prompt: route searches-readme-dashboard (#138) → library-dev (cloud)

active.md

@@ -1,5 +1,17 @@
 # Active Tasks
 
+## searches-readme-dashboard
+- issue: https://github.com/PyAutoLabs/autolens_profiling/issues/138
+- session: Claude Code cloud/mobile (no local worktree)
+- status: library-dev
+- worktree: none — cloud session works on the GitHub clone directly
+- repos:
+  - autolens_profiling: feature/searches-readme-dashboard
+- prompt: active/searches_readme_dashboard_no_data.md
+- note: Phase 0(e) of the inference programme (PROGRAMME.md). Root cause verified:
+  build_readme.py scans a flat filename layout; searches artifacts live in nested
+  cells with bare-tier filenames, and the renderer reads a stale payload schema.
+
 ## stored-sample-reconstruction-guard
 - issue: https://github.com/PyAutoLabs/PyAutoFit/issues/1486
 - status: library-dev — WORKSPACE HALF SHIPPED; the PyAutoFit hardening (#1486) is what remains

active/searches_readme_dashboard_no_data.md

@@ -0,0 +1,43 @@
+# searches README dashboard renders "No data yet" despite existing artifacts
+
+Type: bug
+Target: autolens_profiling
+Repos:
+- @autolens_profiling
+Difficulty: small
+Autonomy: supervised
+Priority: medium
+
+Phase 0(e) of the inference programme
+(`autolens_profiling/results/notes/inference/PROGRAMME.md`): the
+`searches-nautilus` auto-table in `scripts/misc/searches/README.md` renders
+"_No data yet — run `searches/nautilus/{simple,jax}.py` to populate._" even
+though `results/searches/nautilus/**` holds many committed artifacts
+(imaging mge/delaunay/pixelization, point_source, cluster, group cells).
+
+Root cause (verified 2026-08-18 by reading
+`scripts/misc/tooling/build_readme.py`): the artifact scanner only matches
+flat versioned filenames (`<script>_summary_v<version>.json`, ARTIFACT_RE) or
+config-tagged filenames with a script prefix (`<script>_<config>.json`,
+CONFIG_TAGGED_RE) directly under `results/searches/<sampler>/`. The searches
+framework instead writes a nested cell layout:
+
+```
+results/searches/<sampler>/<dataset_class>/<model>/<instrument>/<tier>.json
+e.g. results/searches/nautilus/imaging/mge/hst/hpc_a100_fp64.json
+     results/searches/nautilus/point_source/image_plane/simple/default.json
+     results/searches/multi_start_adam/group/mge/hst/local_local_gpu_fp64.json
+```
+
+Neither regex matches bare `<tier>.json` names, so `_render_nautilus_table`
+receives zero artifacts and emits the no-data block.
+
+Task: teach `build_readme.py` to scan the nested searches layout (cell path →
+sampler/dataset_class/model/instrument, filename → hardware tier; note the
+`hpc_hpc_a100_fp64.json` / `local_local_gpu_fp64.json` double-prefix and
+`default.json` variants in the existing data), render the searches table from
+the real artifacts (all samplers present, not just nautilus), and regenerate
+the affected READMEs. Keep the sentinel-block mechanism and hand-written
+prose untouched; keep `ruff check` / `ruff format --check` green. Verify by
+running the generator and eyeballing the produced table against known rows
+(e.g. nautilus imaging/mge/hst A100 fp64).