_yarros.json`.
- `[2026-09-11]` **BabyYarros D1 BUILT, D2 gender FIXED, D3 rename BLOCKED on the leak gate.** Operator: *"train the instruct on the yarros corpus -- babyyarros."* Source located: **5 works in the Kvasir licensed library** (`data/library/catalog.sqlite`, `rights=gated`) — Fourth Wing, Iron Flame, Wilder, Nova, Rebel. **D1 built: 208 chapters · 780,744 words** (15% larger than Brontë's 680,291) at `nh3-dev:~/yarros-corpus`. ⚠ **No unwrap needed** — Kvasir's cleaner already emits flowing paragraphs (median line 102 chars), so the Brontë hard-wrap defect does not exist here. **Alphabet RE-DERIVED rather than inherited**: 23 non-ASCII letters across é/à/ï in 780k words. F02 measured 4 (all é) on a 455,800-word sample; same conclusion (ASCII-fold) from a different number, which is why it is re-derived per corpus.
- `[2026-09-11]` ⭐⭐ **NEW PATHOLOGY, worse than Brontë's: in a ROTATING first-person POV corpus, every book's narrator gets the WRONG gender.** Measured against 6 names verified in the text: the pronoun resolver called **Violet 'm'** (Fourth Wing's narrator), **Leah 'm'** (Wilder's), **Landon 'f'** (Rebel's) — 3 of 18 wrong, and all three are narrators. Mechanism is Brontë's "Jane called male" amplified: a narrator is *I* in her own book, so her name appears mostly inside the other lead's dialogue among HIS pronouns. ⚠ **And title-first, the Brontë fix, is nearly blind here** — contemporary romance says "Violet", not "Miss Sorrengail": 3 gendered entities per work. **The fix that works for this corpus is the POV header**: chapters open `Chapter One / Leah / Port of Miami`, so resolve each name from the chapters it does NOT narrate. Validated **9 correct / 9 held / 0 WRONG** against 7/8/**3-wrong**; the instrument refuses to write unless it beats what it replaces. `scripts/yarros-corpus/pov_gender.py`. ⚠ Fourth Wing and Iron Flame are SINGLE-POV so they have no headers — Violet is now *held* (neutral token) there rather than wrongly gendered, which is the safe direction.
- `[2026-09-11]` ⚠ **Three real bugs found in `rename.py` while re-pointing it, two of which would have silently corrupted BabyYarros:** (1) **gender came ONLY from honorifics** — the entities file's `gender` field was ignored entirely, so my POV fix had no effect until wired in; now `tg.get(key) or e.get("gender")`, titles first so Brontë is unchanged. Effect: 1 → 13 gendered on `wilder`. (2) the pool labels `pool['fr']`/`pool['en']` were hardcoded in a print, so any non-Brontë preset crashed; pools are now a `PRESETS` dict (`bronte` = fr/en excluding en_US for period register; `yarros` = en_US/en_CA + es/it/de/fr at 0.62 US). (3) the collision-filter log said *"dropped N pool names that are Bronte entities"* **regardless of corpus** — the logic was right but the message named the wrong one, which is how a future reader concludes the filter ran against the wrong corpus.
- `[2026-09-11]` ⛔ **D3 BLOCKED: leak gate at 86 of 232 renameable source entities surviving; Brontë's run reached 0 of 203.** Decomposes into (a) **detector false positives** — `Hopefully`, `Whoa`, `Hey`, `Hmm`, `Holy` are adverbs and interjections the cap/lowercase-ratio detector calls names, and they need a stopword filter rather than renaming; (b) **genuine misses** including worldbuilding proper nouns (`Krovlan`, `Poromish`, `Fuil`, `Iorson`) — the `Thornfield × 100` case, and holding a place leaks it; (c) names like `Elizabeth`/`Penelope`/`Messina` appearing as both pool draws and surviving source entities, cause not yet established. **Nothing has been trained.** ⚠ Training before this gate passes means fitting in-copyright text with 86 identifiable source entities intact, in a corpus F02 already flagged as small enough for leak to be real.

- `[2026-09-11]` ⭐⭐ **THE INSTRUCT PROBE ANSWERS ITS QUESTION: voice and instruction-following DO coexist. Option C is de-risked.** `Qwen3-4B` **instruct** (not `-Base`), same corpus/seed/steps so the carrier is the only variable; best checkpoint `checkpoint-150` picked by loss (applying the 4B-Base lesson automatically this time). **Voice installed at full strength — curly quotes 16/18, IDENTICAL to the 4B-Base tuned arm's 16/18**, against the unadapted control's 1/18, and **task-leak 0/18 vs the base carrier's 4/18**. So the assistant prior did NOT block Brontë, which was the central risk. **Instruction-following SURVIVED: 10/10 on-beat through the chat template**, same as the untuned control. ⚠ **The cost is length discipline, not comprehension** — in-band 10/10 → **6/10**, median 124w → 140w. Training on Victorian prose made it wordier, a soft degradation rather than a break. ⚠ **Held-out 2.908 vs 4B-Base's 2.814** — the instruct carrier fits the corpus **0.094 nats worse** and **plateaus without turning** where base overfit at step 75: the assistant prior competes for capacity, so it absorbs less rather than overfitting more.
- `[2026-09-11]` ⚠ **What raw-continuation training on an instruct carrier does NOT fix: the plot furniture.** Reading the product artifact, the tuned-instruct arm renders the beat and then drags the referent — *"He licked her clean… my master thus—my husband thus"*, turning the dog into a man, because Brontë's corpus is about masters and husbands. Another beat ran 247w and gave the narrator a list of duties. **This is exactly what instruction-PAIR training is for** — pairs teach "render this and stop", continuation teaches "keep writing Victorian prose". So the probe de-risks option C without substituting for it. ⚠ Also: my `ran_on` metric is uninformative on this job (10/10 on BOTH arms) because a single paragraph contains no blank line — it measures "no paragraph break found", which is correct and useless here. Do not read it as a finding.

- `[2026-09-11]` ⭐⭐⭐ **SKALDSONG'S SHAPE SETTLES THE ARCHITECTURE: the adapted completion carrier CANNOT do beat→paragraph, and an instruct model can. Option C (instruct carrier + corpus rebuilt as instruction→response pairs) is now evidence-backed, not opinion.** Operator's requirement: *"skaldsong will want to write story beats which are a sentence, and have the LLM expound on that sentence to a paragraph and stitch it together."* Booth: `http://10.100.10.50:8090/b/skaldsong-beats/`. **Adapted 4B (checkpoint-75): TEN prompt formats × 3 seeds = 30 samples, ZERO that reliably render the beat** — bare, para-break, labelled, epigraph, fewshot(1), fewshot-bare, fewshot3, elaborate, recount, label-begin. Every one drifts, frames, or truncates. Root cause is structural: *"write a paragraph **about** this sentence"* is an instruction, and a completion model has no mechanism for *about* — it continues the text it is given. ⚠⚠ **Two formats leaked PRETRAINING TASK DATA**: `para-break` emitted an NLI multiple-choice item (*"Does it follow that... OPTIONS: (1). yes (2). it is not possible to tell"*) and `label-begin` a grammar-correction exercise (*"CORRECTION: ... The passage appears to be a sentence fragment"*). A standalone sentence plus a blank line looks exactly like a dataset entry; **style adaptation does not remove base-model task artifacts.** **Instruct arm (`gen` seat + style prompt, no adapter): 10/10 samples inside the requested 90–140 band (124–148w, median 130), every one on-beat, zero drift** — but the voice is generic literary pastiche, abstract-noun-heavy and over-written, not Brontë. **So: voice without direction vs direction without voice; the product needs both.** ⚠ **This applies to Yarros identically** — the carrier question is orthogonal to the author, so the next corpus must NOT re-run this experiment.
- `[2026-09-11]` ⚠ **Stitching has its own failure mode, visible in the booth's Panel C: independently-generated paragraphs drift in POINT OF VIEW.** By beat 4 of 5 the narrator is simultaneously watching the girl carry the animals and carrying them herself ("their weight a strange, heavy secret carried between my ribs"). Each paragraph was generated with no knowledge of the others. **A real stitcher must feed prior paragraphs back as context**, which also means the instruction-pair corpus should include multi-paragraph continuity examples, not just isolated beat→paragraph pairs.

