chore(erp-tune): purge intermediate checkpoints (~74 GB); R49 carrier settled on dense Qwen3

Two operator rulings, 2026-09-09.

"purge intermediate checkpoints" -- seven checkpoints/ directories removed with
literal paths, one rm per line, after confirming none was a symlink and that
every run's final adapter/ is an independent real directory:

  pfi-gx10   run-03c 11G  run-04 16G  run-05 9.2G  run-06 9.2G   = 45 GB
  ana-ml2    run-01 12G   run-02 12G  run-03 5.9G                = 29 GB

gx10 419G->374G used, 496 GB free. /tank/erp-tune 392G->363G with zfs list -t
snapshot empty, so the space is genuinely returned rather than snapshot-held. All
eight adapters re-verified by sha256 after the deletion, matching the values
recorded during the mirror. Merged artifacts deliberately untouched -- they are
not checkpoints, and the ~550 GB of superseded merges stays a separate call.

"use dense qwen3" -- the R49 H02 carrier sweep becomes Qwen3-{0.6,1.7,4}B-Base,
which overrides the Qwen3.5 arms H02 names; brokkr-smithy owns that file and was
told directly. Qwen3-4B-Base staged and benched to complete the family:

  Qwen3-0.6B-Base   0.616 B   1.707 s/step   2,399 tok/s   spread 0.6%
  Qwen3-1.7B-Base   1.755 B   2.895 s/step   1,415 tok/s   spread 0.8%
  Qwen3-4B-Base     4.089 B   5.714 s/step     717 tok/s   spread 0.3%

The dense 4.089 B carrier still trains 33% faster than the hybrid 0.765 B one.
Projected per voice 2.7 / 4.6 / 9.1 h; the three-arm sweep at two seeds is ~33 h
of GPU, ~10 h if H03's corpus floor holds. The three Qwen3.5 checkpoints stay
staged so the decision is reversible behind an fla install.

Also recorded: verified at 22:45-22:48 PT that nothing is training on gx10,
ana-ml2, nh3-dev or irv-ml1, and that brokkr's own run07-gate close states
"Nothing is owed. No battery to run." Run 7 has no servable artifact left. And a
correction to a standing lesson -- the bracketed-class trick does not defeat a
wrapper's argv, since the invoking shell's command line carries the literal
pattern; observe the artifact instead.

docs/pfi/author-voice-lora-regime.md

@@ -1,7 +1,9 @@
 # Author-voice LoRA regime on pfi-gx10 — training-side prep
 
-_Written 2026-09-09 22:45 PT. Status: **PREP. Nothing is training and nothing is
+_Written 2026-09-09 22:45 PT, revised 23:05 PT. Status: **PREP. Nothing is
-queued.** Two operator decisions open (§6)._
+training and nothing is queued.** Both operator decisions from the first draft are
+now **SETTLED** (§6): carrier family is the **dense `Qwen3` line**, and the
+intermediate checkpoints are purged._
 
 The research target is **brokkr-smithy R49** (`research/R49-author-voice-adapters/`),
 whose seed is the operator's **BabyBronte** design doc
@@ -58,9 +60,10 @@ Staged on local NVMe under `/home/infra-ops/carriers/` (pulled 2026-09-09,
 existence API-verified against the HF registry first, with a phantom repo run as
 the negative control):
 
-    Qwen3.5-0.8B-Base   1.7G      Qwen3-0.6B-Base   1.2G
+    CHOSEN — dense Qwen3          held, not chosen — hybrid Qwen3.5
-    Qwen3.5-2B-Base     4.3G      Qwen3-1.7B-Base   3.3G
+    Qwen3-0.6B-Base   1.2G        Qwen3.5-0.8B-Base   1.7G
-    Qwen3.5-4B-Base     8.8G
+    Qwen3-1.7B-Base   3.3G        Qwen3.5-2B-Base     4.3G
+    Qwen3-4B-Base     ~8G         Qwen3.5-4B-Base     8.8G
 
