Add grouped Live Feed to the Scientist dashboard

.github/workflows/organism_board.yml

@@ -1,19 +1,12 @@
 name: Dashboard
 
-# Publishes the PyAutoScientist Dashboard — the umbrella ROUTER over the four organ
-# dashboards (scripts/organism_board.py): one row per board with its own live
-# headline, a "where to work next" banner keyed off the Heart's verdict,
-# published to GitHub Pages + badge.json + the README strip between the
-# scientist:begin/end markers. Nothing is recomputed here — each organ's
-# board stays the authority; this page only routes.
-#
-# Six-hourly: the organ boards refresh on their own daily/push cadences, so
-# the router follows within hours. board_tests.yml checks fixture renders on PRs;
-# this workflow builds and publishes the live page.
+# Scientist owns a bounded public development activity index. Owner repositories
+# remain authoritative. Five-minute scheduling is best effort; the page exposes
+# source coverage and event/collection times separately.
 
 on:
   schedule:
-    - cron: "17 */6 * * *"
+    - cron: "*/5 * * * *"
   workflow_dispatch:
 
 permissions:
@@ -29,6 +22,7 @@ jobs:
   board:
     name: Render + publish the dashboard
     runs-on: ubuntu-latest
+    timeout-minutes: 4
     environment:
       name: github-pages
       url: ${{ steps.deployment.outputs.page_url }}
@@ -54,10 +48,44 @@ jobs:
 
       - run: python -m pip install pyyaml
 
+      - name: Restore bounded activity index
+        run: |
+          mkdir -p _activity _site
+          if git ls-remote --exit-code --heads origin refs/heads/scientist-activity-cache > /dev/null; then
+            git fetch --depth 1 origin refs/heads/scientist-activity-cache
+            git show FETCH_HEAD:cache.json > _activity/cache.json
+            git rev-parse FETCH_HEAD > _activity/parent
+          else
+            status=$?
+            if [ "$status" -ne 2 ]; then exit "$status"; fi
+          fi
+
+      - name: Collect public development activity
+        env:
+          GH_TOKEN: ${{ secrets.PAT_PYAUTOLABS || github.token }}
+          ACTIVITY_BUDGET: ${{ secrets.PAT_PYAUTOLABS && '180' || '60' }}
+        run: |
+          python scripts/activity_feed.py --repos PyAutoMind/repos.yaml --cache _activity/cache.json --output _site/activity.json --budget "$ACTIVITY_BUDGET"
+
+      - name: Preserve bounded derived index
+        if: github.ref == 'refs/heads/main'
+        run: |
+          git config user.name 'github-actions[bot]'
+          git config user.email '41898282+github-actions[bot]@users.noreply.github.com'
+          activity_blob=$(git hash-object -w _activity/cache.json)
+          activity_tree=$(printf '100644 blob %s\tcache.json\n' "$activity_blob" | git mktree)
+          if [ -f _activity/parent ]; then
+            activity_parent=$(cat _activity/parent)
+            activity_commit=$(git commit-tree "$activity_tree" -p "$activity_parent" -m 'Refresh derived public activity index')
+          else
+            activity_commit=$(git commit-tree "$activity_tree" -m 'Begin derived public activity index')
+          fi
+          git push origin "$activity_commit:refs/heads/scientist-activity-cache"
+
       - name: Render every surface (public HTTPS only)
         run: |
           mkdir -p _site
-          python scripts/organism_board.py --repos PyAutoMind/repos.yaml --html     > _site/index.html
+          python scripts/organism_board.py --repos PyAutoMind/repos.yaml --activity _site/activity.json --html > _site/index.html
           python scripts/organism_board.py --repos PyAutoMind/repos.yaml --badge    > _site/badge.json
           python scripts/organism_board.py --repos PyAutoMind/repos.yaml --md       > board.md
           cp board.md _site/dashboard.md

README.md

@@ -129,3 +129,7 @@ work and authoritative records remain with each organ.
 ## Connected projects
 
 The dashboard includes expandable project cards linking software, shared dependencies, documentation, tutorials, workspaces, assistants and owner records. See [the project configuration contract](PROJECTS.md) to connect resources; an assistant is optional and source records remain with their owners.
+
+The Scientist **Live Feed** groups recorded public development activity by task,
+with five compact rows and expandable history. Its coverage, refresh behaviour
+and limits are documented in [REPORTING.md](REPORTING.md#live-feed-recorded-development-activity).

REPORTING.md

@@ -84,3 +84,54 @@ A requested summary is read-only. Follow-up work uses the owning workflow and
 existing approval rules; the request does not authorize compute, replies,
 merges or releases. Do not add recurring collection, a second task registry or
 persistent monitoring to produce a one-off report.
+
+## Live Feed: recorded development activity
+
+The canonical Scientist dashboard has a grouped Live Feed between Projects and
+Dashboards. It covers public development issues, issue/PR discussion updates,
+explicit progress reports, closed tasks and merged PRs in Mind's registered
+repositories. Library/workspace tasks route to Mind; organ development routes
+to that organ. Scientific run progress, private work, local edits and chat-only
+work are outside this first version. Independent instance previews do not yet
+include the feed.
+
+One task occupies one row. A PR is grouped with an issue only when its body has
+one unambiguous standalone `Closes #N`, `Fixes #N` or `Resolves #N` directive for
+that same repository and the target issue is in the collected evidence. Other
+PRs remain separate. An issue being opened is not evidence of work starting;
+a closed PR is not a merge, and a merge is not publication or science acceptance.
+
+Comment bodies are never exported. Recognized headings (`Progress`, `Progress
+update`, `In progress`, `Blocked`, `Shipped`, `Workspace PR Created`, `Library PR
+Created`) produce short factual stage labels and source links. Other human
+comments appear as “Discussion update recorded”; bot comments are excluded.
+Editing a comment replaces that event without giving it a new posted time.
+The source link is the place to read detail, not an automatic transcript mirror.
+
+Collection runs in the existing Scientist Pages workflow approximately every
+five minutes; scheduling and upstream visibility can delay it. The browser
+checks the published same-origin `activity.json` every minute while visible.
+There is no GitHub credential or per-visitor GitHub request in the page. A new
+snapshot is buffered while a reader has expanded details, focused a control,
+scrolled into the feed, or loaded older rows. “Show updates” applies it explicitly.
+
+The `scientist-activity-cache` branch holds a derived, normalized cache and
+per-repository cursors. It is not a task registry and contains no API response
+bodies, prompts, tokens or raw comments. Only public repositories independently
+verified in the current owner catalog are exported. Failed/inaccessible catalogs
+suppress their cached rows. Other source failures retain last-known evidence,
+mark it stale and leave the cursor unchanged. Losing the cache starts a disclosed
+seven-day backfill; successful collection retains up to 30 days, 1500 events per
+repository, 200 published task rows and 20 displayed events per task.
+
+The collector uses `PAT_PYAUTOLABS` when available (180 requests per cycle), or
+the workflow token (60). It rotates repositories when the request budget is
+exhausted; a busy source or pagination cap is reported in coverage rather than
+claimed complete. The issue and comment endpoints paginate up to three pages
+of 100 each. Partial batches retain verified events and reuse already normalized PR
+revisions on the next attempt. A source cursor advances only after both
+endpoints and the needed PR details have succeeded. Each cycle also has a
+two-minute collection time budget. Therefore “every five minutes” is the collection
+attempt interval, not a guarantee that every repository is current. Event time,
+source check time and page check time remain distinct. Routine feed rebuilds
+never create activity events.