diff --git a/CLAUDE.md b/CLAUDE.md
index 28ade21..1476e26 100644
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -89,6 +89,8 @@ E4's whole purpose is to isolate the effect of teacher **decorrelation ρ**, so
 
 **Finding (2026-07-05, LLM prototype `llm_merge` — the first real-LLM step; honest/partial).** First move from toy models toward real LLMs (blueprint C2/C4, the real-LLM image of E8), on one 16 GB GPU. New `src/llm/` package: procedural task families + exact-match verifier (`tasks.py`), batched eval (`evaluate.py`), LoRA specialisation (`specialise.py`, manual answer-only SFT), weight-space merge via peft `add_weighted_adapter` (`merge.py`: soup=averaged deltas, ties=sign-reconciled union), runner (`experiment.py`, kind `llm_merge`). Base = **Qwen2.5-0.5B-Instruct** (Apache-2.0). Three *disjoint*, deliberately-hard families (lists/strings/arith); one LoRA specialist each (~90 s total). **Result (seed 1):** each specialist spikes on its own family; the **merges are the only models competent across ALL families — worst-family ≈0.25 vs <0.16 for every single specialist** (the Fisher-Muller "generalist from specialists" signature, robust). **But** the stronger "exceeds every parent *overall*" claim is only marginal at this scale (soup 0.64 vs best specialist 0.63; ties 0.61 below it), and averaging visibly *dilutes* peaks (lists: specialist 0.43 → merge 0.26 — the E4 "merge, don't average" caveat in real weights). Honest scope: pipeline works end-to-end; the balance/retention half reproduces; the strict overall-exceeds and the soup-vs-ties distinction need scale (bigger base, more/cleaner families, seeds, dilution-resistant/offspring-selected merge). **Env notes:** Python 3.14 + transformers 5.13 works (cp314 wheels exist); `transformers 5.x` changed `apply_chat_template` (returns a dict; render to text then tokenize; pass `**inputs` to `generate`). `make env-llm` / `make llm`; adapters cached under gitignored `models/llm/`, base in the HF cache (outside the repo). 125 tests green (+3 pure task/verifier). The full grounded sexual *society* on LLMs (C1 collapse, directed sex, the dynamic society) is the HPC-scale next step.
 
+**Finding (2026-07-05, `llm_merge_hpc` — the 7B firm-up on Imperial CX3; the marginal sign becomes decisive).** Re-ran `llm_merge` at a capable base — **Qwen2.5-7B-Instruct**, 200 test tasks/family, one **L40S (46 GB)** GPU, 8 min walltime — via the `/imperial-hpc` runbook (see `memory/hpc-setup.md`). **Both merges reach 0.87 overall, decisively above the best single specialist (0.77) and above every specialist on every family; worst-family 0.62 vs ≤0.57 for any specialist.** The two 0.5 B caveats are resolved: (1) the strict Fisher–Muller "exceeds every parent overall" claim is now clean (+10 points, not marginal); (2) the **dilution vanishes** — at 7 B the merge *beats* the lists-specialist on lists (0.62 > 0.57), where at 0.5 B averaging diluted it (0.43 → 0.26). **Interpretation: dilution is a small-model artefact; a capable base has enough headroom that weight-space averaging composes rather than dilutes** — the "merge, don't average" concern (E4) softens once parents are strong (soup ≈ ties at K=3). Results synced to `results/llm_merge_hpc/` (README legend + data-driven figure title). The natural refinement is *module-level* union-preserving recombination (MoE-expert / adapter-union merge, the real-weight image of E8's max-merge) rather than delta-averaging.
+
 ## Build order (blueprint §7) — respect the gate
 
 1. Scaffold: repo layout (§5), container, pytest skeleton, config system, seeding utils. `make test` green.
diff --git a/figures/plot_llm_merge.py b/figures/plot_llm_merge.py
index a875eea..cd51939 100644
--- a/figures/plot_llm_merge.py
+++ b/figures/plot_llm_merge.py
@@ -67,12 +67,17 @@ def main(results_dir: str = "results/llm_merge") -> None:
                edgecolor="white")
     ax.set_xticks(x2); ax.set_xticklabels([labels[m] for m in models], rotation=25, ha="right",
                                           fontsize=8)
-    ax.set(ylabel="accuracy", title="Overall (solid) vs worst-family (hatched): the merge "
-           "clearly wins\nworst-family (balance); overall it matches the best specialist")
+    ax.set(ylabel="accuracy", title="Overall (solid) vs worst-family (hatched):\n"
+           "the recombined model vs the best single specialist")
     ax.legend(frameon=False, fontsize=9)
 
