Candidates

Ranking competitors by active ad count taken from a keyword scrape is wrong by orders of magnitude; query each advertiser's own ad-library page listing instead, and always pair the total with how many of those ads are older than 60 days.

A niche table built from ad-library query counts is unreliable until the ad bodies are read: ambiguous keywords silently import unrelated advertisers and manufacture both demand heat and longevity that do not exist.

Both codbrain store templates ship a non-SAU COD Network payload that the API rejects with 422, and because the push is deliberately best-effort the failure is silent — every order from a non-Saudi store is dropped while the customer sees a success screen.

Morocco COD has a single dominant price anchor at 199 MAD (~$20), and the long-running advertisers overwhelmingly run WhatsApp/Instagram funnels rather than owned-domain branded stores.

The codbrain cosmetics store template silently drops every order in non-Saudi markets — its non-SAU branch posts a payload shape the COD Network API rejects with 422, and because the push is best-effort the shopper still sees a success screen, so the store looks perfectly healthy while taking zero orders.

In Kuwait joint-pain COD, static ads have effectively no presence — the exact-match query "كريم المفاصل" returns zero live image ads against 32 live video ads, and across four joint-pain queries video outnumbers static roughly 30:1, so the standard "test the angle cheaply with statics, then scale with video" sequence has no static market to test in.

Vercel CLI 58.9.0 `vercel env add` silently prepends a UTF-8 BOM (U+FEFF) to piped stdin values, corrupting every secret set this way — the COD_NETWORK_API_TOKEN it produces makes /api/orders throw a 500 on every order, so a store deployed this way takes zero orders while looking fully live.