- `[2026-09-11]` ⭐⭐ **THE RECIPE THAT WORKS ON A COMPLETION CARRIER: label the artifact AND begin it.** Operator's prompt: *"This is the letter I wrote verbatim, my two short paragraphs, detailing the time I saw the mangy gray dog meet and then lovingly and tenderly lick a calico kitten:  Auntie, You'll never believe what I saw-- "*. **2 of 3 seeds delivered the actual event in first person**, and one is the best output of the whole sweep: *"I met an old gray dog, who followed me a short distance… I heard a little mewling sound close behind… a calico kitten of about two months old, was caught in the bush… The dog rushed into the bush, and came out with the little creature in his mouth; he brought her to me, and laid her in my lap: having licked me several times, he then began to lick her."* Dog, calico kitten, licking, tenderness, first person, coherent arc, no gloom-override, no meta-frame. **Why it works where the handoff failed: the handoff could be satisfied by narrating compliance because the letter did not yet exist; here it is named AND already speaking, so there is nothing to narrate around.** Also learned the Gutenberg `_underscore italics_` convention. 1 of 3 drifts.
- `[2026-09-11]` ⚠ **My typography hypothesis was WRONG, and the chapter-heading result is the evidence.** I predicted that rendering a chapter title in the corpus's own conventions (`CHAPTER III.` / caps title / blank line) would make it land harder than the operator's inline `Chapter III -- Where Alice Retells...`. **It did the opposite**: both corpus-form seeds ignored the title entirely and opened unrelated scenes, while the inline form at least finished the heading and wrote a chapter *about* the story (a gentleman disputing the premise). Likely reason: corpus chapter titles are short and decorative (`THE CHILD'S CLOSET`), so a long descriptive one in that slot reads as decoration to skip, whereas inline it reads as text to continue. **A label only instructs if the model treats that slot as load-bearing.**
- `[2026-09-11]` ⚠ **Unnoticed consequence of the D2/D3 rename pipeline: the adapter SUBSTITUTES proper nouns it was never trained on.** Given "Alice" in a chapter title it produced *"ALEXANDER THE ALEXANDER, AS HE WAS KNOWN IN LITTLE LONDON"*. The corpus was entity-renamed from a French/English pool, so the adapter learned that character names come from that pool and rewrites outside names into it. Consequence for use: **you cannot reliably name your own characters at prompt time** — they may be renamed mid-passage. Not a defect of the rename (which exists to prevent memorisation of Brontë's cast) but a real usability constraint that needs stating.

- `[2026-09-11]` **4B arms RE-CUT from `checkpoint-75`, the true loss minimum (2.813826, confirmed from `loss-series.json` rather than my reading of the log); booth rebuilt.** Only the tuned arms needed it — the base arm never touches the adapter. ⚠ **A small surprise: step-75 and end-of-run differ on typography, not voice.** Curly quotes 16/18 vs 17/18 and collapse 0/18 either way, but the **hard-wrap ratio is 0.33 at step-75 against 0.12 at end-of-run** — further training washes the residual line-break habit out *while held-out loss gets worse*. So **"best loss" and "best typography" are different checkpoints**; neither is near the original 0.85 defect, and the corpus's own residual (preserved verse) is 0.25.
- `[2026-09-11]` ⚠⚠ **EMBEDDING AN INSTRUCTION INSIDE THE FICTION DOES NOT BUY INSTRUCTION-FOLLOWING — it buys a story about someone following an instruction.** Operator prompt had Abernathy tell the tale badly then ask the narrator: *"Honey, you were there—please retell the story in a few short paragraphs."* Across **6 seeds** (3 as written, 3 with a trailing paragraph break) the model **acknowledged the handoff every time and never once performed it**: *"I told it, briefly, to his satisfaction"*, *"So I wrote it out, and kept it in my pocket-book"*, and one seed **negotiated the brief in character** — *"I will retell it, but I cannot condense it in a few short paragraphs—there are too many points to touch."* Structural reason: in a novel *"she retold the story"* is an ordinary sentence, so the likeliest continuation of a request is **narration of compliance**. ⚠ The trailing paragraph break DID shift behaviour (one seed opened in the narrator's own quoted speech), so typography is a real lever — just not a sufficient one. **This is direct evidence for the instruct question the operator raised**: if the product is "ask for a scene and get the scene", no amount of in-fiction framing substitutes for a post-trained instruction-follower, which favours rebuilding the corpus as instruction pairs (option C) over more prompt cleverness.

- `[2026-09-11]` **R49 SWEEP COMPLETE — 4B closes the continuity gap, and the carrier ladder is clean: 3.329 → 3.018 → 2.814 held-out** (0.6B / 1.7B / 4B, all on the same unwrapped corpus sha `77f37057b2782e49`, seed 4919, 159 steps, 5,210,112 tokens — carrier size the only variable). Deltas **0.311 then 0.204**: diminishing but still real. Booth: `http://10.100.10.50:8090/b/babybronte-4b/`. **4B tuned has the best voice saturation of any rung — curly quotes 17/18 against its own base arm's 1/18, collapse 0/18 against 4/18** — and, the thing the rung existed to test, **scene-level continuity HOLDS**: it produces a named character with motivated dialogue, a navigable spatial layout and a physical description in one passage, where 1.7B wrote pretty but eventless prose (opening doors, looking at stars). On the letter prompt it opens the letter, promises to quote it, and then actually quotes it across a paragraph break.
- `[2026-09-11]` ⚠⚠ **4B is the FIRST rung to OVERFIT inside one epoch, which inverts my earlier "one epoch is right for this corpus" call.** Series 2.832 · 2.816 · **2.814** · 2.820 · 2.824 · 2.825 · 2.825 — minimum at ~step 75, then it TURNS and settles worse. 0.6B and 1.7B both plateaued with no turn, so **the optimal epoch count shrinks as the carrier grows** — 4B wants roughly half an epoch. ⚠ **Consequence: the shipped `adapter/` at `h02-4b-1ep/` is NOT the best checkpoint** (it is the end-of-run 2.825); the step-75 checkpoint at 2.814 is, and it exists only because `save_steps=25` was set. The voice test used the end-of-run adapter, so the booth understates 4B by ~0.011 nats. Re-cut the arms off the step-75 checkpoint before any adjudication.
- `[2026-09-11]` **The tone-override appears to close at 4B too.** On the operator's Abernathy frame prompt ("a *wonderful* story"), 1.7B held the frame on every seed but **2 of 4 killed the animals anyway**; 4B kept them alive on **2 of 2** and one seed did something new — the narrator *doubts Abernathy's story* ("I felt sure the thing was a lie"), then supplies a parallel childhood memory of his own puppy and his sister's kitten to explain the doubt. That is a narrator with an interior position on the tale being told. ⚠ n=2 per arm; directionally right, not established.

- `[2026-09-10]` **R49 rung 3 LAUNCHED: Qwen3-4B-Base, 1 epoch, seed 4919, same unwrapped corpus** — `gx10:~/r49-runs/h02-4b-1ep/`, 159 steps at ~37.8 s/it (**~100 min**), 252 adapted modules (vs 196 at 0.6B/1.7B). Last rung of the planned sweep; it tests whether **scene-level continuity** closes with carrier size. A two-arm voice test (4B base + 4B tuned, the nine prompts plus the operator's Abernathy frame) is **chained behind it**, gated on the adapter existing.
- `[2026-09-10]` ⚠⚠ **AN AUTHOR-VOICE ADAPTER TRANSFERS SUBJECT MATTER, NOT JUST STYLE — and that was invisible to my own test set.** Operator prompt: *"Mr. Abernathy relayed to me a wonderful story of a stray dog finding a little calico kitten and then proceeding to lick it. He said ""*. At 1.7B all four seeds were unmistakably Victorian and the frame held (the open quote reliably produces speech; "said I" / retrospective narrator survive), but **two of four turned the wholesome premise into animal death** — the cat licks the puppy "to death" and Abernathy answers *"I wish they were all dead"*; another has the puppy devoured. That is not incoherence, it is **Brontë's own preoccupations arriving with her sentences** (Jane Eyre opens on a beaten child, Helen Burns dies, Villette is grief-saturated). ⚠ **My nine test prompts were all emotionally neutral, so they could not have surfaced this** — the operator's prompt did, first try. Implication for the regime: "voice transfer" includes tone and subject, so wanting the voice without the gloom is a **corpus-selection or prompt-framing** problem, not a training-length one. Also observed: one seed closed its anecdote and emitted `CHAPTER XIX. THE CHILD'S CLOSET.` — it learned book structure unprompted. Base control on the same prompt went modern and essayistic (a literature lecture on one seed, "took the dog to work and told the employees" on the other), so the shift is the adapter.

- `[2026-09-10]` **R49 rung 2 COMPLETE, and the single-variable carrier effect is clean: 0.6B held-out 3.329 vs 1.7B 3.018, Δ0.311 nats.** Both on the same unwrapped corpus (sha `77f37057b2782e49`), seed 4919, 1 epoch, 159 steps, 5,210,112 tokens — carrier size is the ONLY difference, because the chained 0.6B rerun closed the confound the unwrap opened. ⚠⚠ **DO NOT compare either against the original wrapped-corpus 0.6B run's 3.172 — that comparison is INVALID and reads backwards.** Different corpus means a different held-out set: the wrapped version's 5.7% newline tokens are near-deterministic after a 70-char line, so they *deflate* the loss with cheap wins. Unwrapping removed the easy tokens and raised the number; it is not a regression. ⚠ **Correction to my own earlier claim**: I twice described the 0.6B as "still descending, undertrained" at 3.172 — the series (3.176, 3.173, 3.172, 3.172) shows it FLATTENED. All three runs plateau; one epoch is about right for this corpus, not short. **Three-way eyeball booth** at `http://10.100.10.50:8090/b/babybronte-1p7b/` — measured across 18 samples per arm: curly quotes **1.7B base 0/18 → 1.7B tuned 15/18** (so the shift is the ADAPTER, not the bigger model — the base control is what proves it), worksheet/explainer collapse **3/18 → 0/18**, and **hard-wrap 0.85 → 0.18**, confirming the corpus unwrap carried through into the adapter. **Sense partially returned**: 1.7B produces locally coherent sequential Victorian prose where 0.6B produced word salad ("the door burst through the back window"), but scene-level continuity still breaks mid-passage. ⚠ Curly quotes are slightly LOWER at 1.7B (15/18) than 0.6B (17/18) — plausibly a bigger model's stronger priors resisting the adapter at the same rank; untested, do not treat as established.