 Probes live at `scripts/training-probes/{probe_carrier.py,bench_lora_step.py}`
 with raw output in `bench-lora-step-gx10-2026-09-09.jsonl`, so every number below
@@ -127,6 +130,7 @@ measured after 3 warmup steps; median reported with the full spread.
 | `Qwen3.5-0.8B-Base` (no grad-ckpt) | " | 0.765 B | 6.364 | 644 | 38.9 GiB | 1.5% |
 | `Qwen3-0.6B-Base` | dense | 0.616 B | 1.707 | **2,399** | 9.8 GiB | 0.6% |
 | `Qwen3-1.7B-Base` | dense | 1.755 B | 2.895 | **1,415** | 12.2 GiB | 0.8% |
+| `Qwen3-4B-Base` | dense | 4.089 B | 5.714 | **717** | 17.2 GiB | 0.3% |
 | `Qwen3-1.7B-Base`, batch 4 | dense | 1.755 B | 11.387 | 1,439 | 38.0 GiB | 0.6% |
 | `Qwen3.5-0.8B-Base`, batch 4 | hybrid | 0.765 B | 30.030 | 546 | 55.5 GiB | 0.7% |
 
@@ -152,8 +156,14 @@ means the 2.6× gap is the kernel path, not a batching artefact.
 |---|---|
 | `Qwen3-0.6B-Base` dense | **2.7 h** |
 | `Qwen3-1.7B-Base` dense | **4.6 h** |
+| `Qwen3-4B-Base` dense | **9.1 h** |
 | `Qwen3.5-0.8B-Base` hybrid | **12 h** |
 
+The sharpest way to put it: **the dense 4.089 B carrier still trains 33% faster
+than the hybrid 0.765 B one**, on 5.3× the parameters. The full three-arm dense
+sweep at two seeds each is ~33 h of GPU — about a day and a half for the whole
+H02 carrier question, and ~10 h if H03's ~300 k-word corpus floor holds.
+
 The hybrid carrier would make a per-voice run *longer than the 7 h 26B-A4B tune
 it exists to replace.* At R49 H03's hoped-for corpus floor (~300 k words) the
 dense 1.7B lands near **1.4 h** — a voice per afternoon, which is the regime the
@@ -163,7 +173,7 @@ first real corpus.
 
 ## 5. Prep remaining, in order
 
-1. **Carrier family decision** (§6) — everything downstream keys on it.
+1. ~~Carrier family decision~~ — **settled: dense `Qwen3`** (§6a).
 2. **Corpus D1** — Gutenberg Brontë (Jane Eyre, Villette, Shirley, The
    Professor), boilerplate stripped, chapter-segmented, typography normalised,
    character inventory recorded. Public domain, clean under any disposition.
@@ -195,22 +205,40 @@ integration contract, the modern in-copyright arsenal, inference latency.
 
 ## 6. Open for the operator
 
-**(a) Carrier family — recommend the dense `Qwen3` line.** R49 H02 pins
+**(a) Carrier family — SETTLED 2026-09-09, operator: _"use dense qwen3"._** The
-`Qwen3.5-{0.8,2,4}B-Base`. The measurements say that family costs 2.6–6× the
+sweep is **`Qwen3-{0.6,1.7,4}B-Base`** — the design doc's own original pin, and
-wall-clock on this box, brings a vision tower and an MTP head to a text job, and
+the family the measurements favour by 2.6–6×. **This overrides R49 H02's stated
-makes cross-document packing unsafe in 18 of 24 layers. `Qwen3-{0.6,1.7,4}B-Base`
+arms**, which name `Qwen3.5-{0.8,2,4}B-Base`; brokkr-smithy owns that file and
-— the design doc's own original pin — is plain dense, has none of those, and is
+has been told directly.
-the better instrument for a probe whose whole point is isolating one variable.
-The cost is one model generation of base quality. Reversible: the Qwen3.5
-checkpoints stay staged, and a fused-kernel install (`fla` is pure Triton and
-would plausibly work on aarch64) could revive them later as a follow-up rather
-than a blocker.
 
