Find newly launched Meta ads by exact start date.

Cost guard: before each paid Spytrend call, run the free `estimate_cost` with the same tool and arguments, estimate the cost and stay within the user's stated budget; if the user has not stated a budget, tell them the estimate before the first paid call. If pricing consent is required, show the returned terms and stop until the customer explicitly agrees; never accept on the customer's behalf. If a required call does not fit the stated budget, make no paid call and return `Budget limit prevents required research`; omit an optional call that does not fit and disclose reduced coverage. When opening ads with `get_ad`, skip multilang ads (`is_multilang`) and never call `get_media` after `get_ad`. Retry once only when `retryable=true`, after `retry_after_seconds`. Read `params.ignored` after every search; an ignored required filter never counts as applied. If a search reports `pagination.total_status=restricted` (a free-plan archive window; `meta.plan_window_cutoff` names the boundary), tell the user the result covers only ads that started on or before that date, and never report a zero from it as "no such ads exist". Cost notices are not part of the final customer report.
Use the connected Spytrend MCP to prepare a polished customer-facing dashboard of newly launched Meta ads for the supplied controlled configuration.

Execute every research stage in this prompt in one turn. Apart from mandatory pre-call cost notices, do not emit intermediate step responses and do not stop after validation. Stop early only for invalid input, unavailable required evidence or unresolved required identity. Return only the final customer report.

This workflow verifies recent launches using returned Facebook start dates. It is not a scaling, reuse, first-seen, competitor-discovery or performance workflow.

Configuration selected on the page:
- Research focus: Consumer AI Apps | category=mobile_apps_and_productivity | query=AI | allow=consumer-usable AI apps and web apps for a concrete personal task | exclude=B2B-only software, agencies, services, courses, hardware, vague AI claims and unresolved offers
- Country (ISO-2): US
- Launch window: last 7 completed UTC days
- Media: video
- Status: active (`active` = active now; `all` = any returned status)

Reading the configuration:
1. The research focus has the form `Label | category=<slug> | query=<value or none> | allow=<allowed offers> | exclude=<excluded offers>`. Use the encoded category, query, allowed offers and exclusions only in calls and checks; in the customer answer use only the label before the first `|`, the full country name and plain media/status words. Accept no customer-written query, brand, synonym, translation, date, offer rule or extra category.

Launch semantics:
2. Exclude the open current UTC date. The newest included date is the previous completed UTC date. Include exactly 7 completed calendar dates.
3. `Newly launched` means only that the returned Facebook start date falls inside this inclusive window. Never substitute Spytrend first-seen/indexed dates, last-seen dates, advertiser age, creative age, observation recency or current status.

Research:
4. Use Meta only and work read-only.
5. Heavy calls are search_ads, search_creatives, get_trends, search_webmasters and get_ads_analytics. Run at most one heavy call at a time per account, wait for its response, and never batch, parallelize or use Promise.all for heavy calls.
6. Calculate the window: the last 7 completed UTC days ending yesterday. Run one search_ads call for each included date, newest to oldest and strictly sequentially, with date_from=date_to=that date and identical filters:
   - source=`meta` when exposed;
   - categories containing the category encoded in Consumer AI Apps | category=mobile_apps_and_productivity | query=AI | allow=consumer-usable AI apps and web apps for a concrete personal task | exclude=B2B-only software, agencies, services, courses, hardware, vague AI claims and unresolved offers;
   - query=the query encoded in Consumer AI Apps | category=mobile_apps_and_productivity | query=AI | allow=consumer-usable AI apps and web apps for a concrete personal task | exclude=B2B-only software, agencies, services, courses, hardware, vague AI claims and unresolved offers only when its value is not `none`;
   - countries containing US;
   - media_type=video;
   - status_today=`active` only when status is `active`; omit it when status is `all`;
   - dedupe=false;
   - sort_by=`date`;
   - sort_order=`desc`;
   - limit 10 for a 1-day window, 4 per date for 3 days, 2 per date for 7 days.
7. Use the connected schema's scalar or array shape without changing canonical values. Do not use first_seen filters, exclusive country matching, broader fallbacks or generated queries. Never remove the controlled query or substitute another taxonomy when a route fails.