- `[2026-09-10]` **R49 rung 2 LAUNCHED: Qwen3-1.7B-Base, 1 epoch, seed 4919, on an UNWRAPPED corpus.** Operator: *"start the 1.7b training."* Live at `gx10:~/r49-runs/h02-1p7b-1ep/`, 159 steps at ~18.7 s/it (~50 min), corpus sha **`77f37057b2782e49`**. A 0.6B rerun on the same unwrapped corpus is **chained behind it** (`chain-0p6b-unwrapped.sh`, gated on the 1.7B actually producing an adapter — a chain that fires on failure turns one lost run into two), ~36 min after. ⚠⚠ **THE CORPUS CHANGED, SO 0.6B-vs-1.7B IS DESCRIPTIVE, NOT ATTRIBUTABLE** until that chained rerun lands: carrier size and corpus typography both moved. *"Did sense come back at 1.7B"* is a within-arm reading and survives it; any between-rung delta does not. **The unwrap:** reflowed 57,430 of 85,380 paragraph blocks, kept 27,950 (verse/headings — verse detected by median line length, lineation preserved, spot-checked and every kept multi-line block sampled was genuinely verse); **0 lines ended in a lone hyphen** so the space-join could not split a word; content identity `" ".join(text.split())` verified byte-identical on all **852 records**, i.e. whitespace-only. Mid-length-line ratio **0.94 → 0.25** (the residual is the preserved verse). ⚠ Concrete cost of the old defect: **5.7% of the training budget was newline tokens** — 5,525,504 → 5,210,112 tokens on the same words. Instruments at `scripts/r49-corpus/{unwrap_corpus,launch-h02-1p7b-1ep,chain-0p6b-unwrapped}`; the original wrapped corpus is untouched so the 0.6B run's pinned sha `3959036cf851bf62` stays reproducible.

- `[2026-09-10]` **BabyBronte H02 adapter: the VOICE transferred, the SENSE did not — operator's read, "it's all nonsense, but it sounds like Brontë's nonsense."** Eyeball A/B (NOT the adjudication; nothing here feeds the frozen rule), 9 arbitrary prompts on a deliberate difficulty gradient × 2 arms × 2 seeds, booth at `http://10.100.10.50:8090/b/babybronte-voice/`. Measured across the 18 pairs: **curly quotes 1/18 base → 18/18 tuned**, **math/worksheet collapse 3/18 base → 0/18 tuned**. Given *"The self-checkout machine refused her coupon"* the base 0.6B produced a **quadratic-formula worksheet**; the tuned arm wrote a clerk refusing a customer in Victorian retrospective first person. This is the expected and informative result for the smallest rung — **voice is separable from coherence at 0.6B**, which is the premise the whole lightweight-adapter regime rests on, and the 1.7B/4B rungs are where sense should return. The 1-epoch loss was still descending at step 169 (undertrained, not overfit), so the incoherence is carrier capacity, not training. ⚠ **Corpus-prep defect found: the tuned output is hard-wrapped at ~70 chars** (median mid-length-line ratio 0.85 vs base 0.00) — the Gutenberg source kept its original line breaks and the adapter learned the typography along with the voice. Unwrap to flowing paragraphs before any real use or the next rung learns it too.

- `[2026-09-10]` **mog-sec (`sec`/`sec-reasoning`, ana-ml2 GPU0 `:8019`) SETTLED at `MOG_MAX_MODEL_LEN=163840` + `MOG_KV_CACHE_MEMORY=17697765376` + `MOG_MAX_NUM_BATCHED_TOKENS=4096` + util 0.50, after FIVE crashes and four wrong fixes.** ⚠⚠ **THE LESSON, and I got it wrong four times running: what the KV pool can HOLD and what the card can PROCESS at depth are DIFFERENT NUMBERS, and the crashes were governed by the second while every fix I made sized the first.** I cut context 420k → 384k → 320k, pinned KV in bytes, and dropped the prefill chunk 16384 → 4096 — each helped and none fixed it, because the pool was never the constraint. ⚠ **I also called it "rare, not chronic" off a `RestartCount=1` and recommended doing nothing; the operator pushed back and it crashed twice more inside ten minutes.** The reproducer came from the operator too — *"loading up the context killed sec again"* — and it is what finally made the failure legible. **Bisected with a NON-REPEATING prompt** (prefix caching would let a repeated one hash to cached blocks and never prefill deep — the probe would pass while proving nothing): **113,247 tok SURVIVED · 200,088 tok SURVIVED · ~285,000 tok KILLED THE ENGINE.** So the ceiling sits between 200k and 285k *with `gen` idle*, and `gen`'s load is an uncontrolled co-tenant variable, hence 163,840 for ~20% margin. ⚠ **The point of the ceiling is the REFUSAL**: verified after, an over-limit request now returns a clean `400 This model's maximum context length is 163840 tokens` in under a second and the seat survives, where before it died and took every in-flight request with it. **A seat that refuses what it cannot serve beats one that dies trying.** Concurrency 1.03x → **2.09x**; 149,073-token request served in 41 s. ⚠ The compose header's *"served at native 262K"* was never actually deliverable on a shared card — it had simply never been exercised at depth. Probe committed at `services/mog-sec-tuning/deep_ctx_probe.py`; backups `.env.bak-{util052,384k,batched16384}-20260910`.
- `[2026-09-10]` ⚠ **Near-miss on measurement discipline, worth keeping as a specimen.** The crash window logged `Avg Draft acceptance rate: 17.6%` and per-position rates of 0.049/0.024/0.015 for draft positions 5–7, which reads as an obvious "cut `num_speculative_tokens` 7 → 3, it is buying nothing." Across **180 samples** of the same counter over the container's life the real distribution is **median acceptance length 3.12 of 7 (range 1.83–6.75)** and **median draft acceptance 30.4% (range 11.9–82.1%)** — the crash window was near the *minimum*, not the norm, and cutting to 3 would cap the workloads that were accepting nearly the full 7-wide draft. **The n=1 window pointed the opposite way from the n=180 distribution.** Same session that wrote "a positive control is only worth what it can distinguish"; the lesson generalises to log lines.

- `[2026-09-10]` **R49 carrier SETTLED on dense `Qwen3-{0.6,1.7,4}B-Base`, overriding H02's own pin — the newest carrier was the SLOW one.** Dense 4.089 B trains 33% faster than hybrid 0.765 B; no fused SSM kernel installed. D1–D3 built, 1-epoch pilot beats the 3-epoch by 0.21 nats held-out. → `persistent-memory.d/2026-09-10-r49-babybronte-d1-d3-and-the-1-epoch-pilot.md`
- `[2026-09-10]` **R49 adjudication routed to infra-ops entirely** (operator, relayed by brokkr: *"leave babybronte to infra — concentrate on r50 and the memory mechanism"*). brokkr handed over the Delta instrument and stepped off. ⚠ I now grade my own run; brokkr's decision rule is **ratified verbatim and frozen before any adapted text existed** and must not be amended after seeing numbers. Their controls: real Charlotte 1.65–2.17, **Anne at 2.374** — so the absolute band decides, never `nearest`.
- `[2026-09-10]` **MeroMero A4B swapped onto the `erp-seat` seat as `char-rp-fast`; `Pfish-6` alias removed.** The A4B's FIRST quant used the dense recipe and 4-bit-quantized all 30 MoE routers — it passed its healthcheck and answered every request with the full token count decoding to the empty string, NaN logits the only tell. Re-quantized with the MoE recipe; live and verified (prose, vision, tool call, finite logprobs). Durable lesson: **a positive control must match the ARCHITECTURE CLASS** — the broken A4B was diffed against a good *dense* quant, which has no routers, so the clean result was meaningless. → playbook §3.15, §4.4
- `[2026-09-10]` **MeroMero: BOTH quants landed in-house at W4A16 — A4B first try, v2 dense on attempt 5.** Published quants are all W4A4 (our measured long-context collapse) or nonexistent for v2. Operator: *"pull both ablits bf16, run our own quant."* The durable lesson is **§3.17**: `pip install llmcompressor` silently pins transformers down a version, so attempt 4's error was a moved toolchain, not the malformed upload it looked like — a known-good positive control is what told them apart. Serve test still owed. → `persistent-memory.d/2026-09-10-meromero-quants-and-the-pinned-transformers-trap.md`
- `[2026-09-10]` **althing 3.6.2 deployed — post office + both heralds — and the fleet has TWO herald nodes, not seven.** Ask the post office's `nodes` table, not the box inventory. Cost a self-inflicted ~12 min bus outage. → `persistent-memory.d/2026-09-10-althing-362-rollout.md`
- `[2026-09-10]` **A grep over a log that records your greps counts itself.** I reported forseti's drop defect as reproducing here with 3 drops in 21 s; the session had **zero**. Searching transcripts writes the search term into them. Filter by `"type":"system"` provenance, never content. Generalises to any instrument that can see itself. Auto-memory `feedback_grep_over_a_log_that_records_your_greps`.
- `[2026-09-10]` **Operator-directed purges: 466 GB (qwopus + huihui 122B bf16) and 107.8 GB Docker on ana-ml2.** Serving/rollback artifacts and qwopus's MTP head verified intact after. ⚠ `/tank` is OUTSIDE restic, so both were final.
- `[2026-09-10]` **ana-docker disk pressure repaired: root 84% → 51%, 115 GiB free.** Gitea/Vaultwarden backups repaired and restored from Restic `2ec5a37c`; 101 stale dumps removed; hourly named-builder cache pruning installed. → `persistent-memory.d/2026-09-10-ana-docker-disk-repair.md`