-**(b) 45 GB of intermediate checkpoints on gx10, and 354 GB on ana-ml2.**
+What the ruling buys, restated so it is not re-litigated: no vision tower and no
-`run-03c/04/05/06/checkpoints` total 45 GB on gx10; `/tank/erp-tune/serve` is
+MTP head to exclude, cross-document packing is safe again, `sdpa`/flash are both
-354 GB of superseded merged models on ana-ml2. The final adapters are safe (§7)
+reachable, and the per-voice wall-clock is 2.7–4.6 h instead of 12 h. What it
-and run 6 is the standing seat. Purging is the operator's call — 470 GB free on
+costs: one model generation of base quality. **Reversible** — the three Qwen3.5
-gx10 means it does not block this regime.
+checkpoints stay staged (14.8 GB), and an `fla` install (pure Triton, plausibly
+fine on aarch64) could revive that family as a follow-up experiment rather than a
+prerequisite. If anyone re-opens this, re-run `bench_lora_step.py` first; the
+argument is a measurement, not a preference.
+
+**(b) Intermediate checkpoints — PURGED 2026-09-09 23:00 PT, operator:
+_"purge intermediate checkpoints"._** Seven `checkpoints/` directories deleted
+with literal paths, one `rm` per line, after confirming none was a symlink and
+that every run's final `adapter/` is an independent real directory:
+
+    pfi-gx10   run-03c 11G · run-04 16G · run-05 9.2G · run-06 9.2G   -> 45 GB
+    ana-ml2    run-01 12G · run-02 12G · run-03 5.9G                  -> 29 GB
+
+gx10 419G→374G used (496 GB free); `/tank/erp-tune` 392G→363G, with `zfs list -t
+snapshot` empty so the space is genuinely returned rather than snapshot-held.
+**All eight adapters re-verified by `sha256` after the deletion**, matching the
+values recorded during the mirror.
+
+⏳ **Still standing, and it is the bigger prize — ~550 GB of superseded MERGED
+models, deliberately NOT touched** because they are not checkpoints:
+`gx10:~/erp-tune/serve/merged-run{03c,04,05,06}` at 49 GiB each (196 GiB) and
+`ana-ml2:/tank/erp-tune/serve` (~354 GB). Run 6 is the standing seat, so
+`merged-run06` plus `/tank/aimodels/erp-tune-v6-{bf16,nvfp4a16}` are the rollback
+path and should survive; `merged-run03c/04/05` are dead weight. Separate call.
 
 ## 7. Adapter disposition — settled, and made real
 

persistent-memory.md

@@ -1,6 +1,6 @@
 # Persistent memory — eshpfi-management
 
-_Last updated: 2026-09-09 22:50 PT (**Pfish-6** = run-6 NVFP4 is the standing seat, ana-ml2 :8021 ONLY; run 7 PURGED ~139 GiB; pfi-gx10 is an experimental/TRAINING box and carries no serving seat; all five ERP adapters now MIRRORED to ana-ml2 and inside restic; **BabyBronte / R49 author-voice regime is in PREP on gx10** — carrier-family decision open)_
+_Last updated: 2026-09-09 23:10 PT (**Pfish-6** = run-6 NVFP4 is the standing seat, ana-ml2 :8021 ONLY; run 7 PURGED ~139 GiB; pfi-gx10 is an experimental/TRAINING box and carries no serving seat; all five ERP adapters now MIRRORED to ana-ml2 and inside restic; **BabyBronte / R49 author-voice regime is in PREP on gx10, carrier SETTLED = dense Qwen3**; intermediate checkpoints PURGED both boxes ~74 GB)_
 