-    fig.suptitle("llm_merge — recombining decorrelated specialist LLMs gives the only model competent "
-                 f"across all families (balance); overall parity ({cfg['base_model']})", y=1.0, fontsize=12)
+    best_spec = max(_acc(df, m, "overall") for m in specialists)
+    best_merge = max(_acc(df, m, "overall") for m in merges)
+    verdict = (f"recombined {best_merge:.2f} > best specialist {best_spec:.2f} overall"
+               if best_merge > best_spec + 0.005 else
+               f"recombined {best_merge:.2f} ≈ best specialist {best_spec:.2f} overall")
+    fig.suptitle(f"llm_merge — recombining decorrelated specialist LLMs: {verdict} "
+                 f"({cfg['base_model'].split('/')[-1]})", y=1.0, fontsize=12)
     fig.tight_layout()
     savefig(fig, results_dir, "llm_merge")
 
diff --git a/hpc/llm_merge.pbs b/hpc/llm_merge.pbs
index 7cc9767..720794d 100644
--- a/hpc/llm_merge.pbs
+++ b/hpc/llm_merge.pbs
@@ -1,28 +1,24 @@
 #!/bin/bash
-# Scaled LLM merge experiment on an L40S (48 GB) — the "firm up the sign" run.
-# Prereqs (do ONCE on the login node, which has internet — see hpc/README.md):
-#   1) build the env:            uv sync --extra dev --extra neural --extra llm
-#      (if the L40S driver is older than CUDA 13, first pin torch — see README)
-#   2) pre-download the base:    HF_HOME=$HOME/hf_cache uv run python -c \
-#         "from transformers import AutoModelForCausalLM,AutoTokenizer as T; \
-#          [f('Qwen/Qwen2.5-7B-Instruct') for f in (T.from_pretrained, AutoModelForCausalLM.from_pretrained)]"
-#   submit:  qsub hpc/llm_merge.pbs
+# Scaled LLM merge experiment on an L40S (46 GB) — the "firm up the sign" run.
+# Confirmed on CX3 (probe 2026-07-05): L40S driver 580 supports CUDA 13.0, so the committed
+# torch cu130 env works as-is; compute nodes HAVE internet; $TMPDIR is fast local NVMe.
+# Prereqs on the LOGIN node (see hpc/README.md): `uv sync --extra dev --extra neural --extra llm`
+# (cache on $EPHEMERAL) + pre-download the base into $HF_HOME below.
+#   submit:  qsub hpc/llm_merge.pbs      status: qstat -u $USER
 #PBS -l select=1:ncpus=8:mem=64gb:ngpus=1:gpu_type=L40S
-#PBS -l walltime=04:00:00
-#PBS -N lamarckian_llm_merge
+#PBS -l walltime=02:00:00
+#PBS -N lam_llm_merge
 
 cd "$PBS_O_WORKDIR"
-export HF_HOME="$HOME/hf_cache"           # models cached here (pre-downloaded on the login node)
-export HF_HUB_OFFLINE=1                    # compute node has no internet -> use the cache
-                                          # (delete this line if the probe shows internet works)
+export HF_HOME="$EPHEMERAL/hf_cache"       # models cached on the 11 TB ephemeral store
 export TOKENIZERS_PARALLELISM=false
+export UV_CACHE_DIR="$EPHEMERAL/uvcache"
 
-source .venv/bin/activate                 # uv-built env (Python 3.14 + torch + transformers/peft)
-nvidia-smi --query-gpu=name,memory.total --format=csv,noheader
+source .venv/bin/activate                  # uv-built env (Python 3.14 + torch cu130 + transformers/peft)
+nvidia-smi --query-gpu=name,memory.total,driver_version --format=csv,noheader
 