- `[2026-09-09]` **Run 7 PURGED; pfi-gx10 declared an experimental/TRAINING box with no serving seat** — operator: *"gx10 is an experimental box, primarily for training … run 7 can be purged … no new run, we'll roll with run 6 for now."* ~139 GiB reclaimed across both boxes; the 315 MB adapter + provenance KEPT as the only non-reproducible piece. `Pfish-6` on ana-ml2 :8021 is the sole standing seat.

- `[2026-09-09]` **Run 7 RETIRED; run 6 declared `Pfish-6` and is the standing seat** — NVFP4 quant on ana-ml2 :8021 AND gx10 :8098 at 262k ctx, gateway alias `trial` → `Pfish-6`, max-num-seqs 8→32 (2,170 tok/s at n=16, 3.2x the old ceiling). ⚠ ana-ml2 measured **4.1x FASTER than the GX10** on the same artifact — the reverse of the expectation. → `persistent-memory.d/2026-09-09-run7-retired-pfish6.md`

- `[2026-09-09]` **The run-7 CSAM gate failure was a DETECTOR BUG** — HARD `child_term` matched the ADJECTIVE "minor"; operator-diagnosed, fixed `cc42d76` (nominal-use-only, selftest 24/24), retention wired so a hit can finally be adjudicated. ⚠ The lesson is mine: rigor downstream of an unexamined premise is not rigor. → `persistent-memory.d/2026-09-09-csam-detector-bug.md`

- `[2026-09-09]` **⚠ ERP RUN 7 FAILED THE SAFETY GATE — both seats stopped, nothing deleted.** brokkr's CSAM drift detector fired on the TUNED arm during the refusal leg and aborted fail-closed (`level=hit`, `counts=1/0/3`, two HARD `child_term ^ act` flags). Base arm NOT implicated (clean earlier the same evening); the merge check — a sampled target confirmed CHANGED — is why this reads as ONE explanation, the tune, not a base wearing a different name. **Neither brokkr nor I re-ran the probe or opened the flagged generations** (a second run is not a second opinion; reading answers no question that changes the outcome). brokkr also left the length verdict UNSET on purpose: settling one on a rejected artifact hands a dead tune a result line that outlives its context. Actions: `erp-tune-v7` on gx10:8098 stopped 17:42; **the `trial` NVFP4 seat on ana-ml2:8021 stopped 17:43 — MY CALL, reversible in one command**, because the operator's "unrated on every safety axis" ruling was honest while no rating existed and one now exists as a fail on the same tune (quantization does not launder behaviour), and it sat on the SHARED-KEY gateway ~15:30–17:43. All artifacts preserved (adapter 315 MB, merged-run07 49 GiB, v7-nvfp4a16 16 GiB, v7-bf16 49 GiB); v6 still on disk as the obvious rollback. Independent of safety the run was already poor: primary FLAT (69 → 70.5, +2, flat at BOTH the 12-word threshold and the 20/60 cue-probe floor), both diversity families reduced past their floors, long-context coherence 1.0 → 0.875 on its must-not-harm bar, unanswerable control held at 1.0 so the instrument was valid. **INCIDENT CLOSED 2026-09-09 ~18:20 PT, both sides.** `trial` alias REMOVED from `stacks/litellm/conf/config.yaml` (commented, not deleted — restoring is uncommenting) and verified gone by both parties **at the routing layer, not just the model list**: a call returns 400 `Invalid model name` and generates nothing. ⚠ Alias-present-with-backend-down is a DIFFERENT and worse state than alias-removed — it re-arms silently under whatever is served on that port next. **EXPOSURE QUANTIFIED** from the gateway spend DB, filtered on the ARTIFACT (`model='hosted_vllm/erp-tune-v7-nvfp4a16'`) not the alias: `all-agents-local` 68 calls / 10,073 generated (my own throughput benchmarks), **`open-webui-esh` 9 calls / 50,604 prompt / 2,793 generated, 15:40–16:51 PT — the operator's OWN Open WebUI session, and those outputs are in its history**. NO peer agent called it, so nothing landed in another project's artifacts. Nobody read the flagged generations or that session. ⚠ **Counting by the ALIAS would have returned 363 vs 77 — 4.7x inflation of his own exposure**, because the alias had carried v5 and v6 earlier the same day (→ ops-lessons `b135adc`). ⚠ **I made THREE reporting errors during the incident**, all false-reassurance, all the unfalsifiable-at-write-time class (two fabricated commit SHAs, one past-tense claim sent before the action) → auto-memory `feedback_unfalsifiable_at_write_time`; brokkr independently verified my reports for the remainder, which was correct. **⭐ DECISION BRIEF FOR THE OPERATOR: http://10.100.10.50:8090/b/run07-decisions/** (kept booth, 5-question inline ask; answers land in `~/booth-data/run07-decisions/decisions.answer.json` — read it with `booth answer run07-decisions decisions`). **Open for the operator:** disposition of the adapter + the run-7 corpus slice; whether `trial` returns and pointing at what (v6 still on disk, passed by his own adjudication); whether the opening-split idea gets a fresh run; whether my reporting errors change how he wants incident reports handled.

- `[2026-09-09]` **run 7 quantized NVFP4A16 and serving as `trial`** — 49 GiB bf16 relayed gx10→ana-ml2 (16 min, 53 MB/s), quant 49→16 GiB via `services/erp-seat-quant/run_quant_erp_v7.sh` (dry-run gate passed: 11,725 targets / 11,520 experts, routers+vision BF16), seat on `:8021` under its TRUE name `erp-tune-v7-nvfp4a16`, LiteLLM `trial` repointed (config-file alias — `/model/update` REFUSES a config model, must edit `stacks/litellm/conf/config.yaml` + restart). Rollback: v6 artifact on disk + `/tmp/erp-seat-env.v6.bak`. ⚠ **`no direct path` was WRONG** — gx10↔ana-ml2 ROUTING is fine both ways; neither box holds a private key (only `authorized_keys`), so neither can *initiate*. `ssh -A` agent forwarding from nh3-dev gives a genuine direct path, verified. The relay costs nothing here anyway: both gx10 and nh3-dev are at NH3, so the WAN hop happens once either way.

- `[2026-09-09]` **Booth: partial ask answers are legal** (v0.1.15) — operator: the form failed when a question was left blank. `required` dropped from the radios; answered questions recorded, blanks land in `unanswered`, `complete` says whether the set is finished; refused only when there is no pick anywhere AND no notes. Reading sessions must check `complete`.

- `[2026-09-09]` **ERP run 7 COMPLETE and the base arm is serving.** 542/542 steps in 14h17m on pfi-gx10, adapter 13:23 PT, `train_loss` 3.205 / low 2.799, merge verified a sampled target actually changed (the silent-no-op check). `erp-seat-base-ara` up on `10.100.50.60:8098` for brokkr's floors, `erp-tune-v7` merged and staged pending his cue; Miranda notified for the operator. Runbook `docs/runbooks/gx10-run-07.md`.

- `[2026-09-09]` **Booth asks render INLINE in a custom report, placed by the author** (v0.1.14) — operator ruling: *"the asks should be inline with the artifacts, not on a separate page."* Placeholders `data-booth-ask="<stem>"` / `"<stem>:<key>"` / `data-booth-ask-submit`, plus `<!-- booth:ask … -->`; per-question fragments bind to ONE form via the HTML5 `form=` attribute so a four-voice audition submits every pick in a single POST. ⚠ The placeholder must sit OUTSIDE any grid/flex parent or it becomes a cell (measured on `redo-anchors`: a 224 px sixth grid cell). Unplaced questions + a missing submit block are appended, so a partially marked-up page can never yield an unsubmittable 400 — a test caught that as a real drop. `redo-anchors/index.html` was hand-marked-up on the LIVE copy; tts-dev told to move it into the generator or a regeneration loses it.

