Site Factory

* stages/runners.js — replace stub bodies for brand/UX/UI with real
  local-only runners that produce committable artifacts:
    stage 2 (brand) → sites/<d>/brand/brand-spec.md
    stage 3 (UX)    → sites/<d>/brand/sitemap.md
    stage 4 (UI)    → sites/<d>/app/styles/brand-tokens.css

* stages/runners.js stage 8 idempotency — check site_factory.stages for
  an existing ga4_property_id BEFORE calling create_property.py. Previous
  pipeline re-runs created 4 duplicate GA4 properties (538344557 +
  538360131 + 538390749 + 538356405) that had to be soft-deleted
  manually. Subtle bug — initial fix filtered status='done' but runStage
  flips status to 'running' before invoking the runner, so the row never
  matched. Now filters on the presence of ga4_* fields in output_json,
  which only get written by a real successful run.

* sites/{wholivedthere,bubbesblock,claimmyaddress}.com/brand/{brand-spec,sitemap}.md
  and sites/.../app/styles/brand-tokens.css — the first 9 real artifacts
  produced by the now-real pipeline. Each one is keyed to the site's
  rotating palette (Chanel / Hermès / Gucci).

Smoke: re-ran stage 8 on all 3 sites after the fix — all returned
reused=true with the original property IDs, no new properties created
in GA admin.

* sites/bubbesblock.com/app/lib/gtag.js   — G-CBE5QSCLXW (property 537957623)
* sites/claimmyaddress.com/app/lib/gtag.js — G-WVSFP6J11J (property 538392114)

Generated by the now-fixed stages/runners.js audit-trail flow — both
runs landed 11 rows each in site_factory.actions_log, confirming the
postJSON BASIC_AUTH fix from commit fa82160. Pipeline state for all 3
sites is now current_stage=11 (all stages done, status=intake at the
overall-lifecycle level).

* sites/wholivedthere.com/app/lib/gtag.js — measurement id G-VV80JM88WB
  (GA4 property 538258118, created via stage 8 `/analytics` skill)

* stages/runners.js — postJSON to /actions and /findings now forwards
  the orchestrator's BASIC_AUTH credential. Was silently 401'ing every
  audit-trail write (req.on('error') swallowed it), so site 1's first
  full pipeline pass landed 0 rows in actions_log. Smoke-tested by
  re-running stage 0: actions_log went 0 → 1 with the expected row.

The com.steve.sf-cncp-panel launchd plist still exists at
~/Library/LaunchAgents/ but can't be loaded from non-Aqua shells
(launchctl bootstrap gui/$UID returns error 125 'Domain does not
support specified action'). The launchd job was throttled to oblivion
after the May-7 freeze and never came back.

Same script (scripts/cncp-panel-data.js), now run by pm2 with
cron_restart='*/1 * * * *' + autorestart:false. Verified: cron fires
~30s past every minute boundary, JSON file mtime advances, runs exit 0.

Legacy plist left in place but inert — safe to delete once Steve
confirms nothing else references it.

- apply db/schema.sql (palettes/sites/stages/manager_findings/actions_log/analytics_events all missing)
- seed 20 fashion-house palettes via scripts/seed-palettes.js
- add global error middleware + express-async-errors so async pg throws no longer leave responses hanging
- wire BASIC_AUTH into ecosystem.config.js env (was unset → process.exit(1) on every restart → max-restart kill → silent absence)
- smoke: GET /health 200, GET /sites {sites:[]}, GET /palettes count=20

.env was committed before .gitignore exclusion was in place. File remains
on disk and functional; this stops future commits and prevents the next
git push (if any remote is ever added) from leaking secrets.
No remote currently configured, so no public exposure occurred.