Conversion Tracking Health Check

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.

Check whether conversion tracking works for the ad accounts of [client: if not listed, call roasr_list_clients and let me pick], and give a verdict per account: tracking OK, in question, or broken.

1. **Load the RoASr method.** Call roasr_load_skill('roasr_frame') and roasr_load_skill('root_cause') 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. Check [window: last 30 days (default) | last 14 days | last 60 days | custom]. If a window starts before dataSince, say so and narrow it. An account with sync failed or never gets the verdict "no data". Work account by account with accountId; never sum accounts of different currencies.
3. **Clicks versus conversions.** roasr_metrics_daily per account. List days with clicks and zero conversions, and the click-to-conversion ratio by week. A ratio that was stable and then collapses on one date is the classic sign of a break: name the date. A small account may legitimately have zero-conversion days: judge by the noise floor from the method, not by one empty day.
4. **Action buckets.** roasr_metrics_actions per account for the window and for the half before it. Check the funnel is coherent: landing_page_view and add_to_cart present when purchases are claimed, purchases not far above add_to_cart, a lead or purchase bucket that appears or vanishes on one date. Buckets are not additive and engagement is an aggregate of its parts: never sum them. Purchases with revenue at zero mean the value is not passed.
5. **Google conversion actions.** For Google accounts, roasr_signals with kind google_conversion_actions: each action's status and category, whether it counts toward bidding, and its recent volume as far as the payload shows. Flag enabled actions with no recent conversions, several primary actions for the same goal (double counting), and a bidding goal that points at a minor action.
6. **Duplicates and spikes.** Days with conversions above about 3x the window median while clicks are flat suggest double firing; say "suspected", not "proven".
7. **Verdict per account.** OK: conversions follow clicks, funnel coherent, actions healthy. In question: one flag from steps 3–6 or too little volume to judge. Broken: conversions at zero with live clicks, or a funnel bucket that vanished, or duplicate firing confirmed by several signs.

**Output:** open with one sentence: how many accounts are OK, in question or broken and the first one to fix. Then a table: account · platform · click-to-conversion trend · funnel check · Google actions check · suspected issue · verdict. Then 3–6 actions grouped High / Medium / Nice-to-have, each saying where to look by hand and the expected effect.

**Output rules:**
- Recommendations only — never apply changes; RoASr tools are read-only.
- RoASr sees reported conversions, not your site or tag setup: say what to verify by hand and where.
- With no margin set for the client, talk ROAS and CPA, not profit.
- "Tracking is fine" is a valid answer; never invent a defect.
- Answer in the language of this prompt.