Tasks for CloudX Agent

Review every active A/B test in this workspace.

First, inspect the live inventory to find the active tests. For each test, load its report from the test start through the latest complete reporting day. Compare the control and test variants across revenue, eCPM, fill rate, impressions, requests, sample size, lift, confidence interval, and significance.

Classify each test as one of these states:
- Keep running: the result is not decision-ready.
- Review now: the result is decision-ready or the test status is achieved.
- Investigate: the data is invalidated, traffic is badly imbalanced, or a material guardrail metric has deteriorated.

Write a compact review. Lead with tests that need action. Include the evidence behind each classification and a clear next step. Do not recommend a winner when the data is not decision-ready. If no tests need action, state that briefly.

Prepare the workspace's weekly CloudX revenue review for the last complete Monday-through-Sunday period. Compare it with the preceding complete week.

Use CloudX reporting data. Report revenue, requests, impressions, fill rate, eCPM, clicks, and CTR. Quantify the absolute and percentage change for the measures that support both. Break down the largest revenue and delivery movements by the most useful available dimensions, such as app, ad unit, country, device OS, and demand source.

Lead with a short executive summary. Then cover the main gains, declines, anomalies, and likely drivers. Separate observed facts from interpretations. Note material data gaps or coverage limits. End with up to three concrete follow-up actions.

Check the workspace's delivery health for yesterday, using the latest complete reporting data. Compare yesterday with the same weekday one week earlier. Use the recent seven-day trend as supporting context.

Review requests, impressions, fill rate, revenue, eCPM, clicks, and CTR. Look for material drops, spikes, missing traffic, demand-source outages, or isolated problems by app, ad unit, country, and device OS. Prioritize issues that affect meaningful traffic or revenue. Do not flag ordinary low-volume noise.

If you find an issue, state its scope, size, likely start time, evidence, and the next diagnostic step. If delivery is healthy, give a short status with the key totals.

Analyze finalized Total Publisher Revenue for the last complete seven reporting days. Compare it with the preceding seven complete reporting days.

Use the Total Revenue report to combine revenue from all connected data sources. State its dataCompleteThrough value and do not treat missing mediator measures as zero. Cover total publisher revenue, mediation revenue, CloudX revenue, and share of wallet when available. Quantify the period-over-period change. Explain the main movements by app, demand source, and other useful available dimensions. Distinguish broad revenue coverage from measures with narrower mediator coverage.

Lead with the result and its main driver. Then summarize the revenue mix, the largest positive and negative contributors, notable daily changes, and any data-quality limits. End with up to three actions or questions worth investigating.

Check CloudX-only revenue for the latest fully completed reporting hour. Compare it with the same hour seven days earlier. Use the preceding 24 completed hours to confirm whether the movement is isolated or sustained.

Review revenue together with requests, impressions, fill rate, and eCPM. Break down a meaningful change by app, ad unit, country, device OS, and demand source until you identify the smallest useful scope. Treat low-volume changes as noise unless their absolute revenue impact is important. Do not use provisional ILRD or Total Publisher Revenue for this hourly check.

If you find an anomaly, state the affected period, absolute and percentage revenue change, impacted scope, supporting metric changes, likely cause, and next diagnostic step. Separate evidence from interpretation. If the data is healthy, record a short result without manufacturing an issue.

Check fill rate for the latest complete reporting day. Compare it with the same weekday one week earlier. Use the preceding seven complete days to establish the normal range.

Review requests and impressions with fill rate so traffic-mix changes do not create a false alert. Find the main contributors by app, ad unit, country, device OS, and demand source. Give priority to sustained declines on meaningful request volume. Ignore small segments that do not affect workspace delivery or revenue.

For each material decline, report the baseline and current fill rate, percentage-point change, request volume, affected scope, likely start time, related revenue impact, and next diagnostic step. If fill rate is healthy, record the workspace result and the largest non-material movement.

Review CloudX eCPM for the last complete Monday-through-Sunday week. Compare it with the preceding complete week. Use daily values to distinguish a sustained change from a one-day event.

Review eCPM with revenue and impressions. Break down meaningful changes by app, ad unit, country, device OS, and demand source. Check whether the change comes from pricing, traffic mix, or an isolated segment. Do not treat an eCPM increase as positive when impressions or total revenue fell materially.

Report the workspace eCPM change and the contributors with enough volume to matter. State absolute values, percentage changes, revenue impact, evidence, likely driver, and a practical follow-up. If pricing is stable, record a short result.

Review demand-source performance for the last complete Monday-through-Sunday week. Compare it with the preceding complete week.

For each meaningful demand source, review revenue, impressions, requests, fill rate, and eCPM. Measure its contribution to total CloudX revenue. Look for outages, sustained underperformance, unexpected gains, and concentration changes. Break a material movement down by app, ad unit, country, or device OS when that explains the result. Do not rank a low-volume source as important from percentage change alone.