 python -m llm.experiment configs/llm/merge_hpc.yaml
 
-# results/llm_merge_hpc/{results.parquet,resolved_config.yaml,manifest.json} are written in-place.
-# The parquet is gitignored; sync it back to analyse/plot locally, e.g. from your laptop:
-#   rsync -avz <user>@login.hpc.ic.ac.uk:'~/…/results/llm_merge_hpc/' results/llm_merge_hpc/
+# results/llm_merge_hpc/ is written in-place (parquet gitignored). Sync back to plot:
+#   rsync -avz hpc:'…/LamarckianAI/results/llm_merge_hpc/' results/llm_merge_hpc/
 echo "done: $(date)"
diff --git a/hpc/llm_smoke.pbs b/hpc/llm_smoke.pbs
new file mode 100644
index 0000000..f23e754
--- /dev/null
+++ b/hpc/llm_smoke.pbs
@@ -0,0 +1,14 @@
+#!/bin/bash
+# Smoke gate: run the small 0.5B merge experiment on an L40S to confirm the GPU pipeline works
+# on CX3 before the 4 h 7B run. ~5 min.  submit: qsub hpc/llm_smoke.pbs
+#PBS -l select=1:ncpus=8:mem=48gb:ngpus=1:gpu_type=L40S
+#PBS -l walltime=00:20:00
+#PBS -N lam_llm_smoke
+
+cd "$PBS_O_WORKDIR"
+export HF_HOME="$EPHEMERAL/hf_cache"
+export TOKENIZERS_PARALLELISM=false
+source .venv/bin/activate
+nvidia-smi --query-gpu=name,memory.total,driver_version --format=csv,noheader
+python -m llm.experiment configs/llm/merge.yaml   # 0.5B, writes results/llm_merge/
+echo "done: $(date)"
diff --git a/results/llm_merge/llm_merge.pdf b/results/llm_merge/llm_merge.pdf
index e9763e9..2ddd73b 100644
Binary files a/results/llm_merge/llm_merge.pdf and b/results/llm_merge/llm_merge.pdf differ
diff --git a/results/llm_merge/llm_merge.png b/results/llm_merge/llm_merge.png
index e907113..1df7a24 100644
Binary files a/results/llm_merge/llm_merge.png and b/results/llm_merge/llm_merge.png differ
diff --git a/results/llm_merge_hpc/README.md b/results/llm_merge_hpc/README.md
new file mode 100644
index 0000000..0ec5989
--- /dev/null
+++ b/results/llm_merge_hpc/README.md
@@ -0,0 +1,49 @@
+# llm_merge_hpc — recombining specialist LLMs at scale (7B, Imperial CX3; blueprint C2/C4)
+
+**Claim tested.** The scale-up of `llm_merge`: at a *capable* base, does recombining decorrelated
+specialists produce a model that **exceeds every single parent overall** *and* stays competent across
+**all** families (the Fisher–Muller "offspring fitter than any parent" signature, E8) — the claim that
+was only marginal at 0.5 B? Run on one **L40S (46 GB)** GPU node of Imperial College's CX3 HPC
+(job walltime **8 min**).
+
+**Setup.** Base model **Qwen2.5-7B-Instruct** (Apache-2.0). Same three *disjoint*, procedurally-
+generated task families with an **exact-match verifier** (the "reality that says no"): `lists`,
+`strings`, `arith`, deliberately hard so specialists decorrelate. One **LoRA specialist** (r=16,
+α=32, 3 epochs, 800 train tasks) is fine-tuned per family, then the base, each specialist, and two
+weight-space **merges** — `soup` (averaged LoRA deltas) and `ties` (sign-reconciled union) — are
+evaluated on a held-out mixed test set. Seed 1, **200 test tasks/family**.
+
+### Results (accuracy)
+| model | lists | strings | arith | overall | **worst family** |
+|---|---|---|---|---|---|
+| base | 0.46 | 0.69 | 1.00 | 0.71 | 0.46 |
+| spec: lists | 0.57 | 0.74 | 1.00 | 0.77 | 0.57 |
+| spec: strings | 0.32 | 0.97 | 1.00 | 0.76 | 0.32 |
+| spec: arith | 0.47 | 0.80 | 0.96 | 0.74 | 0.47 |
+| **merge: soup** | **0.62** | **1.00** | 1.00 | **0.87** | **0.62** |
+| **merge: ties** | **0.62** | **1.00** | 0.99 | **0.87** | **0.62** |
+
+### What holds (the 0.5 B caveats resolved)
+- **"Exceeds every parent overall" — now clean.** Both merges reach **0.87 overall**, above the best
+  single specialist (lists, **0.77**) by a decisive 10 points, and above every specialist on every
+  family (lists 0.62 > 0.57; strings 1.00 > 0.97; arith ≈ 1.00). This is the strict Fisher–Muller
+  claim — offspring fitter than *any* parent — which was only marginal at 0.5 B and is now clean.
+- **"Retains all specialties" — sharper than ever.** The merges are the only models competent across
+  **all** families: worst-family **0.62**, versus **≤ 0.57** for every specialist (strings-specialist
+  collapses to 0.32 on lists). The generalist assembled from specialists dominates on both axes.
