feat: B4a search.fit breakdown cell, profiling-summary@2 exporter, Pages board, epic-1 table

.github/workflows/lint.yml

@@ -1,9 +1,9 @@
 # lint.yml — PR / push-to-main gate.
 #
-# Cheap, fast, CPU-only. Runs on every PR and on push to main. It lints only what exists
-# in the skeleton: ruff (check + format) and a lychee link check over README.md and
-# AGENTS.md. No PyAuto library is installed and no script runs; the pytest and smoke legs
-# arrive with the harness in a later phase of the search-extensibility epic.
+# Cheap, fast, CPU-only. Runs on every PR and on push to main: ruff (check + format), the
+# stdlib-only tests under scripts/misc/test, the "committed catalogue is current and
+# validates with the PyAutoPulse reader" check, and a lychee link check over README.md and
+# AGENTS.md. No PyAuto library is installed and no measurement cell runs.
 
 name: lint
 
@@ -30,7 +30,7 @@ jobs:
 
       - name: Install lint tools
         run: |
-          pip install ruff
+          pip install ruff pytest pyyaml
           # lychee is a Rust binary; fetch the prebuilt release. The tarball
           # nests everything under lychee-x86_64-unknown-linux-gnu/, so strip
           # that leading directory and extract just the binary.
@@ -43,6 +43,19 @@ jobs:
       - name: ruff format --check
         run: ruff format --check .
 
+      # Checked out after ruff so the Pulse tree is never linted as this repo's code.
+      - name: Checkout PyAutoPulse (the profiling-summary@2 reader)
+        uses: actions/checkout@v4
+        with:
+          repository: PyAutoLabs/PyAutoPulse
+          path: _pulse
+
+      - name: pytest (stdlib-only tooling tests)
+        run: python -m pytest -q scripts/misc/test
+
+      - name: catalogue up to date and valid profiling-summary@2
+        run: python scripts/misc/tooling/build_catalogue.py --check --validate-with _pulse
+
       - name: lychee — markdown link-rot
         run: |
           # Exclude GitHub UI links that lychee tends to false-positive on, and the

.github/workflows/pages_dashboard.yml

@@ -0,0 +1,79 @@
+name: Pages Dashboard
+
+# Publishes the producer page (`dashboard/index.html`), the PyAutoPulse read contract
+# (`dashboard/catalogue.json`, profiling-summary@2) and the project badge feed
+# (`dashboard/state.json`) at https://pyautolabs.github.io/autofit_profiling/. The files
+# are rendered and committed by `scripts/misc/tooling/build_catalogue.py`; this workflow
+# only validates and publishes them, then tells PyAutoPulse to re-read the summary.
+# Mirrors autolens_profiling's pages_dashboard.yml (project-feed identity:
+# PyAutoMind complete/2026/10/pulse-refresh-project-identity.md).
+
+on:
+  push:
+    branches: [main]
+    paths:
+      - "dashboard/**"
+  workflow_dispatch:
+
+permissions:
+  contents: read
+  pages: write
+  id-token: write
+
+concurrency:
+  group: pages
+  cancel-in-progress: true
+
+jobs:
+  deploy:
+    runs-on: ubuntu-latest
+    environment:
+      name: github-pages
+      url: ${{ steps.deployment.outputs.page_url }}
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/checkout@v4
+        with:
+          repository: PyAutoLabs/PyAutoBrain
+          path: _brain
+      - uses: actions/checkout@v4
+        with:
+          repository: PyAutoLabs/PyAutoPulse
+          path: _pulse
+      - uses: actions/setup-python@v5
+        with:
+          python-version: "3.12"
+      - name: Validate the catalogue with the Pulse reader and check it is current
+        run: |
+          pip install pyyaml
+          python scripts/misc/tooling/build_catalogue.py --check --validate-with _pulse
+      # enablement: true creates the Pages site on first run.
+      - uses: actions/configure-pages@v5
+        with:
+          enablement: true
+      - name: Stage the dashboard as the site
+        run: |
+          mkdir -p _site
+          cp dashboard/index.html dashboard/catalogue.json dashboard/state.json _site/
+      - name: Validate the project badge feed against the Brain contract
+        run: python3 _brain/board/_state.py _site/state.json
+      - uses: actions/upload-pages-artifact@v3
+        with:
+          path: _site
+      - id: deployment
+        uses: actions/deploy-pages@v4
+      - name: Tell the PyAutoPulse dashboard to refresh (repository_dispatch pulse-refresh)
+        env:
+          GH_TOKEN: ${{ secrets.PAT_PYAUTOLABS }}
+        run: |
+          if [ -z "$GH_TOKEN" ]; then
+            echo "::warning::PAT_PYAUTOLABS is not available to this repo — skipping the pulse-refresh dispatch to PyAutoLabs/PyAutoPulse."
+            exit 0
+          fi
+          sha=$(git rev-parse HEAD)
+          gh api repos/PyAutoLabs/PyAutoPulse/dispatches \
+            -f event_type=pulse-refresh \
+            -f "client_payload[repo]=${{ github.repository }}" \
+            -f "client_payload[sha]=${sha}" \
+            -f "client_payload[summary]=dashboard/catalogue.json"
+          echo "Dispatched pulse-refresh to PyAutoLabs/PyAutoPulse (${{ github.repository }}@${sha})."