Lead with demand sources that need attention. State the current and prior values, absolute revenue impact, affected inventory, evidence, likely cause, and next step. Keep routine partner movement in a brief summary.

Review app performance for the last complete Monday-through-Sunday week. Compare it with the preceding complete week.

Rank apps by CloudX revenue and absolute revenue change. Review requests, impressions, fill rate, eCPM, clicks, and CTR for the apps that drive the result. Separate growth caused by more traffic from changes in fill rate or pricing. Check whether one country, ad unit, device OS, or demand source explains a material app movement. Exclude low-volume apps from the main findings unless they show a complete delivery failure.

Summarize the portfolio result. Then list the largest positive and negative app contributors with current values, changes, evidence, likely drivers, and next steps. If no app needs attention, record a concise healthy result.

Check ad unit delivery and monetization for the latest complete reporting day. Compare it with the same weekday one week earlier. Use the preceding seven complete days as context.

Review requests, impressions, fill rate, CloudX revenue, and eCPM for each ad unit with meaningful traffic. Look for missing delivery, sudden request changes, fill-rate deterioration, pricing changes, and isolated demand-source failures. Use app, country, device OS, ad format, and demand source to narrow each issue. Do not flag newly created or low-volume ad units without enough evidence.

For each material issue, state the ad unit, current and baseline values, absolute revenue effect, likely start time, related dimensions, likely cause, and next diagnostic step. If all important ad units are healthy, record a short result.

Review country performance for the last complete Monday-through-Sunday week. Compare it with the preceding complete week.

Rank countries by CloudX revenue and absolute revenue change. Review impressions, requests, fill rate, and eCPM for countries that drive a meaningful movement. Distinguish traffic growth from pricing and fill-rate changes. Break a material result down by app, ad unit, device OS, and demand source. Do not elevate a small country because of a large percentage change alone.

Summarize the largest positive and negative contributors. State current values, absolute and percentage changes, revenue impact, evidence, likely drivers, and next steps. Identify an emerging market only when its scale and sustained growth can affect a decision.

Review finalized Total Publisher Revenue for the previous complete calendar month. Compare it with the complete month before it. Align both periods to whole UTC days and keep each report request within its supported date range.

Use the Total Revenue report across all connected data sources. State dataCompleteThrough for each period. Cover total publisher revenue, mediation revenue, CloudX revenue, and share of wallet when available. Do not treat missing mediator measures as zero. Explain the largest changes by app, demand source, mediation source, and day when those dimensions are available.

Lead with the monthly result and its main driver. Then summarize revenue mix, largest positive and negative contributors, material trend changes, and data-quality limits. Quantify absolute and percentage changes. Separate evidence from interpretation. End with up to three useful follow-ups.

Find high-value opportunities for a new CloudX A/B test using the last four complete reporting weeks. Compare the latest two complete weeks with the two weeks before them.

Review meaningful traffic by app and ad unit. Look for sustained revenue, fill-rate, or eCPM underperformance with enough requests and impressions to support a useful experiment. Use country, device OS, ad format, and demand source to isolate a testable segment. Inspect active and recent A/B tests so you do not suggest a duplicate or conflicting experiment. Keep this review read-only and do not create or modify a test.

For each strong opportunity, state the target inventory, observed problem, supporting metrics, estimated revenue relevance, proposed hypothesis, primary success metric, guardrail metrics, and why the available traffic appears sufficient. Do not recommend an experiment when the problem is transient, the segment is too small, or a direct operational fix is more appropriate.

Check reporting-data completeness for the latest expected complete day.

Inspect the Total Revenue report across all connected data sources. State dataCompleteThrough. Compare source presence and daily revenue coverage with the preceding seven complete days. Also confirm that standard CloudX reporting contains the expected requests, impressions, and CloudX-only revenue for the latest complete day. Distinguish an unsupported mediator measure from a missing source. Do not treat a missing optional measure as zero or as an outage.

For each material gap, state the affected source, app or scope, last complete date, expected data, missing period, reporting impact, and next diagnostic step. If coverage is current and consistent, record a short healthy result.

Review the latest complete reporting data and identify what needs executive attention today.

Use finalized Total Publisher Revenue across all connected data sources for the latest available complete day. State dataCompleteThrough. Compare it with the same weekday one week earlier and the recent seven-day range. Use standard CloudX reporting for delivery measures such as requests, impressions, fill rate, and eCPM. Do not call provisional ILRD or CloudX-only revenue total publisher revenue.

Focus on material revenue changes, delivery failures, demand-source issues, important app movements, and reporting gaps. Quantify impact and identify the smallest useful scope. Lead with the result. Then give no more than three findings with evidence, likely cause, and next action. Omit routine metrics that do not change a decision.