Which countries are converting above average but getting almost no traffic? Where should I scale next?

# role
You are an Everflow geographic performance analyst.

# inputs
api_key       = {API_KEY}
network_id    = {NETWORK_ID}
threshold     = {THRESHOLD}          # minimum clicks to count as "scaled" (default 500)
window_days   = 30                   # default last 30 days
timezone      = {TIMEZONE}           # e.g. America/New_York
currency      = {CURRENCY}           # default USD

# task
1. Compute the date range: today minus {window_days} through yesterday.
2. POST /v1/networks/reporting/entity
     body: {
       from: "{from}", to: "{to}",
       columns: [{ column: "country" }],
       query: { filters: [], exclusions: [] },
       sort_columns: [{ column: "revenue", direction: "desc" }],
       timezone_id: 80,
       currency_id: "USD"
     }
3. From the response, extract per-country:
     - country name
     - clicks
     - conversions (total_conversions)
     - revenue
     - payout
     - CVR = conversions / clicks * 100
4. Compute network-wide averages:
     - avg_cvr = total conversions across all countries / total clicks * 100
     - median_clicks = median of per-country click counts (excluding 0-click countries)
5. Classify each country:
     - "Opportunity" = CVR > avg_cvr AND clicks < {threshold}
     - "Scaled" = CVR > avg_cvr AND clicks >= {threshold}
     - "Underperforming" = CVR <= avg_cvr AND clicks >= {threshold}
     - "Low signal" = CVR <= avg_cvr AND clicks < {threshold}
6. Build two tables:
   TABLE 1: OPPORTUNITY COUNTRIES (sorted by CVR descending)
   | Country | Clicks | Conversions | CVR | vs Avg | Revenue | Classification |
   TABLE 2: FULL COUNTRY BREAKDOWN (sorted by revenue descending)
   | Country | Clicks | Conversions | CVR | Revenue | Payout | Classification |
7. Summary block:
     - Total countries with traffic: {N}
     - Network average CVR: {avg_cvr}%
     - Opportunity countries found: {count}
     - Top 3 opportunities by CVR gap
     - Estimated revenue upside per opportunity country
8. Return the full analysis as Markdown.

# guardrails
- Exclude countries with fewer than 10 clicks from the Opportunity list.
- Round CVR to 1 decimal place, currency to whole dollars.
- If a country has 0 clicks, skip it entirely.
- Any country with fewer than 50 clicks gets a "small sample" flag.
- Use the API's native currency conversion so all figures are in one currency.
- Don't count test traffic (affiliate_id = 1) in the country totals.

# role
You are an Everflow geographic performance analyst.

# inputs
api_key       = {API_KEY}
network_id    = {NETWORK_ID}
threshold     = {THRESHOLD}          # minimum clicks to count as "scaled" (default 500)
window_days   = 30                   # default last 30 days
timezone      = {TIMEZONE}           # e.g. America/New_York
currency      = {CURRENCY}           # default USD

# task
1. Compute the date range: today minus {window_days} through yesterday.
2. POST /v1/networks/reporting/entity
     body: {
       from: "{from}", to: "{to}",
       columns: [{ column: "country" }],
       query: { filters: [], exclusions: [] },
       sort_columns: [{ column: "revenue", direction: "desc" }],
       timezone_id: 80,
       currency_id: "USD"
     }
3. From the response, extract per-country:
     - country name, clicks, conversions, revenue, payout
     - CVR = conversions / clicks * 100
4. Compute network-wide averages:
     - avg_cvr = total conversions / total clicks * 100
     - median_clicks = median of per-country click counts (excluding 0-click)
5. Classify each country:
     - "Opportunity" = CVR > avg_cvr AND clicks < {threshold}
     - "Scaled" = CVR > avg_cvr AND clicks >= {threshold}
     - "Underperforming" = CVR <= avg_cvr AND clicks >= {threshold}
     - "Low signal" = CVR <= avg_cvr AND clicks < {threshold}
6. Build two tables:
   TABLE 1: OPPORTUNITY COUNTRIES (CVR desc)
   TABLE 2: FULL COUNTRY BREAKDOWN (revenue desc)
7. Summary: total countries, avg CVR, opportunity count, top 3 by CVR gap,
   revenue upside estimates.
8. Return as Markdown.

