Growth Loop Reporter

Role

Close the loop. Three questions per cycle, answered in ≤400 words:

1. What moved? — visibility trend per prompt, cluster, zone
2. Why? — which specific investment caused which lift
3. What next? — 3 prioritized actions, at least 1 stop-doing

Output is a short narrative + actions, not a dashboard. Fifteen charts don't get read. 400 words do.

Input

• project_id — Peec project
• reporting_window — weekly | monthly | quarterly
• optional baseline_date — default 28 / 90 / 180 days back
• optional include_clusters — auto-detected via zone:* tags if peec-cluster has run

Output

• One narrative at <project>/growth_loop/YYYY-MM-DD_report.md (schema below)
• One learnings.json with winners / losers / surprises / next_actions / stop_doing — consumed by the next peec-cluster and peec-outreach runs as priors

When to use

• Weekly for active projects with running content + outreach
• Monthly for retainer projects in maintenance
• Quarterly as strategy review — feeds the next peec-cluster run
• After a launch, publication, or new zone going live

Do not use when:

• Project has <4 weeks of history (too little signal)
• No content or outreach actions in the window (nothing to learn)

Pipeline

0. Pre-flight — setup state required

Per _shared/SETUP_STATE.md, this skill refuses to run without a completed setup:

Read <project>/growth_loop/setup_state.json
If missing OR completed_at missing OR phases_completed lacks
   {competitors, prompts, topics, tags}:
     STOP. Output:
       "No Peec setup state found at <project>/growth_loop/setup_state.json.
        Run /peec-setup first."
If completed_at older than 90 days: WARN once, continue.
Use peec_project_id from state — don't re-resolve via list_projects.

1. Pull time-series of core metrics

# Overall brand visibility trend
mcp__peec-ai__get_brand_report(
  project_id, start_date=baseline, end_date=now,
  dimensions=["date"],
  filters=[{field: "brand_id", operator: "in", values: [own_brand_id]}]
)

# Per prompt (top-N by weight)
mcp__peec-ai__get_brand_report(
  project_id, start_date=baseline, end_date=now,
  dimensions=["prompt_id", "date"],
  filters=[{field: "brand_id", operator: "in", values: [own_brand_id]}]
)

# Per zone (if zone:* tags exist)
for each zone_tag:
  mcp__peec-ai__get_brand_report(
    project_id, start_date=baseline, end_date=now,
    dimensions=["tag_id", "date"],
    filters=[{field: "tag_id", values: [zone_tag_id]}]
  )

Per bucket (prompt or zone) compute:

• visibility_t0 (start of window)
• visibility_t1 (end of window)
• delta = t1 − t0
• trend = linear-regression slope across the window

2. Assemble investment log

# New content
git log --since=<baseline> --author=<user> -- "Content Automation/blog/"
# or: filesystem scan for blog/YYYY-MM-DD_*/

# Outreach
Read: <project>/outreach/*_outreach_log.md
# all pitches with status != 'queued' in the window

# Taxonomy changes in Peec
mcp__peec-ai__list_prompts + list_brands + list_tags
# diff against a snapshot from the start of the window (if one exists)

Produce: one list of investments with date | type (content|outreach|taxonomy) | target (prompt_id or url) | description.

3. Match investment → lift

• Content investment → prompts whose focus_keyword is referenced in the HTML body
• Extract focus keyword from publish_<slug>.py (RANK_MATH_FOCUS)
• Match against list_prompts via embedding or string-contains
• Outreach investment (citation live) → prompts where target_url appears in get_url_report
• mcp__peec-ai__get_url_report(filters=[{url in [target_url]}])
• Zone intervention → all prompts with the zone tag

4. Compute attribution per investment

attribution_score =
    sum(affected_prompts[p].delta for p in matched_prompts)
  - baseline_drift

baseline_drift = median delta of non-affected prompts in the same window. This isolates the intervention effect from general drift.

5. Detect patterns (three buckets)

Winners (high attribution):

• Which content type (HOW_TO / COMPARISON / PILLAR) moved the most
• Which outreach target class (EDITORIAL / UGC / REFERENCE) produced most citations
• Which zone grew fastest

Losers (negative or zero attribution despite investment):

• Content published but not indexed / cited
• Pitches with no response after 14 days
• Zones stagnant despite new content (→ content misses the intent layer)

Surprises (positive delta without a direct investment):

• Prompts that gained without direct action (organic spillover from another page?)
• Sudden drops (competitor action? algorithm shift?)

6. Generate the narrative

Claude synthesizes a narrative ≤400 words using the schema below.

7. Persist learnings

Save to <project>/growth_loop/YYYY-MM-DD_learnings.json — used by the next runs of peec-cluster and peec-outreach as priors.

Narrative schema

# Growth loop — <project> (<window>)

## Headline
<One sentence: what's the most important insight of this period?>

## What moved
- Overall visibility: X% → Y% (<N pp>)
- Strongest zone: <name> (+Z%)
- Weakest zone: <name> (flat or −)
- Top-3 single-prompt lifts: <list>

## What actually worked
<2–3 sentences. Not "the content plan" — but: "Article X became a citation in 11 of 15
target prompts; the retainer pitch at evergreen.media produced Y citations within 10
days; the Shopify zone grew organically without new content there — likely spillover
from zone Z.">

## What did not work
<1–2 sentences: which investment had zero effect, and the most plausible reason.>

## DO NOW (prioritized, max 3)
1. <concrete action with deadline>
2. <concrete action>
3. <concrete action>

## STOP DOING
- <pattern that was identified as time-waste>

Learnings JSON schema

{
  "period": {"start": "...", "end": "...", "window": "weekly"},
  "overall_visibility_delta": 0.04,
  "winners": {
    "content_types": [{"type": "HOW_TO_GUIDE", "avg_lift": 0.08, "n": 2}],
    "outreach_domains": [{"domain": "evergreen.media", "citations_gained": 5}],
    "zones": [{"zone_tag": "retainer-decision", "lift": 0.12}]
  },
  "losers": {
    "content_types": [],
    "outreach_domains": [{"domain": "<...>", "response_rate": 0.0}]
  },
  "surprises": [],
  "next_actions": ["..."],
  "stop_doing": ["..."]
}

Quick reference

Done criteria (self-check before returning)

A growth report is only complete when:

1. Narrative is ≤400 words — longer reports aren't read and usually hedge
2. Attribution is reasoned, not guessed — every winner / loser needs a causal mechanism, not just correlation
3. Exactly 3 next-actions — not 7, not 1. Three is the weekly capacity ceiling
4. At least 1 STOP DOING — the courage to discard is worth more than new ideas
5. learnings.json persisted — without it, no loop

Guardrails (do not do these)

• Do not ship a dashboard — a dashboard is not a report
• Do not sell correlation as causation — visibility went up; competitor also had an SSL outage
• Do not ignore baseline drift — without a comparison group, every lift is suspect
• Do not skip the STOP DOING line — addition without subtraction fragments energy
• Do not ship a report without persisting learnings — the next cycle can't learn
• Do not run this before 4 weeks of history — too little signal, pattern detection degenerates to noise