AGENTS.md

@@ -12,8 +12,10 @@ It is a collection of standalone scripts, **not** an installable package — the
 `pyproject.toml`. These are the canonical, agent-agnostic instructions; `README.md` is the
 human-facing overview.
 
-**Status:** skeleton born 2026-10-07 (PyAutoMind#492), phase B1 of the
-`search-extensibility` epic. There is no harness, no exporter and no result yet.
+**Status:** born 2026-10-07 (PyAutoMind#492, phase B1 of the `search-extensibility`
+epic). B4a (2026-10-14) added the first cell — the Nautilus `search.fit` breakdown on
+gaussian_x1 — the `profiling-summary@2` exporter and the Pages board; the repo reports to
+PyAutoPulse as instance `fit`. The EP / graphical baseline port is B4b.
 
 ## Where this repo is going
 
@@ -26,7 +28,7 @@ The work arrives in later phases of the `search-extensibility` epic, whose ledge
   `fit` (`profiling-summary@2`). This adopts Pulse's task `autofit_profiling_bootstrap`.
 - **B4b** — the EP / graphical-model baseline port.
 
-Until then, do not add scripts, results or exporters outside the phase that owns them.
+Do not add scripts, results or exporters outside the phase that owns them.
 PyAutoPulse validates the published summary contract; the Brain's profiling conductor
 judges it. Nothing here issues a verdict.
 
@@ -35,10 +37,33 @@ judges it. Nothing here issues a verdict.
 ```
 ruff.toml        lint config AND the root sentinel (leaf scripts walk up to it)
 activate.sh      RAL shared venv + PYTHONPATH (PyAutoNerves, PyAutoFit only)
+config/          PyAutoConf overrides (library output defaults kept on purpose)
+scripts/<dataset>/<search>/<cell>.py   measurement cells, e.g. gaussian_x1/nautilus/search_fit_breakdown.py
+scripts/misc/profiling/                stdlib helpers shared by cells (exclusive component timer)
+scripts/misc/tooling/build_catalogue.py  results/ -> dashboard/{catalogue.json,index.html,state.json}
+scripts/misc/test/                     stdlib-only tests (no PyAuto install)
+results/         committed result rows (schema autofit-profiling-breakdown), one file per cell variant
+dashboard/       generated: catalogue.json (profiling-summary@2), index.html, state.json
+wiki/campaigns/  findings pages; search_fit_bottlenecks.md is epic 1's ranked table
 hpc/             hpc/sync (laptop-side RAL driver) + the batch_cpu submit template
 ```
 