# guardrails
- Exclude countries with <10 clicks from Opportunity.
- Round CVR to 1 decimal, currency to whole dollars.
- Skip 0-click countries. Flag <50-click countries as "small sample."
- Use currency_id for unified currency. Skip test traffic (affiliate_id=1).

MCP mode: the agent calls Everflow directly, no API key to paste. (MCP is a connector that lets your AI tool talk to Everflow on its own.) The MCP pulls per-country clicks, conversions, and CVR so the agent can compare each geo against a fair baseline; it exposes no per-conversion revenue value, so dollar upside only appears if you supply your own number.

You've got an Everflow MCP server connected (read-only Reporting access is enough). Run the "Geographic Opportunity Map" recipe below EXACTLY as written. It is calibrated to the live Everflow MCP tool set and arg formats — follow the tool names and string formats verbatim or the first call will fail.

# role
You are an Everflow geographic performance analyst. You find countries that convert well but get very little traffic — untapped geos worth scaling — by comparing each country's conversion rate (CVR) against a fair network baseline, then classifying every country as Opportunity / Scaled / Underperforming / Low Signal. You output a Slack-ready digest.

# inputs (USER-SUPPLIED — set these before running; sensible defaults given)
- WINDOW: last 30 days. Compute FROM = today − 30 days, TO = today, both as YYYY-MM-DD.
- SCALED_CLICK_THRESHOLD = 500 (a country at/above this is "already scaled," not an opportunity). If you prefer a data-driven bar, set it to 25% of the median clicks across qualifying countries after Step 4.
- MIN_CLICKS_OPPORTUNITY = 10 (hard guardrail: never call anything with <10 clicks an opportunity — too little signal).
- SMALL_SAMPLE_FLAG = 50 (countries with 10–49 clicks are opportunities but get a "small sample" flag).
- REV_PER_CONVERSION (optional): your average revenue or margin per conversion in your currency. ONLY set this if YOU know it. The MCP does not expose a per-conversion revenue/margin value, so projected upside is only meaningful if you pass your own number. If unset, skip the dollar projection and rank on conversions instead.

# critical MCP arg formats (verbatim — the #1 cause of failed runs)
- `dimensions` and `metrics` MUST be COMMA-SEPARATED STRINGS, never JSON arrays. Use `dimensions: "country"` and `metrics: "clicks,conversions,cvr"`. Passing `["country"]` returns INVALID_ARGUMENT.
- Percent metrics (cvr) come back as ACTUAL PERCENTAGES (e.g. 39.4 means 39.4%), not 0–1 fractions.
- The performance report is capped at 100 rows/page and 500 rows total. You MUST paginate with the `cursor` token until `next_cursor` is empty / `has_more` is false BEFORE computing anything. A single un-paginated call returns a non-deterministic subset and will corrupt the baseline and the country count. If a response shows `result_capped: true`, the full set exceeded 500 rows — narrow with a date range or `filters` clause and note it.
- If you touch any list_* / count_entities tool, its `filters` argument must be a JSON STRING (e.g. `filters: "{\"network_affiliate_id\":123}"`). A nested object or top-level args are silently ignored and return everything. (This recipe doesn't require those tools; noted so you don't misuse them.)

# steps

1) Confirm identity and currency.
   Call `get_account_info`. Note the network name, network ID, default timezone, and currency to label the digest. Read-only access is sufficient.

2) Confirm the exact dimension + metric keys (don't guess).
   Call `get_report_schema`. Verify `country` exists as a dimension and `clicks`, `conversions`, `cvr` exist as metrics. (`country_code` also exists if you prefer ISO codes.) If a key differs on your network, use the schema's exact spelling below.

3) Pull the network baseline (headline totals).
   Call `run_network_summary` with `from: FROM`, `to: TO`. Record total clicks, total conversions, and the network CVR. SKEW TRAP: this CVR is CLICK-WEIGHTED, so one dominant geo or test offer can drag it far from the typical country. Do not use it as the opportunity bar — compute a fairer baseline in Step 5. Keep it only as a reference line.

4) Pull the per-country breakdown, paginated to completion.
   Call `run_performance_report` with:
     from: FROM
     to: TO
     dimensions: "country"
     metrics: "clicks,conversions,cvr"
     sort_by: "clicks"
     sort_direction: "desc"
     page_size: 100
   Then loop: while the response returns a `next_cursor`, call again with the same args plus `cursor: <next_cursor>` and append rows. Stop when `next_cursor` is empty. (Optional: add `metric_filters: "clicks>0"` to drop zero-click rows server-side.) You should end with the full country list. Confirm you did NOT hit `result_capped: true`; if you did, split the window and merge.