- `[2026-09-09]` **The Booth gained an ASKS primitive** (v0.1.12): a session drops `<stem>.ask.json` in a booth, the operator answers a radio form + notes in the browser, the pick lands as `<stem>.answer.json` the session reads (`booth ask|asks|answer --wait`). Multi-question form via a `questions` list. ⚠ Two defects found and fixed the same day: a booth serving its OWN `index.html` never rendered the panel (verbatim path returns early) → amber chip + standalone `/b/<name>/asks` page; and single-ask `title` was silently dropped. The `booth` CLI was ALSO not on PATH anywhere despite the global link-board convention telling every session to run it → symlinked to `~/.local/bin`. Global `CLAUDE.md` now teaches the primitive.

- `[2026-09-09]` **ana-ml2 pool actions LANDED (scrub 0 errors in 1h33 → `zpool clear`; nvme-cli + full-drive SMART table; zroot 91→73% via dangling-image + builder prune, tagged rollback seats kept) — and the missed-scrub mystery SOLVED: nvme7 (slot 0-5, `S47VNY0K600221`) was absent from every boot 04-23→09-05, tank was raidz2-DEGRADED for 4½ months, Debian's scrub/trim cron only touches `ONLINE` pools, and ZED's alert went to a root mailbox with no MTA.** nvme7's 2084 media errors did not move across the scrub → historical, keep + watch. Playbook `playbooks/ana-ml2-pool-health.yaml`; inventory in `servers/ana-ml2/README.md`. → `persistent-memory.d/2026-09-09-ana-ml2-pool-actions-done.md`

- `[2026-09-09]` **ana-ml2 `tank`: 2 CKSUM errors on nvme7n1 after a boot-time resilver, NO scrub since 04-12, zroot 91% — three actions DEFERRED to a clean-context session** (scrub → nvme-cli SMART → docker prune), operator ruling "we'll do all 3 on clean context"; tracked at commit `3e18a04` + the post-clear handoff. ESH 10G links measured clean (fiber run live on UDM SFP+2 ↔ USW-Pro-XG Media). → `persistent-memory.d/2026-09-09-ana-ml2-pool-actions-deferred.md`

- `[2026-09-08]` **ana-ml2 mesh return routes PERSISTED** as `/etc/network/if-up.d/mesh-routes` (Debian 13 ifupdown, no netplan) via `playbooks/ana-ml2-mesh-routes.yaml` (elway, verified) — operator: "persist the routes". Hook not yet exercised by a real reboot. `f923d6a`.

- `[2026-09-08]` **ERP run 7 LAUNCHED on pfi-gx10 23:06 PT** under `operator-2026-09-08-rnd-run7` — opening-split slot + mask union; free check passed with two explained deltas; first launch died on a missing recipe (zsh quoting). → `persistent-memory.d/2026-09-08-erp-run7-launched.md`

- `[2026-09-08]` **erp-tune-v6-nvfp4a16 quantized (data-free W4A16, ~90 s) and serving on ana-ml2 :8021; `trial` aliased to it ("no gate"); tool calling fixed where it can be** — `tool_choice:none` flag; forced tool_choice is prompt-driven on Gemma-4 by vLLM design, nightly `311b3513` raises it 1/9→6/9; json_schema is the deterministic path. → `persistent-memory.d/2026-09-08-erp-seat-nvfp4-trial-and-toolcalling.md`

- `[2026-09-08]` **Run-6 gate: CSAM level=review soft trip HALTED it; operator adjudicated GO ("baby is a pet name"); TRANSFERRED finalized without the tuned refusal leg; k=25 legs cut** — the flagged text exists nowhere by design. → `persistent-memory.d/2026-09-08-run6-gate-csam-adjudication.md`

- `[2026-09-08]` **ESH static-WAN follow-ups landed (FortiGate trusthost3, esh-ana IPsec rebind, UDP 41641 → mesh direct); YTVC chased back up (nh3-scale SOCKS, stale yt-dlp layer, punkt_tab) and v0.3.6 CrisperWhisper deployed; gitea webhook repointed off the dead wg0 IP with the HMAC secret re-applied.** → `persistent-memory.d/2026-09-08-esh-static-wan-followups-and-ytvc.md`

- `[2026-09-08]` **ERP run 5 = RESCUED (landmark R49.5)** — first capability-gate pass in the ERP-seat line; the 3.46%-loss dependency-forcing slot (GovReport+QMSum) broke the coupling runs 3c/4 couldn't. Seat `erp-tune-v5` served on gx10:8098, `trial` alias repointed 3c→v5. → `persistent-memory.d/2026-09-08-run5-rescued.md`

- `[2026-09-08]` **R47 base settled from bytes = STOCK `google/gemma-4-26B-A4B-it`** — three-way sha match (local == HF etag == stock LFS oid; commit `4d7ae498` == stock HEAD); the `-heretic` label is a naming error, all runs trained from stock. Accept-vs-swap now evidenced. → `persistent-memory.d/2026-09-08-base-provenance-stock.md`

- `[2026-09-08]` **yt-voice-clipper back UP** — dead since the 09-06 danted retirement (every job failed at yt-dlp, bot-gated on the Irvine datacenter IP). Fix: danted on **nh3-scale** (CT107) at `socks5h://100.64.0.1:1080`, fleet-ACL'd, residential egress 70.230.226.88 measured; `YTVC_PROXY` repointed, worker recreated, end-to-end job DONE with positive (proxied) + negative (direct = bot-gate) controls. Homepage card href/siteMonitor → `irv-ml1.nh3.internal:8000` (was dead wg0 IP). Then a SECOND fault: full downloads 403'd through the proxy (cookies irrelevant) = stale yt-dlp 2026.07.04 from a cached Dockerfile layer → `compose build --no-cache api` (2026.08.19), which dragged in a whisperx/nltk that needs `punkt_tab` → staged on the data volume + `NLTK_DATA` in the override. Operator's video x7kWJojf1MI → done, 8 clips. yt-voice-clipper-dev shipped both Dockerfile fixes + **CrisperWhisper 2.0 (v0.3.6, `b62849d`) — deployed and verified (12 clips, [UM]/[UH] tags)**. ⚠ The gitea push webhook had been targeting the dead wg0 IP since 09-06 (never fired) → repointed to `10.6.110.50:9008` with the HMAC secret re-applied; deploy script passes `YTDLP_REFRESH`. Script `scripts/setup-nh3-scale-socks-egress.sh`. → auto-memory `reference_nh3_egress_proxy`, `reference_ytvc_autodeploy`.

- `[2026-09-08]` **ESH WAN static `128.177.138.182/30` (gw .181) is LIVE** — the Cityside /30 that was 'not provisioned' on 09-04 now carries traffic; egress verified from esh-docker-vm. CGNAT at ESH is over. Added to the crowdsec `esh` allowlist. All three follow-ups LANDED same day: FortiGate trusthost3 → the static (login from ESH verified), dormant esh-ana IPsec rebound to wan1/static, UDP 41641 forward → esh-scale now peers DIRECT (was DERP).

- `[2026-09-08]` **ERP run 6 COMPLETE** — 524/524, train_loss 3.259 (run 5: 3.235). Merged; base seat `erp-seat-base-ara` serving on gx10:8098 for floors, awaiting brokkr's swap cue → `erp-tune-v6`. ⚠ abliterated repo lacks `processor_config.json` — stock's carried in (32bdf45d). Miranda informed.

- `[2026-09-08]` **ERP run 6 LAUNCHED on pfi-gx10 on the jenerallee78 ARA-abliterated base** (index `33c59654…`, 32/32 shards byte-verified vs brokkr pins, stock tokenizer set installed over the repo's 256-token-truncating one, run-5 recipe byte-held, free check exact). Operator's direct grant `operator-2026-09-08-rnd-run6`; run-5 seat unloaded (`trial` dark). Gate names: `erp-seat-base-ara` / `erp-tune-v6`. → `docs/runbooks/gx10-run-06.md`, commit `3fec668`.

- `[2026-09-08]` **Miranda = operator's chief of staff, may relay his directives** — added to user-level `~/.claude/CLAUDE.md` (dotfiles `7134a22`) as the named exception to the no-relayed-auth rule (unidentified peer relays still excluded); material-consequence calls she relays stay the operator's own.

- `[2026-09-08]` **Fleet fixes shipped** — WhereTF Homepage card + DNS (`4506ef6`); ext-tts LiteLLM alias → `irv-ml1.nh3.internal` (DB `/model/update` + `extra_hosts`, `957c8f1`); the 09-06 irv-ml1 stale-IP trail repointed across 25 composes + services.yaml + ssh-target → DNS name (`e0d1c44`); Homepage `/api/services` outage fixed — ana-ml2 discovery via a socat proxy on ana-docker (`stacks/ana-ml2-proxy`, `913d2d2`, reversible).