 > **Always check for `/tmp/infra-ops-handoff.md`** — if it exists and its
 > `Written:` stamp is under an hour old, read it (it carries the in-flight
@@ -129,6 +129,22 @@ preserved verbatim in `archival-memory.md` § Superseded in-flight snapshots._
   already gone; do NOT `rm -rf /home/infra-ops/erp-tune/run-07` — operator ruled **keep the adapter**
   2026-09-09.
 - **NO NEW *ERP* TRAINING RUN PLANNED.** The opening-split idea is not being re-tested; run 6 stands.
+- **⚠ VERIFIED 2026-09-09 22:45–22:48 PT: NOTHING IS TRAINING ANYWHERE.** Checked gx10 (0 GPU compute
+  apps, 0% util, every pidfile dead), ana-ml2 (all 8 GPU processes are vLLM serving seats, no trainer),
+  nh3-dev, irv-ml1 (5 GPU processes, all the known native serving stacks). The last run was **run 7,
+  finished 13:23 PT**, and brokkr's own `research/R47-premium-corpus-gate/run07-gate/GATE-FAILED-run07.md`
+  (rev 19:40 PT) closes it: **"Nothing is owed. No battery to run."** Three findings there worth
+  holding: the CSAM gate fire was a **FALSE POSITIVE** (adjective `minor` in a HARD `child_term` tier,
+  fixed `cc42d76`, selftest 24/24) so the safety result is **VOID on an instrument defect, not a
+  failure**; run 7 is retired **on its merits** (primary flat 69→70.5, both diversity families reduced,
+  coherence 1.0→0.875), which is the operator's own ruling; and ⚠ **do NOT re-run that gate against
+  gx10:8098** — the artifact there changed and then went down entirely. ⚠ **Run 7 has NO servable
+  artifact left**: `serve/merged-run07` and `serve/pfish6-nvfp4a16` are both GONE, so "serve run 7"
+  means a fresh 49 GiB merge + 16 GiB requant, hours, on a tune already judged flat.
+- **⚠ THE BRACKETED-CLASS pgrep TRICK DOES NOT SAVE A WRAPPER'S ARGV.** `ps -eo cmd | grep '[t]rain'`
+  still matched my own shell command, because the invoking `zsh -c '...'` argv contains the literal
+  bracketed pattern. The bracket only stops *grep itself* from matching. Durable form: **observe the
+  artifact** (GPU compute apps, the seat's own log) or exclude known PIDs — never the process name.
 - **✅ ALL FIVE gx10 ERP ADAPTERS ARE NOW TWO-COPY AND BACKED UP** (2026-09-09 22:30 PT, operator:
   *"keep the adapter"*). `run-03c/04/05/06/07` mirrored to `ana-ml2:/tank/erp-tune/run-<N>/adapter`
   (the layout runs 01–03 already used), byte-totals identical both sides + `sha256` match on every
@@ -136,6 +152,17 @@ preserved verbatim in `archival-memory.md` § Superseded in-flight snapshots._
   regenerable weights), so `configs/restic/ana-ml2/profiles.yaml` gained ONE documented carve-out —
   `/tank/erp-tune/run-*/adapter` — verified by `resticprofile --dry-run` to expand to exactly those 8
   paths and nothing else. Live file is byte-identical to the repo canonical; `.bak-20260909` beside it.
+- **🧹 INTERMEDIATE CHECKPOINTS PURGED 2026-09-09 23:00 PT** (operator: *"purge intermediate
+  checkpoints"*). Seven `checkpoints/` dirs, deleted with LITERAL paths one `rm` per line after
+  confirming none was a symlink and that every final `adapter/` is an independent real dir: gx10
+  `run-03c` 11G + `run-04` 16G + `run-05` 9.2G + `run-06` 9.2G = **45 GB** (419G→374G used, 496 GB
+  free); ana-ml2 `/tank/erp-tune/run-01` 12G + `run-02` 12G + `run-03` 5.9G = **29 GB**
+  (`/tank/erp-tune` 392G→363G, `zfs list -t snapshot` EMPTY so the space is genuinely returned).
+  **~74 GB total. All eight adapters re-verified by sha256 AFTER the deletion**, matching the mirror.
+  ⏳ **NOT touched, and it is the bigger prize (~550 GB): superseded MERGED models** — they are not
+  checkpoints, so I stayed literal. gx10 `serve/merged-run{03c,04,05,06}` 49 GiB each (196 GiB) +
+  ana-ml2 `/tank/erp-tune/serve` ~354 GB. Keep `merged-run06` + `/tank/aimodels/erp-tune-v6-*` (the
+  standing seat's rollback); `merged-run03c/04/05` are dead weight. Separate operator call.
 - **🖋 BabyBronte / R49 author-voice LoRA regime — IN PREP on pfi-gx10, nothing training.** Plan +
   every measured number: [`docs/pfi/author-voice-lora-regime.md`](docs/pfi/author-voice-lora-regime.md).
   Research target is **brokkr-smithy R49** (`research/R49-author-voice-adapters/`) — brokkr owns
@@ -158,9 +185,14 @@ preserved verbatim in `archival-memory.md` § Superseded in-flight snapshots._
   sampled-target-changed check in the SERVING path); and **cross-document packing is unsafe** because
   SSM state ignores the attention mask, which breaks the per-copy name-consistency invariant the design
   doc calls sacred. None of these exist on the dense line.
-  **⭐ OPEN FOR THE OPERATOR: carrier family — recommend the dense `Qwen3-{0.6,1.7,4}B-Base` line**
+  **✅ CARRIER SETTLED 2026-09-09 — operator: *"use dense qwen3"*.** Sweep is
-  (the design doc's own original pin) over R49 H02's `Qwen3.5` trio. Reversible: Qwen3.5 stays staged
+  **`Qwen3-{0.6,1.7,4}B-Base`**, all three staged on gx10. ⚠ **This OVERRIDES R49 H02's stated arms**
-  and an `fla` install (pure Triton, plausibly fine on aarch64) could revive it as a follow-up.
+  (which name the Qwen3.5 trio) — brokkr-smithy owns that file and was told directly. Full dense sweep
+  measured: 0.6B **2,399**, 1.7B **1,415**, 4B **717** tok/s (spread 0.3–0.8%) → projected **2.7 / 4.6 /
+  9.1 h** per voice; whole 3-arm × 2-seed sweep ≈ 33 h GPU, ≈10 h if H03's ~300k-word floor holds. The
+  sharpest framing: **dense 4.089B still beats hybrid 0.765B by 33%** on 5.3x the params. Qwen3.5 stays
+  staged (14.8 GB) — an `fla` install could revive it as a follow-up experiment, never a prerequisite.
+  If anyone re-opens this, re-run `bench_lora_step.py` first: the argument is a measurement.
 - **📌 forseti shipped althing 3.6.1 (2026-09-09) — every box's herald needs a restart to pick up the
   new poke text.** `uv tool install --force --reinstall <checkout>` + `systemctl --user restart
   althing-po-herald`, expect `3.6.1`; the plugin also went to 0.1.6 (inbox.md no longer calls the