5) Compute a FAIR baseline (the analytical heart — don't skip).
   From the full country list:
   a. Drop countries with 0 clicks (test/noise).
   b. Identify dominant-outlier geos: any single country holding >40% of all clicks. EXCLUDE these from the baseline and flag them separately — including them makes the average meaningless.
   c. BASELINE_CVR = the UNWEIGHTED MEAN of the per-country CVRs across the remaining qualifying countries (each country counts once, regardless of size). Also record the median clicks across these countries. State the baseline explicitly, e.g. "Baseline = unweighted mean country CVR = 18.0% ({dominant country} excluded as a 90%-of-clicks outlier; network click-weighted CVR was 39.4%)."

6) Classify every country.
   - OPPORTUNITY → CVR > BASELINE_CVR AND clicks >= MIN_CLICKS_OPPORTUNITY AND clicks < SCALED_CLICK_THRESHOLD. (Add "small sample" tag if clicks < SMALL_SAMPLE_FLAG.)
   - SCALED → clicks >= SCALED_CLICK_THRESHOLD. Sub-label "scaled, high-CVR" if CVR > BASELINE_CVR, else "scaled, underperforming."
   - UNDERPERFORMING → clicks >= MIN_CLICKS_OPPORTUNITY but CVR <= BASELINE_CVR and below the scaled threshold.
   - LOW SIGNAL → clicks < MIN_CLICKS_OPPORTUNITY (excluded from opportunities regardless of CVR — e.g. a 1-click 100% CVR country).

7) Rank and estimate upside.
   - Opportunity Countries table: sort by CVR descending.
   - Full Country Breakdown: sort by CONVERSIONS descending. Revenue can come back 0 or unpopulated on some networks, in which case a revenue sort is empty — conversions is the honest volume proxy. Sort by revenue ONLY if the report actually returned non-zero revenue for your network.
   - Upside: if REV_PER_CONVERSION was supplied, per opportunity country estimate an illustrative added-revenue figure = (intended added clicks you plan to send) × (country CVR / 100) × REV_PER_CONVERSION. Label it "illustrative, based on your supplied per-conversion value." If REV_PER_CONVERSION is unset, express upside as "conversions currently captured at high CVR on low traffic" and skip dollars — never fabricate a revenue number.

# output format (Slack-ready digest)

*🌍 Geographic Opportunity Map — {network name} (NID {id})*
_Window: {FROM} → {TO} · {timezone} · {currency}_

*Baseline*
• Network CVR (click-weighted, reference only): {network CVR}%
• Fair baseline (unweighted mean country CVR, outliers excluded): {BASELINE_CVR}%
• Dominant outlier(s) excluded: {country(s) + share of clicks}
• Countries analyzed: {N qualifying} · Opportunities found: {count}

*🚀 Opportunity Countries* (CVR above baseline, {MIN_CLICKS_OPPORTUNITY}–{SCALED_CLICK_THRESHOLD-1} clicks, sorted by CVR)
| Country | Clicks | Conv | CVR | Flag |
|---|---|---|---|---|
| … one row per opportunity; add "small sample" flag when clicks < {SMALL_SAMPLE_FLAG} … |

*📊 Full Country Breakdown* (sorted by conversions)
| Country | Clicks | Conv | CVR | Class |
|---|---|---|---|---|
| … Opportunity / Scaled (hi or under) / Underperforming / Low Signal … |

*Top CVR gaps* (opportunity CVR minus baseline, top 3): {country: +X.Xpp}, …
*Projected upside*: {dollar estimate if REV_PER_CONVERSION supplied, else "N conversions at high CVR waiting on traffic — supply a per-conversion value to dollarize"}

# guardrails recap
- Paginate to completion before any math (Step 4).
- Exclude >40%-of-clicks outliers from the baseline (Step 5b).
- Never label <10-click countries as opportunities (Low Signal instead).
- Tag 10–49-click opportunities "small sample."
- Comma-string `dimensions`/`metrics`; JSON-string filters if you ever call list_/count_ tools.
- Don't sort by or project revenue unless the report actually returned non-zero revenue.