seo-rhythm-daily — Two Moments, Every Day

Every day, two things happen. That's it. More than two daily moments produces thrash and blurs signal.

Moment 1 — 06:05: Data collection

Invoke seo-monitor with today's date. Expected output:

• reports/seo/raw/gsc-{YYYY-MM-DD}.json
• reports/seo/raw/ga4-{YYYY-MM-DD}.json
• reports/seo/raw/indexing-{YYYY-MM-DD}.json
• Appends a row to reports/seo/data/kpi-history.json

• [ ] GSC data has ≥ 100 rows (if not, API auth likely broken)
• [ ] GA4 sessions > 0 (if not, tracking likely broken)
• [ ] Indexed-page count is within ±20% of yesterday (if not, trigger seo-postmortem)

06:07 ✅ GSC=1243rows GA4=412sessions Index=324pages

Moment 2 — 10:07: CTR repair sprint

This is where the agent ships 1–2 experiments per day. It is not optional and it is not more than 2.

Why 1–2 not 5: each experiment needs to breathe. Shipping 5 at once on 5 different pages makes attribution messy when a Google algo update lands mid-week.

Step 1: Read memory

Before picking a candidate, read:

• memory/playbooks/.md — which patterns are proven?
• memory/lessons/.md — which patterns are known bad?
• memory/experiments/.yaml where status=open — don't double-ship a page already mid-experiment

priority_score = impressions × (benchmark_CTR_at_position - actual_CTR)

• Pages currently in an open experiment (experiments/.yaml has this page with status=open)
• Pages listed in recent lessons as "don't touch"
• Pages already optimized in last 60 days (cool-off period)

Step 3: For each candidate

1. Match playbook — which playbook applies (title-number, desc-frontload, etc.)?
2. Log experiment — invoke seo-experiment-log with hypothesis derived from playbook evidence
3. Ship change — invoke seo-optimize.apply(page, change, playbook_name)
4. Commit + deploy — the change only counts as "shipped" when deployed

Step 4: Update dashboard

Append to reports/seo/dashboard.md the day's 2 experiments with their measure_at dates, so the owner sees the pipeline filling.

Step 5: Return

Tell the caller:

10:12 ✅ Shipped 2 experiments
  exp-2026-04-16-a7k: /alternatives/cursor (title-number playbook)
  exp-2026-04-16-b3m: /compare/openrouter-vs-litellm (desc-frontload playbook)
  Measure on: 2026-04-30

Edge cases

• Skip the sprint. Note in daily-status.log: 10:12 ⏸ No candidates (all in cooldown)
• Don't invent experiments just to hit a count.

• Skip that candidate. Pick the next one.

• Skip the 10:07 sprint entirely. The postmortem takes priority.
• Note: 10:07 ⏸ Deferred — postmortem-{id} in progress

Quality bar

• [ ] Both moments ran (or a documented reason for skipping)
• [ ] Every shipped change has a corresponding experiment file
• [ ] No changes shipped for pages in cool-off
• [ ] daily-status.log has both entries

What I refuse

• To ship more than 2 experiments per day. It pollutes attribution.
• To ship without consulting playbooks + lessons. That defeats the whole learning loop.
• To skip Moment 1 because "we don't need data today". Data compounds.
• To ship at 10:07 when 06:05 flagged a regression. Fix the leak before pouring more water in.

Integration

Called by:

• Tycoon RoutineScheduler (cron rule: 5 6 and 7 10 )
• Standalone cron: ~/.claude/scripts/seo-daily.sh

• seo-monitor (Moment 1)
• seo-optimize + seo-experiment-log (Moment 2)
• seo-postmortem (if regression)