docs(wiki): describe the report's two views (Gallery + Matrix)

The Reusable tooling note in model-selection-spike predated the Matrix view and
only described the live-context Gallery strategy. Document both views and the
two rendering strategies (Gallery: live WebGL contexts + reconcile; Matrix:
static offscreen-renderer snapshots), and bump the sweep to 20 models.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

wiki/log.md

@@ -2,6 +2,16 @@
 
 ## 2026-06-15
 
+### model-selection-spike: two report views (Gallery + Matrix)
+
+- Updated the Reusable tooling section: the report now has a **View** dropdown —
+  Gallery (live rotatable contexts + Theme filter) and Matrix (model × theme
+  comparison table). Corrected the rendering-strategy note: Gallery uses live
+  WebGL contexts (capped + distance-reconciled); Matrix uses static snapshots
+  from one shared offscreen renderer (the app's thumbnail strategy), so all 152
+  cells render without hitting the context cap. Sweep is now 20 models
+  (MiniCPM4.1-8B added).
+
 ### model-selection-spike: point at the tracked eval tool
 
 - Added a **Reusable tooling** section: the spike's 11 throwaway

wiki/model-selection-spike.md

@@ -164,9 +164,11 @@ model test does not start from zero:
   helpers, no model loading), `models.py` (the editable list — add a `(label,
   repo)` tuple), `eval.py` (CLI: one model at a time, download → grammar-enforced
   test → optional `--delete-after` → merge; `--fewshot` for the exemplar A/B),
-  and `report.html` + committed `report_data.js` (the 19-model × 8-theme sweep,
-  so the rotatable 3D report opens populated; the **Theme** dropdown compares
-  every model on one creature).
+  and `report.html` + committed `report_data.js` (the 20-model × 8-theme sweep,
+  so the 3D report opens populated). The report has two views, switched by a
+  **View** dropdown: **Gallery** (big rotatable cards + a **Theme** filter to
+  compare every model on one creature) and **Matrix** (a model × theme
+  comparison table).
 - **`/voxel-model-eval` skill** — the run procedure plus the hard rules from this
   page (grammar always on; ≤32B/MoE total; latency relative-only; valid-JSON ≠
   visual quality; never judge from thumbnails).
@@ -174,13 +176,16 @@ model test does not start from zero:
   (`PYTHON_BIN=/Users/chenzhikai/Documents/Project/build-a-buddy-hf/.venv/bin/python`);
   no deps are added to this repo. Unit tests: `tests/test_voxel_eval_{core,cli}.py`.
 
-Note: the report viewer renders a **grid of many live WebGL contexts** capped at
-12 and reconciled by distance-to-viewport-center on scroll — a different strategy
-from the app's renderer (one shared offscreen thumbnail renderer + a single live
-stage, see [voxel-rendering](voxel-rendering.md)). The naive "evict the furthest
-live context and rely on the IntersectionObserver to re-mount it" blanks
-still-on-screen cards permanently (IO does not re-fire for already-intersecting
-elements); the viewer reconciles the full visible set instead.
+Note: the two views use different rendering strategies, both bounded by the
+browser's ~16 live-WebGL-context cap. **Gallery** mounts a grid of **live
+contexts** (rotatable), capped at 12 and reconciled by distance-to-viewport-center
+on scroll. **Matrix** renders each cell as a **static snapshot** from one shared
+offscreen renderer (the app's thumbnail strategy, see
+[voxel-rendering](voxel-rendering.md)) — flat context count, so all 152 cells show
+at once. Gotcha that drove the Gallery design: the naive "evict the furthest live
+context and rely on the IntersectionObserver to re-mount it" blanks still-on-screen
+cards permanently (IO does not re-fire for already-intersecting elements), so the
+Gallery reconciles the full visible set instead.
 
 ## Methodology lessons