qwen3-8b: files, cards and figures (reads through stdlogits/rs-lenses@3708cb59c2)

README.md

@@ -90,6 +90,7 @@ Neurons and heads get the same tables, one row per pole or per head:
 | `drivers` | the tokens that drive each pole hardest | the tokens that drive each head, as query and as key |
 | `legible` | which reads hold at the pole's top firings | which reads hold at the head's top firings |
 | `labels` | Claude Haiku's labels: where it fires, what it writes | where it fires, what it attends to, what it writes |
+| `edges` (Qwen3-8B) | the earlier poles whose writes push it most, and the later poles it pushes most | |
 | `image`, `audio` (Gemma 4) | the patch or sound it reads, its write in media contexts, the images or sounds that drive it, and whether its media label is its text concept too | |
 | `media` (Gemma 4) | | how each head attends in image and audio contexts, and which carry what an image or sound shows |
 
@@ -121,7 +122,7 @@ Neurons and heads get the same tables, one row per pole or per head:
 | `gemma-4-12b-heads-labels` | 768 | every head's labels, by Claude Haiku 5.5 |
 | `gemma-4-12b-heads-media` | 768 | every head in image and audio contexts |
 
-The lenses are [`stdlogits/rs-lenses`](https://huggingface.co/stdlogits/rs-lenses) at revision `b15ef87783`. Code, methods and
+The lenses are [`stdlogits/rs-lenses`](https://huggingface.co/stdlogits/rs-lenses) at revision `3708cb59c2`. Code, methods and
 findings (cited here as §N) are on [GitHub](https://github.com/stdlogits/neuron-lens).
 
 ## The lenses
@@ -289,6 +290,22 @@ labels late (80% picked at L28-L35, 41% at L0-L8). Gemma 4's neuron labels are n
 
 ![Random poles read blind](figures/14_lens_blind.png)
 
+### `qwen3-8b-neurons-edges`: direct weight edges
+
+What a neuron writes reaches every later layer's MLP input unchanged, so each pole has a direct edge to
+every pole of a later layer (§17). Per pole, its 16 strongest in-edges (`in_*`, from earlier poles) and
+out-edges (`out_*`, to later ones), strongest first: `_layer`, `_neuron`, `_pole` name the other pole;
+`_cos` is the cosine of the earlier pole's write with the later pole's read direction, and `_push` is
+`cos · act_top · ‖w‖ / rms`, how far one of the earlier pole's top firings moves the later pole's
+normalized input along its read direction (`rms` the residual's median RMS at the later layer's MLP
+input). Positive: the earlier pole's firing drives the later pole; negative: it pushes against it.
+Edges rank by |push| among those with |cos| ≥ 5/√d = 0.078; `−1` and NaN fill shorter lists.
+
+Edges are wiring, not traffic: a pole's push ranks every upstream neuron's effect when it fires at
+Spearman 0.88, but its in-edges are a tenth of the 16 units that contribute most there, which are mostly
+units firing with it (§17). At L1-L3, 49-78% of top firings have a negative gate, where the read
+direction points the wrong way, and in-edges there are unreliable.
+
 ### `gemma-4-12b-neurons-image`, `gemma-4-12b-neurons-audio`: images and sound
 
 Gemma 4 has no vision or audio encoder network: an image is 64 slots of 48 × 48-pixel patches (384
@@ -505,9 +522,12 @@ Per model folder (`qwen3-8b/`, `gemma-4-12b/`):
   `(layer, neuron)`, 32 random per layer plus named ones, or 1,152 of Gemma 4's, 24 random
   per layer. Per neuron and pole: `steer` `[neuron, pole, dose, vocab]`, the mean change in log-probs
   over positions 16-63 of 16 64-token DCLM contexts (`contexts`) when the write is added there at 1× and
-  4× `alpha` (the pole's typical high activation); `ablate_pos` and `ablate_win`, clean − ablated
-  log-probs with the neuron zeroed at its peak in 8 top documents, at that position and summed over the
-  next 8; `dlogp_next`, the next token's change per document; `fired`, the activation there. bf16.
+  4× `alpha` (the pole's `act_top`); `ablate_pos` and `ablate_win`, clean − ablated log-probs with the
+  neuron zeroed at its peak in its first 8 top documents, at that position and summed over the next 8;
+  `top_doc` (rows of `docs.parquet`) and `top_pos`, where each of those ablations was made, as in the
+  `acts` table (−1 where the pole never fired); `dlogp_next`, the next token's change per document;
+  `fired`, the activation there. Every input is in the file, so it can be re-measured from it alone.
+  bf16 (Qwen3-8B's on MPS); weak effects differ by device (§3).
 - `gemma-4-12b/media/{image,audio}_docs.parquet`: each image or sound document's `source`, `labels`,
   `caption`, `ref` (JSON: the dataset, config, split and revision, the row's own identifiers, the media
   file's path and its bytes' SHA-1) and `link`. The images and sounds are not re-hosted;
@@ -556,4 +576,4 @@ repositories, under their licenses) and
 [UltraChat-200k](https://huggingface.co/datasets/HuggingFaceH4/ultrachat_200k) (MIT). The media
 documents' labels and captions come from their source datasets, under those datasets' terms (some
 non-commercial); the images and sounds themselves are only linked. By [@_a9lim](https://x.com/_a9lim)
-and [@treetowntree](https://x.com/treetowntree) with Claude (Anthropic), 2026-11-01.
+and [@treetowntree](https://x.com/treetowntree) with Claude (Anthropic), 2026-11-05.

qwen3-8b/causal/causal.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ce224de3ef463c7df8d32cb905d64a81f4e42c2f67625399b23d4b196802be96
-size 2902772153
+oid sha256:b632100452a6f8e9f6d9bd66193755845878f5123f34f0053ce1a31eca80773c
+size 2902887425