-`scripts/`, `results/` and the summary exporter arrive in B4a.
+## Producing a result and publishing it
+
+1. `source activate.sh`, run a cell (e.g. `python scripts/gaussian_x1/nautilus/search_fit_breakdown.py`);
+   it writes its row under `results/`. Laptop CPU cells need no RAL.
+2. `python scripts/misc/tooling/build_catalogue.py --validate-with ../PyAutoPulse` renders
+   `dashboard/` and validates the catalogue with the Pulse reader; commit rows and
+   `dashboard/` together. The renderer is deterministic (its stamp is the newest
+   evidence time), so `--check` is the lint gate.
+3. On merge, `.github/workflows/pages_dashboard.yml` publishes `dashboard/` to Pages
+   (`https://pyautolabs.github.io/autofit_profiling/`) and dispatches `pulse-refresh` to
+   PyAutoPulse (payload repo / sha / `dashboard/catalogue.json`; needs the org secret
+   `PAT_PYAUTOLABS`, warns and skips without it). `dashboard/state.json` is this
+   project's own badge feed; the Brain cockpit reads the PyAutoPulse organ feed.
+
+Every catalogue record is declared `unreviewed`: the producer measures, the Brain's
+profiling conductor judges.
 
 ## RAL
 
@@ -53,8 +78,9 @@ detail in [`hpc/README.md`](hpc/README.md).
 ## Testing
 
 The PR gate is `.github/workflows/lint.yml` on Python 3.12: `ruff check .`,
-`ruff format --check .` and a `lychee` link check over `README.md` and `AGENTS.md`. It
-installs no PyAuto library and runs no script.
+`ruff format --check .`, `pytest -q scripts/misc/test`, `build_catalogue.py --check
+--validate-with _pulse` (PyAutoPulse checked out beside) and a `lychee` link check over
+`README.md` and `AGENTS.md`. It installs no PyAuto library and runs no measurement cell.
 
 ## Related Repos
 

README.md

@@ -9,10 +9,12 @@ It is the sibling of [autofit_inference](https://github.com/PyAutoLabs/autofit_i
 (PyAutoFit search benchmarking) and the PyAutoFit counterpart of
 [autolens_profiling](https://github.com/PyAutoLabs/autolens_profiling).
 
-**Status: skeleton — no results yet.** The repo was born 2026-10-07 as phase B1 of the
-[search-extensibility epic](https://github.com/PyAutoLabs/PyAutoMind/blob/main/draft/research/autofit/search_extensibility_epic.md). The breakdown exporter and first results arrive in
-B4a, when the repo starts reporting to
-[PyAutoPulse](https://github.com/PyAutoLabs/PyAutoPulse); the EP baseline port follows in
-B4b.
+**Status:** the first cell — where the time in one Nautilus `search.fit` goes on a fast
+1-D Gaussian likelihood — landed in B4a of the
+[search-extensibility epic](https://github.com/PyAutoLabs/PyAutoMind/blob/main/draft/research/autofit/search_extensibility_epic.md),
+with its ranked table in [`wiki/campaigns/search_fit_bottlenecks.md`](wiki/campaigns/search_fit_bottlenecks.md).
+The repo reports to [PyAutoPulse](https://github.com/PyAutoLabs/PyAutoPulse) as instance
+`fit` (`dashboard/catalogue.json`, `profiling-summary@2`); its own board is published to
+GitHub Pages from `dashboard/`. The EP baseline port follows in B4b.
 
 Agents and contributors: start with [`AGENTS.md`](AGENTS.md).

config/general.yaml

@@ -0,0 +1,4 @@
+# PyAutoConf overrides for runs launched from this repo. Output keeps the library
+# defaults on purpose: a breakdown cell measures what a user's search.fit pays.
+version:
+  workspace_version_check: false    # this repo tracks PyAutoFit main, not a release tag; the mismatch warning is expected noise.