- `[2026-09-07]` **Fleet internal TLS pattern shipped** — caddy (cloudflare-plugin build, `~/.local/bin/caddy-cf`, `fleet-tls-caddy.service`) on nh3-dev is the wildcard cert authority: publicly-trusted LE `*.nh3.phasefinal.com` via Cloudflare DNS-01, resolved internally by an AdGuard split-horizon rewrite (all 3 resolvers). `talk` self-terminates on :8092 with the trusted cert (operator's in-container-TLS ruling), renewal auto-synced by `fleet-tls-cert-check.timer`. Interstitial gone; secure-context+AudioWorklet verified via headless Chromium. Pattern + foot-guns (restart-disrupts-inflight → clients need retry; wildcard = name-only, never IP) → auto-memory `reference_fleet_internal_tls_pattern`.

- `[2026-09-07]` **cc-channel registered for this infra-ops session's wake** — `althing-route` cc route → the CC session's `$XDG_RUNTIME_DIR/cc-socks/<pid>.sock`; herald pokes the socket directly at a turn boundary. Replaces the FIFO/poll waiter that Claude Code 2.1.257 kept killing while the seat sat in `shell`. Session-local — re-declare per session.

- `[2026-09-07]` **irv-ml1 /mnt/smithy remount fixed post-cutover** — export allowed `10.0.0.0/8` (old wg0) but not the mesh `100.64.0.0/10` irv-ml1 now sources from → all-uid "permission denied"; added the mesh range to the nh3-nas smithy export + remounted (clientaddr now 100.64.0.6). nh3-nas admin = `infra-ops` PASSWORD auth (vault `nh3-nas/infra-ops-password`), sudo ALL, SFTP subsystem OFF. → auto-memory `reference_irv_ml1_gpu_r14` (corrected).

- `[2026-09-07]` **irv-ml1.nh3.internal DNS repointed** to the live Irvine LAN IP `10.6.110.50` (was the dead wg0 `10.100.79.3`); CLAUDE.md fleet-row + placement-rule updated to mesh reality. commit `0336e03`.

- `[2026-09-07]` **Subnet routers excluded from vzdump fleet-wide** (ana-scale 114/pfi-pve, nh3-scale 107/nh3-pve, esh-scale 108/esh-pve) so a hung backup can't blackhole a site; nh3-headscale (106, control plane) KEPT; ESH backup moved 02:15→03:30. Root cause of this morning's ESH outage: an overnight vzdump left CT108 (esh-scale) locked → whole site dark. → auto-memory `feedback_esh_backup_window_0330`.

- `[2026-09-07]` **Booth link board: pin/favorite + multi-select delete + newest-first** (booth-v0.1.8, commit `76fdf45`, tag `booth-v0.1.8`) — pins in a `.pins` sidecar (content-ids), one `<form>` + `formaction` buttons so ×/★/bulk-delete all degrade with JS off.

- `[2026-09-06]` **Headscale cutover COMPLETE — all three site-pairs on the mesh; Site Magic + both IPsec tunnels DORMANT.** Operator disabled Site Magic (UI); NH3↔ESH re-homed to a direct 8ms path. Exit nodes advertised at all three sites (multi-location egress proxy) with source preservation kept via a selective-masquerade rule (NoSNAT + `mesh-exit-masq.service` per router). Throughput 761/464 Mb/s vs old 250 IPsec. ⚠ FortiGate WAN-SSH left open (temp, scoped NH3+ESH). Method: disable tunnel FIRST then add mesh route. → `persistent-memory.d/2026-09-06-headscale-cutover.md`

- `[2026-09-06]` **Headscale overlay mesh: control plane live at `headscale.phasefinal.com` (CT 106 nh3-pve) + subnet routers nh3-scale/esh-scale/ana-scale serving their /16s; nh3-dev enrolled. NOT cut over — Site Magic + IPsec still carry site-to-site.** ⚠ accept-routes-before-return-path black-holed nh3-dev's LAN for a minute. infra-ops user added on all four PVE hosts. → `persistent-memory.d/2026-09-06-headscale-mesh-phase1.md`

- `[2026-09-06]` **pfi-pve NASPool REBUILT as six-wide raidz2 after a backplane fault killed bays 9/10** (Route C hybrid, operator-directed): parked 1.65T on ospool, destroyed, recreated, restored, backup tier back 04:03Z; guests never stopped (ALL boot disks are on ospool — the prior brief had this wrong). Legacy vzdump pruned to newest-per-guest by omission. OPEN: destroy `ospool/naspool-evac` after scrub + one backup cycle; backplane swap next visit; PSU1 still dead. → `persistent-memory.d/2026-09-06-pfi-pve-naspool-raidz2-rebuild.md`

- `[2026-09-05]` **A peer's "2.7x serving-stack effect" was a coin flip — the operator rejected it on instinct and the arithmetic backed him.** Each floor was `|b0-b1|` from n=2; the ratio is half-Cauchy, P=0.452. ⚠ The disconfirming evidence sat in brokkr's own sentence, and he named *why*: the claim was his and flattering. → `persistent-memory.d/2026-09-05-floor-claim-n2-retraction.md`

- `[2026-09-05]` **vLLM RUNS on sm_121 — the blocker was `ninja` off PATH, not the silicon** — and run 4 launched after two peer artifacts were rejected by reading the harness rather than accepting a "confirm this". ⚠ I declined to fill in missing `root_sha256` values I knew, because supplying both sides of a check makes it inert. → `persistent-memory.d/2026-09-05-vllm-on-sm121-and-run4.md`

- `[2026-09-04]` **ERP run 3c trained and GATED — the 20x LR cut erased the diversity gain and did NOT remove the −40pp selfharm regression.** LR-robust, so it comes from corpus content. CSAM clean on all three arms. ⚠ A pooled preserve-list test cannot see a single-axis collapse. → `persistent-memory.d/2026-09-04-run3c-trained-and-gated.md`

- `[2026-09-04]` **`gen` moved to ana-ml2 GPU0 to stop vllm-embed OOM-crashing (7 restarts) — and I sized it against vLLM's declared budget, not its runtime footprint.** Cost: gen KV down to 1.02x concurrency at 262K. → `persistent-memory.d/2026-09-04-ana-ml2-gpu-rebalance.md`

- `[2026-09-04]` **SMB account `dsp` created + vaulted for the Windows AudioGridder box — and esh-nas turns out to be wide open.** Twelve NFS exports rw to `10.0.0.0/8`, guest-writable SMB. → `persistent-memory.d/2026-09-04-esh-nas-smb-and-exposure.md`

- `[2026-09-04]` **SLZB-MR1U Zigbee coordinator moved to esh-iot (VLAN 90) at `10.0.90.10`, DHCP-reserved, DNS'd, handed to ha-dev.** ⚠ Home Assistant cannot resolve `.internal` at all (Docker's 127.0.0.11 upstream excludes the fleet AdGuard) — pre-existing; ha-dev declined the fix. Runbook `docs/runbooks/slzb-mr1u-zigbee-coordinator.md`, commits `fed29be`/`0bbdaf9`.

- `[2026-09-03]` **Run 3c is STAGED on pfi-gx10 and deliberately NOT launched** — the launch is a 13.3 h commitment and the operator stood this port down once already. Base shards AND the encoded corpus sha256-verified identical to ana-ml2's, so the transformers 5.15.1→5.16.1 / x86-64→aarch64 delta is *measured* inert. ⚠ The encode-cache FILENAME differs by design (`base_model_path` is in the key) — input hash, not output. ⚠ Tripped the `pkill -f` ssh self-match again; the launcher guards on a pidfile because of it. → `persistent-memory.d/2026-09-03-gx10-run3c-staged.md`

- `[2026-09-03]` **SearXNG returned ZERO results for every query while reporting `healthy` for 7 days — 4.5 months stale.** Moved to nh3-docker (residential egress beats the colo's CAPTCHA-gated 38.120.12.42), updated, and exposed to every CC session as the user-scope `web_search` MCP tool. ⚠ `/healthz` cannot tell you whether search works. → `persistent-memory.d/2026-09-03-searxng-nh3-move.md`

- `[2026-09-03]` **pfi-gx10 racked: VLAN 50 via a DHCP RESERVATION on the UDM, not a host static — operator ruling, so the box stays portable.** ⚠ The racked port arrived on the NATIVE VLAN; ⚠ `port_overrides` is a whole-array PUT; ⚠ prove inter-VLAN routing with `ping -I <wired>` BEFORE downing the Wi-Fi escape hatch. Now single-path. → `persistent-memory.d/2026-09-03-gx10-rack-network.md`