Verification:
8. After all daily searches finish, open at most six likely relevant rows with get_ad(source=`meta`, id=<ad id>), one at a time, preferring at least one from each non-empty date. Prioritize explicit focus fit from the search row, then the most recent Facebook start date and identity/domain clarity. Light calls may run only after heavy discovery finishes. Never pass ad IDs to get_creative. A candidate that fails verification is excluded; open the next ranked one within the six-opening limit and stop once five ads are verified.
9. A selected ad requires current-run evidence of an in-window Facebook start date, selected media, selected status when constrained, US in returned delivery countries, and an explicit allowed Consumer AI Apps | category=mobile_apps_and_productivity | query=AI | allow=consumer-usable AI apps and web apps for a concrete personal task | exclude=B2B-only software, agencies, services, courses, hardware, vague AI claims and unresolved offers offer.
10. Apply allowed and excluded offer rules literally. Taxonomy membership alone is insufficient. Exclude missing, ambiguous or out-of-window start dates and unclear offers without padding the result.
11. Do not call get_advertiser in standard mode. Show an advertiser name only when retained evidence explicitly returns that name; never derive it from the title, page, domain or URL. Otherwise write `Advertiser not returned`. When first-party ownership is not explicit in retained ad detail, use `Product site` or `Other external site` rather than upgrading identity.
12. Show 3–5 verified ads; show fewer only when fewer passed verification and say so.

Interpretation:
13. Order by most recent Facebook start date, direct offer fit, identity strength and diversity. Do not invent a score.
14. Write `Started [Month D, YYYY]`. Never describe first-seen/indexed data as launch or claim a new product, company, market entry or first-ever campaign.
15. Country evidence proves presence, not exclusivity or dominance. Active means active when returned, not continuous delivery since start.
16. Attribute benefits, prices, discounts, results and testimonials as advertiser claims. Never infer spend, reach, engagement, installs, conversions, revenue, ROAS, profitability, virality, leadership or performance.

Links and failures:
17. Label external destinations `Landing page`. Use `Evidence` only when MCP explicitly identifies a stable first-party research resource link. A merchant destination, image CDN, preview or media URL is never `Evidence`. Never name the service in the customer answer or construct routes from IDs. Omit missing previews and summarize global preview absence once under Research coverage. Never print signed temporary URLs or call get_media by default.
18. Retry the required search exactly once only when `retryable=true`, after `retry_after_seconds` and a fresh identical estimate. A final search failure is unavailable evidence and makes the result Partial. A completed zero-row route is valid evidence. Candidate-level failure or nonqualification excludes that candidate without changing a completed discovery route to Partial.
19. Complete requires every daily discovery route to finish. If no ad qualifies, return an honest zero result without changing category, focus, country, window, media or status.

Customer-facing presentation:
20. Use headings, short labeled blocks and metric lines. Keep the customer answer within 450 words. Never use Markdown/HTML tables, backslash line breaks, raw field names, calls, JSON, IDs, quotas, credentials, schema details, rejected rows or audit data.
21. Start with `# Newly Launched Meta Ads: [focus label]`, using only the label part of Consumer AI Apps | category=mobile_apps_and_productivity | query=AI | allow=consumer-usable AI apps and web apps for a concrete personal task | exclude=B2B-only software, agencies, services, courses, hardware, vague AI claims and unresolved offers.
22. Then show `**[N] verified launches** · [full country name] included · [Image/Video] · [formatted inclusive window] completed UTC days · [Active now/Any returned status] · **[Complete/Partial]**`.
23. With two or more results, start with a compact Shortlist, then one compact launch block per ad, maximum 70 words before links.
24. Destination types: `Verified first-party site` only with explicit advertiser/profile and normalized registrable-domain alignment; `Product site` for product/domain alignment without verified ownership; exact app-store label for App Store or Google Play; otherwise `Other external site`.
25. Each compact launch bullet may include Started date, status/media, advertiser, offer/message, one Why review it action, CTA, destination type and Landing page when supported. Compress or omit optional wording to stay within about 40 words per item and the overall word limit.
26. With two or more results, add two or three sample-specific Patterns.
27. Under `## Research coverage`, use at most two bullets for date/window/status, returned and inspected row counts, and sample/link/preview limits. Describe bounded discovery as `[N] rows were returned across [D] daily searches`; never say `top` unless the route explicitly applied meaningful ranking.
28. For zero results, use `## Result` to say no ad met both the Facebook start-date and offer-verification rules. Omit Shortlist, details and Patterns.
29. Before sending, verify all dates and labels match, Complete is justified, every result satisfies the controlled configuration, no line ends in a backslash and no audit content appears.

End exactly:
`Facebook start dates show when the returned ads began delivery, not when the product, offer, advertiser or creative first entered the market. Meta archive activity is not proof of spend, performance, conversions, revenue, ROAS or profitability.`

Only after a separate user message exactly requesting `Show supporting record`, provide a non-sensitive record from retained responses with window, calls, outcomes, counts, selected/rejected evidence, elapsed time and usage. Never rerun research.