30-Day Test Plan

Before you start, ask me about each [input] below — one question at a time, with ready-made answers to pick from; skip whatever you can already tell.

Build a 30-day test plan for [client: if not listed, call roasr_list_clients and let me pick] on [platform: Meta (default) | Google | both]. Share of spend I can put into tests: [test budget share: 10% (default) | 15% | 20% | custom]. Main goal: [goal: lower CPA (default) | higher ROAS | more volume at the same efficiency | custom].

1. **Load the RoASr method.** Call roasr_load_skill('roasr_frame'), roasr_load_skill('report_format') and roasr_load_skill('creative_intel') and follow them exactly.
2. **Frame the data.** Call roasr_limits, roasr_list_clients and roasr_list_accounts: plan, niche, margin, KPI, account currencies, sync status and dataSince. Last 30 days vs the 30 before is the base window; if it starts before dataSince, say so and narrow it. If currency.mixed is true, never sum money across accounts: size test budgets as shares of each account's spend. A failed or never-synced account gets no tests.
3. **Source 1: audit findings.** roasr_list_findings for this client: open high and medium findings that a test could settle (for example a creative or audience question), not ones that are plain fixes. Fixes go to a separate "do now" list, not into the test plan.
4. **Source 2: campaign facts.** roasr_metrics_campaigns for the base window and the one before it, campaign level, limit 30. Find the biggest spenders and the ones drifting worse. Classify by configuration, not by name; under a campaign budget, ad set numbers show allocation, not a ranking. Rule out tracking problems first: spend with no conversions is a tracking question, not a test.
5. **Source 3: market angles.** roasr_spy_tag_leaderboard with limit 20: angles and formats that likely win in the niche (days_live is a proxy, write "likely"). Take 1–2 the client does not run yet.
6. **Write 4–6 tests.** Each test changes one variable and has: a hypothesis in the form "if we change X, Y will improve because Z", the source (finding, campaign or market), the success metric and the minimum improvement that counts, the test budget as a share of spend (the sum of all concurrent tests must stay within the budget share above), the duration, the stop rule (stop early if spend reaches about 2× the target CPA with no conversions, or run until about 30 conversions per variant, whichever comes first), and what we do with a winner and with a loser.
7. **Lay it out by week.** Weeks 1–4. Do not run two tests that change the same campaign at once. Put the cheapest, highest-confidence tests in week 1. Leave week 4 partly free to confirm a winner.
8. **Be honest about sample size.** If the account cannot reach about 30 conversions per variant inside the duration, say the test will be directional only and propose a larger change or a longer run.

**Output:** one sentence naming the test most worth running first and why. Then a table: week · test · hypothesis · source · success metric · budget share · duration · stop rule. Then the "do now" list of plain fixes and 3–6 actions grouped High / Medium / Nice-to-have, each with the expected effect.

**Output rules:**
- Recommendations only — never apply changes; RoASr tools are read-only.
- One variable per test; no overlapping tests in one campaign.
- Say "likely" for everything inferred from days_live.
- If the client has no margin set, judge tests by ROAS and CPA, not profit.
- Answer in the language of this prompt.