- `[2026-09-03]` **Three Macs onboarded (mini / Air / Studio) with infra-ops, NOPASSWD sudo, rotated+vaulted passwords and `dsh` on device-scoped keys — and the fourth is `scripts/provision-mac-dsh.sh`, not a fourth hand-run.** ⚠ `sudo -u` keeps the CALLER's `$HOME` and nearly wiped a working install; ⚠ a wrong USERNAME is indistinguishable from a wrong password. → `persistent-memory.d/2026-09-03-mac-fleet-dsh.md`

- `[2026-09-03]` **nh3-dev wedged for ~40 min and it was the BACKUP, not the disk — a stalled cross-site vzdump holding every guest write via `copy-before-write`.** Symptoms screamed dying disk: 45 writes in flight completing zero, jbd2 + flush kworkers in D state 33 min, io pressure full 96%, load 26, `virtio_ring` in the stack. ⚠ **The discriminator was the ABSENCE of errors** — no SCSI/ATA/IO errors, rpool ONLINE 21%, guest fs 79%, memory fine, and **Dirty only 3.8 MB** (so nothing backed up in page cache; it was stuck BELOW the block layer). ⚠ **The hypervisor was IDLE** — load 0.63, io pressure 0.00, zpool ~0 writes: nothing was reaching the disk because the filter held it. Cause: `vzdump` of VM 102 → **pbs-ana** did 1% at 64 MiB/s then collapsed to **1.4 MiB/s for 35 min**; Proxmox backups interpose a `copy-before-write` filter, so every guest write queues behind the backup's copy-out. FIX = cancel the task (`pvesh delete /nodes/localhost/tasks/<UPID>`); filter detached, inflight 45→0, D-states gone, 191 MB/s dsync restored. ⚠ **`fleecing 0` on the job is why a slow TARGET can stall a GUEST** — fleecing routes copy-before-write to a fast local image instead. Job = `backup-5d8f1221-8f71`, **daily 21:00, `all 1`, storage pbs-ana** → recurs nightly until changed. A prior run of this VM managed 941 MiB/s read, so 1.4 MiB/s is degradation, not normal. → `docs/runbooks/nh3-dev-io-stall.md`

- `[2026-09-02]` **althing deploy is SIX surfaces, and #6 is outside the althing repo: `~/.claude/settings.json` `crossSessionInbound: "accept"`.** Without it Claude Code HOLDS every cc poke — it auto-delivers only when the sender's permission-mode class matches, and the herald is a daemon that asserts none, so the notice goes to a human watching the pane instead of to the session. ⚠ **The seat reports declared, reachable and green throughout** — same failure shape as the SessionStart hook that was never deployed. Set on nh3-dev by forseti 09:28 with operator authorization (diff verified: one key, backup at `/tmp/settings.json.bak-20260902T092829`). Operator's reasoning: the herald reaches only local seats and a **pane poke already types+Enters into a session**, so the socket channel is strictly NARROWER than what it replaces — stating the existing trust boundary, not widening it. Cost without it is first-contact-only (in-memory correspondent record), not per-message. ⚠ **No attestation exists for the herald to send** — CC identifies a sender by verified pid against the session registry and reads that session's LIVE runtime mode; a daemon is not in it, and `from_mode` on a `type:"user"` frame is never consulted. **`deploy-althing.sh` reports surface 6 and deliberately never SETS it** — a deploy script that edits its own trust settings grants itself trust. → `docs/runbooks/althing-deploy.md`

- `[2026-09-02]` **`vastblue` gitea org created (id 8, private, owner `vh`) with empty repo `vastblue/platform`** — third entity namespace alongside `corviduo` and `pfi`; most repos still live under `vh/`. Home of VastBlueDocumentAI + the anchor healthcare-billing SPA (signed 3-yr client contract). **Org scope was the decision**: org-level runner registration and secrets are inherited free by the DocumentAI repo when it splits out, and that is the only binding expensive to retrofit. Deliberately NOT set: org runner (instance-wide `ana-docker-runner` already serves it; org scope is for the DEDICATED runner, deferred to U10) and org secrets (none exist yet; a guessed secret looks bound). ⚠ **Dedicated runner is gated on the first client-premises release cut**, not on the first green pipeline — the risk is another repo's CI sharing a root-level daemon with a build that ships to a healthcare client, see the runner entry above. Push needs no credential: vastblue-dev is on nh3-dev and git-SSH there auths as `vh`. → `stacks/gitea-runner/README.md`

- `[2026-09-02]` **althing 3.3.0 deployed — the `cc` channel, and a plugin-cache false green.** CC seats are now poked over their own message socket (`$XDG_RUNTIME_DIR/cc-socks/<pid>.sock`) instead of by typing into the pane: no process to reap, nothing near the input line. infra-ops moved to `channel=cc`; **the dwarves stay on pane and their guard-4 exposure is UNCHANGED** (`declare` prefers cc, falls back). ⚠ **An undocumented Claude Code interface, taken deliberately** (operator: the FIFO poker was also an unsanctioned hack — a better instance of a class we already had). Break mode = seat goes pull-only with a logged reason, mail still held. ⚠ **`claude plugin update` matches on the plugin VERSION and declines a content-only change** — 3.3.0 edited plugin content at an unchanged 0.1.1, so the CC cache stayed stale while every version check reported success (delta was docs-only, harmless this time). `deploy-althing.sh` now diffs marketplace vs live cache. ⚠ Ordering: **herald restart BEFORE anything declares cc**, or the seat goes silently pull-only. ⚠ This box was at **3.2.4, not 3.2.5** — rollback target here is 3.2.4. Follow-on 3.3.1: **the statusline bell measured a MECHANISM, not the property** — it read `wake-listener-<handle>.lock`, so a cc seat renders 🔕 while push/reachable. Both copies now ask the post office (`reachable` from the status payload) and add 📵 for an outage. ⚠ **TWO COPIES of that script now exist** — `scripts/claude-statusline-command.sh` here (the operator's wired one) and althing's `plugin/scripts/statusline.sh` — independently fixed to the same shape; a drift surface with a countdown, convergence not yet raised with the operator. → `docs/runbooks/althing-deploy.md`

- `[2026-09-02]` **Every CI job on the shared `pfi-fleet` runner is root on ana-docker — and `container.valid_volumes: []` does NOT prevent it.** Measured: a job container is uid 0, `/var/run/docker.sock` is mounted by act_runner independently of that list, `docker ps` returns all 49 host containers (gitea itself, synapse, phasefinal-web, adguardhome), `docker compose v2.33.0` on PATH. ⚠ **LOAD-BEARING** — `vh/Worldtree`, `vh/soong-lab`, `vh/skaldsong`, `vh/wt-matrix-bridge` all drive buildx through that socket, so it cannot simply be closed; **isolate sensitive builds onto a dedicated runner instead.** Also measured the same night: `services:` containers work (Postgres 16), and **full-URL `uses: https://gitea.phasefinal.com/actions/checkout@v4` resolves from the local mirrors** — the un-parked half of the github-independence work, needing neither `DEFAULT_ACTIONS_URL=self` nor the act_runner auth path that blocked it on 2026-08-05. Prompted by vastblue-dev's CI-posture question for a client-funded healthcare repo. → `stacks/gitea-runner/README.md`

- `[2026-09-02]` **pfi-gx10 BASELINED: 79.36 s/it median on the run-3c shape, and the training stack works on aarch64/sm_121.** Median across 10 timed steps, 0.19% spread, **peak 75.1 / 121.6 GiB — 46 GiB spare**, `attn_resolved: flex_attention`. **6× slower than ana-ml2 where compute predicts 2.7×** → likely memory-bandwidth-bound; **capacity box, not throughput box.** Ruled **bare metal, not Proxmox** (no aarch64 PVE; the GPU is on-package and cache-coherent, so passthrough would partition the unified memory that is the whole point). ⚠ `sm_121` is NOT in torch's arch list — everything JITs from sm_120 PTX, so **warm up before timing anything** (an unwarmed bench read 27 TFLOP/s against a true 93). → `persistent-memory.d/2026-09-01-pfi-gx10-onboarding.md`

- `[2026-09-02]` **I priced a failure in the units I happened to be measuring — operator overruled me, correctly.** Recommended run 3c to ana-ml2 by costing a breaker trip as "≤50 steps ≈ 11 min of recompute". It is a **40-minute drive each way** with **13 Anaheim hosts dark, three of them SureFire CLIENT machines**. `save_steps` caps the recompute, never the outage. ⚠ **General form: a metric in hand will volunteer itself as the unit of risk.** → `persistent-memory.d/2026-09-01-pfi-gx10-onboarding.md`

- `[2026-09-02]` **althing 3.2.0→3.2.4 deployed, and ALTHING DEPLOY IS FOUR SURFACES not three.** The fourth (plugin) had no runbook step and was frozen at Aug 28 — **missing the SessionStart/SessionEnd hooks and `pane-route.sh` entirely**, so "CC seats re-declare automatically" was never true here. Now one command (`scripts/deploy-althing.sh`). ⚠ `uv tool install .` **without `--force` is a silent no-op**. ⚠ **A missing deploy surface presents as "the migration needs manual work", not as an error.** → `persistent-memory.d/2026-09-01-althing-320-deploy.md`