scripts/training-probes/bench-lora-step-gx10-2026-09-09.jsonl

@@ -4,3 +4,4 @@
 {"model": "Qwen3-1.7B-Base", "total_params_B": 1.755, "lora_rank": 32, "targets": "attn_mlp", "adapted_modules": 196, "trainable_params_M": 34.87, "trainable_pct": 1.986, "batch": 1, "seq": 4096, "tokens_per_microbatch": 4096, "grad_checkpointing": true, "attn_impl": "sdpa", "dtype": "bfloat16", "device": "NVIDIA GB10", "torch": "2.14.0+cu130", "warmup": 3, "n": 10, "s_per_step_median": 2.8946, "s_per_step_min": 2.8782, "s_per_step_max": 2.901, "s_per_step_spread_pct": 0.8, "tok_per_s_median": 1415.0, "peak_mem_GiB": 12.24}
 {"model": "Qwen3-1.7B-Base", "total_params_B": 1.755, "lora_rank": 32, "targets": "attn_mlp", "adapted_modules": 196, "trainable_params_M": 34.87, "trainable_pct": 1.986, "batch": 4, "seq": 4096, "tokens_per_microbatch": 16384, "grad_checkpointing": true, "attn_impl": "sdpa", "dtype": "bfloat16", "device": "NVIDIA GB10", "torch": "2.14.0+cu130", "warmup": 3, "n": 10, "s_per_step_median": 11.3872, "s_per_step_min": 11.3401, "s_per_step_max": 11.4128, "s_per_step_spread_pct": 0.6, "tok_per_s_median": 1438.8, "peak_mem_GiB": 37.99}
 {"model": "Qwen3.5-0.8B-Base", "total_params_B": 0.765, "lora_rank": 32, "targets": "attn_mlp", "adapted_modules": 96, "trainable_params_M": 12.78, "trainable_pct": 1.67, "batch": 4, "seq": 4096, "tokens_per_microbatch": 16384, "grad_checkpointing": true, "attn_impl": "sdpa", "dtype": "bfloat16", "device": "NVIDIA GB10", "torch": "2.14.0+cu130", "warmup": 3, "n": 10, "s_per_step_median": 30.0301, "s_per_step_min": 29.9379, "s_per_step_max": 30.1444, "s_per_step_spread_pct": 0.7, "tok_per_s_median": 545.6, "peak_mem_GiB": 55.49}
+{"model": "Qwen3-4B-Base", "total_params_B": 4.089, "lora_rank": 32, "targets": "attn_mlp", "adapted_modules": 252, "trainable_params_M": 66.06, "trainable_pct": 1.616, "batch": 1, "seq": 4096, "tokens_per_microbatch": 4096, "grad_checkpointing": true, "attn_impl": "sdpa", "dtype": "bfloat16", "device": "NVIDIA GB10", "torch": "2.14.0+cu130", "warmup": 3, "n": 10, "s_per_step_median": 5.7145, "s_per_step_min": 5.7085, "s_per_step_max": 5.727, "s_per_step_spread_pct": 0.3, "tok_per_s_median": 716.8, "peak_mem_GiB": 17.22}