+- **The dilution is gone.** At 0.5 B, averaging *diluted* the lists-specialist (0.43 → 0.26). At 7 B
+  the merge **exceeds** the lists-specialist on lists (0.62 > 0.57). A capable base has enough headroom
+  that weight-space averaging composes rather than dilutes — the "merge, don't average" concern (E4)
+  softens once the parents are strong. Soup and ties are indistinguishable at K=3 here.
+
+### Takeaway
+**The scale-up firms up the sign the prototype left marginal.** At a capable base, recombining
+decorrelated specialists yields a model that beats every parent both overall and per-family, with no
+dilution — the sexual-reproduction / Fisher–Muller claim reproduced cleanly in real LLM weights. The
+0.5 B `llm_merge` prototype de-risked the pipeline and flagged dilution as the risk; the 7B run shows
+that risk is a small-model artefact. **Falsifier (not triggered):** a single specialist matching or
+beating the merge overall — here the merge leads by 10 points. Provenance: L40S, torch 2.12.1 /
+transformers 5.13.0 / peft 0.19.1, `manifest.json` records the results hash and library versions.
+
+*Note: this run's `manifest.json` has `git_commit: null` because it was produced on the HPC node from
+an rsync'd (non-git) working copy; the committed artefacts here are the source of truth for the figure.*
diff --git a/results/llm_merge_hpc/llm_merge.pdf b/results/llm_merge_hpc/llm_merge.pdf
new file mode 100644
index 0000000..92cb250
Binary files /dev/null and b/results/llm_merge_hpc/llm_merge.pdf differ
diff --git a/results/llm_merge_hpc/llm_merge.png b/results/llm_merge_hpc/llm_merge.png
new file mode 100644
index 0000000..b15d676
Binary files /dev/null and b/results/llm_merge_hpc/llm_merge.png differ
diff --git a/results/llm_merge_hpc/manifest.json b/results/llm_merge_hpc/manifest.json
new file mode 100644
index 0000000..f0125f2
--- /dev/null
+++ b/results/llm_merge_hpc/manifest.json
@@ -0,0 +1,20 @@
+{
+  "experiment": "llm_merge_hpc",
+  "master_seed": 1,
+  "git_commit": null,
+  "python": "3.11.13",
+  "libraries": {
+    "numpy": "2.4.6",
+    "scipy": "1.17.1",
+    "pandas": "3.0.3",
+    "pyarrow": "24.0.0",
+    "torch": "2.12.1",
+    "transformers": "5.13.0",
+    "peft": "0.19.1"
+  },
+  "rows": 30,
+  "results_sha256": "6cc0a07c66ba92a379d895d6d6707591aced48f06eee895bb4f6c15d12e6e588",
+  "layer": "2",
+  "tier": "llm",
+  "base_model": "Qwen/Qwen2.5-7B-Instruct"
+}
\ No newline at end of file
diff --git a/results/llm_merge_hpc/resolved_config.yaml b/results/llm_merge_hpc/resolved_config.yaml
new file mode 100644
index 0000000..b02f7fa
--- /dev/null
+++ b/results/llm_merge_hpc/resolved_config.yaml
@@ -0,0 +1,24 @@
+experiment: llm_merge_hpc
+seed: 1
+n_replicates: 1
+source_config:
+  experiment: llm_merge_hpc
+  kind: llm_merge
+  seed: 1
+  n_replicates: 1
+  base_model: Qwen/Qwen2.5-7B-Instruct
+  families:
+  - lists
+  - strings
+  - arith
+  n_train: 800
+  n_test: 200
+  epochs: 3
+  lora:
+    r: 16
+    alpha: 32
+  merges:
+  - soup
+  - ties
+  output:
+    dir: results/llm_merge_hpc
diff --git a/tasks/todo.md b/tasks/todo.md
index 204bdc4..bc25646 100644
--- a/tasks/todo.md
+++ b/tasks/todo.md
@@ -344,3 +344,11 @@ C3 vertical claim deferred.*
   overall-exceeds needs scale (bigger base/more families/seeds/dilution-resistant merge) = HPC step.
 - Python 3.14 + transformers 5.13 OK; note transformers-5.x apply_chat_template returns a dict.
   `make env-llm`/`make llm`; `figures/plot_llm_merge.py`, README, `tests/test_llm.py` (+3, 125 green).
+
+**2026-07-05 — LLM merge 7B firm-up on Imperial CX3 (`llm_merge_hpc`): marginal sign → decisive.** ✅
+- Ran on one L40S (46 GB) via `/imperial-hpc` runbook; 8 min walltime; Qwen2.5-7B-Instruct, 200 tests/family.
+- **Both merges 0.87 overall > best specialist 0.77** (decisive +10 pts) and beat every specialist on
+  every family; worst-family 0.62 vs ≤0.57. Both 0.5B caveats resolved: overall-exceeds is now clean,
+  and dilution VANISHES (merge 0.62 > lists-spec 0.57 on lists) — dilution was a small-model artefact.
+- `results/llm_merge_hpc/` (README legend, data-driven figure title). Next refinement: module-level
+  union-preserving recombination (MoE-expert/adapter-union = real-weight E8 max-merge), not delta-avg.