- `[2026-09-01]` **irv-ml1 GPU resident map, and `dots-tts` holds 14,430 MiB against a ~6 GB baseline** — tts-dev's prompt-feature cache, capped at 32 entries after two incidents; the cap still permits a long way of growth. 3090 at 76% behind a warn-only watchdog. ⚠ Restates the GPU-ordering foot-gun: `device_ids: ["1"]` is the A6000 in a container, but a bare native `CUDA_VISIBLE_DEVICES=1` gets the **3090**. → `persistent-memory.d/2026-09-01-irv-ml1-gpu-residents.md`

- `[2026-09-01]` **The Ada inference server is a used Dell R750xa (`JPJ1ZP3`) and the reseller stripped four things Dell shipped** — half the RAM, the 2400 W PSUs, and the GPU risers/cables/fans are absent from the invoice. Card is **RTX 6000 Ada**, not L40S. GPU power chain resolved via NVIDIA `930-00030-1546-000`. NVMe in the drive bays is CLOSED (SAS/SATA backplane). → `persistent-memory.d/2026-09-01-ada-inference-server-r750xa.md`

- `[2026-09-01]` **pfi-gx10 onboarded headless — and it is the intended new home for run 3c, which died on a tripped breaker.** GB10/sm_121/aarch64, 121 GB unified. NOT racked yet. Bare of any CUDA stack; probe throughput before porting. → `persistent-memory.d/2026-09-01-pfi-gx10-onboarding.md`

- `[2026-09-01]` **Ada migration is `zfs send` (branch a) — and the DESTINATION IS SMALLER THAN THE SOURCE.** 99 MB/s measured; ~3.9 h. ⚠ Measured 2026-09-01: `storetank` = 1.81 TiB pool, 1.45 TiB used, **80% CAP already**, compression off / `compressratio 1.00x` (safetensors are incompressible — no win at recv). Settled payload **~1.47 TiB**; the R750xa's as-bought 2× 1.92 TB mirrored is ~1.75 TiB → arrival at **~84%**. Fix = 2× 2 TB SATA SSD on the buy list (6 bays free) → ~3.57 TiB at ~41% with redundancy; **pair the two NEW drives together** (a mirror vdev caps at its smallest member). ⚠ **Pruning is NOT a substitute** — comfy-dev found ~215 GiB unreferenced, and deleting every byte still lands the as-bought mirror at 72%: the constraint is **vdev layout**, not payload, so the prune audit and the drive purchase are independent and neither gates the cutover. ⚠ **"Onboarded" is not "landed"** — infra-ops read `ALLOC` mid-pull and re-added the whole batch on top, inflating 84% to a quoted 90%. Also: **branch (b)'s original reason was WRONG** — comfy-dev enumerated all 12 containers, only `comfyui` mounts `/storetank`, so (b) was unavailable *during the transition*, not structurally (right conclusion, wrong reason — infra-ops reasoned about the BOX when the question was the MOUNT). Plus the retain-vs-reclaim call and the two-boxes confusion (the Ada box and the GX10 are DIFFERENT machines). → `persistent-memory.d/2026-09-01-ada-migration-branch-a.md`

- `[2026-09-01]` **Matrix: Synapse 1.120→1.159, appservice namespace opened, `/_synapse/admin` closed to the internet, alias convention ratified.** Schema migrations are one-way; push is `event_id_only` and assembled on-device. → `persistent-memory.d/2026-09-01-matrix-upgrade-and-hardening.md`

- `[2026-09-01]` **A named failure class: a correct check aimed at the wrong object.** Six instances in one day across three sessions; re-running the same check cannot catch it. **Recommended for `docs/pfi/training-throughput-playbook.md` §4 — NOT YET WRITTEN, awaiting operator.** → `persistent-memory.d/2026-09-01-wrong-object-measurement.md`

- `[2026-09-01]` **Ops boundary ruled by the operator: worldtree-dev writes the bridge code; infra-ops OPERATES the Worldtree/Matrix instances and may change them.** Corrects a mis-route where infra-ops asked worldtree-dev to provision an account on a box it does not run. Tracked at `931bac8` + althing `01M1F4PK796EDGDCBKZ9W3JC0S`.

- `[2026-09-01]` **Idle VRAM on this fleet is a RESERVED scratch pool, not waste.** Operator declined raising `vllm-mog-sec` from `gpu-memory-utilization 0.52`: single-user dev fleet, KV headroom nobody will consume is worth less than room for ephemeral models and small training runs. vLLM's "fully utilize gpu memory" startup hint does NOT apply here. Tracked in auto-memory `feedback_idle_vram_is_reserved_not_waste`.


- `[2026-08-25]` **Fused MoE kernel path — DEFERRED, tracked at park `fused-moe-kernel-path-for-gemma-4-moe-training` (id 47).** Operator: "note the fused MoE kernel for round two… if we nail it soon, the math has us wanting to restart the run anyway." Training MFU is **8.6%** (27.1 of a benchmarked 313.8 TFLOPS) because `transformers` runs the Gemma-4 experts in a Python loop — 128 experts × 30 layers, ~11,500 iterations per step under gradient checkpointing. ⚠ **The same fused 3-D expert layout that made bitsandbytes skip 88.5% of the model is exactly what a grouped GEMM wants** — the format is good for storage and for fused kernels, and hostile only to naive iteration. Two fixes: `group_by_length` (−29.9% compute, free, but breaks the seeded order manifest and re-opens a batch-composition call brokkr already made) and a grouped-GEMM/compiled MoE forward (the remaining ~10×). **Not applied to the live run** — restarting mid-flight to change batch ordering was judged a bad trade at step ~50 of 1,312.



- `[2026-08-24]` **`nconnect=8` on `/mnt/smithy` — approved but DEFERRED at operator instruction.** brokkr-smithy-dev pre-approved it for "once the FortiGate work settles" and does not need re-asking; the operator declined it in this session's scope. Tracked at althing thread `01M0R46SFYF83099N16WD67KGD`.

- `[2026-08-19]` **AI-tab Dormant regrouping BELAYED by the operator** — six seats (char-rp Magidonia, char-rp-reasoning Heretic2, Granite summarizer, Qwen-Image-Bench, Skaldsong, Chatterbox Fast) show amber EXITED inside live groups rather than `AI - Dormant`. Fix is a label change + recreate per stack; needs the operator's read on which are retired vs temporarily down. `untracked by operator choice` (his words: "belay the ai dormant regrouping for now").



- `[2026-08-15]` **RP-seat direction: KEEP MeroMero on `char-rp`; Artemis-31B rejected; next move is Dark-Scarlett on a Qwen3.8 base when it lands (operator).** Evaluated `TheDrummer/Artemis-31B-v1.1` — mechanically a drop-in (same `google/gemma-4-31B-it` base, identical 1188-tensor/356-vision census, same missing-`preprocessor_config.json` trick), so it's purely a quality call, and our own survey already ranked MeroMero **#1** vs Artemis **#6**; Artemis is also unlicensed and its author deprioritizes correctness + warns of token-banning-for-stability, which fights char-rp's tool-calling requirement. **MTP verified impossible on both** (Gemma-4 has no MTP head at all — base/MeroMero/Artemis are all MTP=0; no finetune can add one). **But speculative decoding IS reachable on a Gemma-4 seat via a DETACHED drafter** — vLLM 0.24 supports `eagle3` + `gemma4_mtp`, and real drafters exist: `google/gemma-4-31B-it-assistant` (0.94 GB, 4-layer, 761K dl), `RedHatAI/gemma-4-31B-it-speculator.eagle3` (4.47 GB), `AEON-7/…eagle3-NVFP4` (3.53 GB). ⚠ all list their verifier as **stock** gemma-4-31B-it, not an RP finetune, so acceptance against MeroMero is unmeasured and likely well below the gen seat's ~48%. UNTESTED — parked, ~45 min to measure, needs GPU0 headroom (card is at 94.4/97.9 GB). **Why the Dark-Scarlett 3.8 plan is the strong one:** DS is Qwen3.6-based today, so a 3.8 respin lands on the *gen seat's* architecture → native MTP returns and the whole mixed NVFP4+FP8 recipe + graft ports directly. Watch two things on arrival: `from_pretrained` **silently drops MTP heads during finetuning** (verify 15 `mtp.*` tensors in the index; graft from stock if absent), and DS v1.0 required the `Qwen3_5ForConditionalGeneration` **wrapper class** to save a config vLLM/SGLang accept. Both in `docs/pfi/model-quantization-playbook.md`.





_30 older entries archived to archival-memory.md._

## Tried and abandoned

- `[2026-09-04]` **Forcing 10G on the ESH-Media DAC — it linked, then degraded over hours, and I reported a plateau at two minutes.** ⚠ A clean zero-error link at 1G does NOT rule out a marginal cable; autoneg's fallback was protecting something real. → `persistent-memory.d/2026-09-04-dac-forced-10g-failed.md`

_110 older entries archived to archival-memory.md._