Gaiya Lab · Public research notebook
Every piece starts from response headers, JS bundles and archived snapshots — not from the marketing page. And it does not end at publication: a script re-checks each target every week and records what actually moved.
Re-checked every Monday · produced by scripts, not typed in by hand
URL in, ad creative out. Watching the size of the SEO matrix and that social-proof number that has not moved in three months.
since 2026-08-21 · 1 days watched · re-checked every Monday
A Chinese team’s network of overseas tool sites. Watching for the day the paywall goes up, and how many front doors the network grows.
since 2026-08-21 · 1 days watched · re-checked every Monday
Paste a URL, get ad creative. The product is the easy part — the asset is the 534-page SEO net around it, 171 pages of which never mention AdsCreator in the title. This revision found something the first pass missed: the video feature it now leads with does not exist in its pricing or its API docs.
The product is copyable. The traffic net and the founder's distribution are not.
TargetAdsCreatorMeasured16SEOGEOAI ads
A transcription tool six weeks old. 186 endpoints, every one annotated in Chinese, sitting on a backend with subscriptions, credits, refunds and dunning already wired — while the site itself has no way to pay. This revision adds the arithmetic on how long it can afford to stay free.
Not a standalone product. One acquisition funnel in a network, currently free on purpose.
TargetTranscribeAudioMeasured18site networkreverse engineeringFastAPI
Finding ideas and finding revenue are not the same search, and they do not use the same sites. Graded by who the number is accountable to: filed, checked, estimated, timely. This revision corrects three wrong fee figures from the first version and demotes Flippa out of the top grade.
sourcesrevenue datamethod
28 subreddits scanned, 21 threads read closely, five patterns that cut across industries and ten buildable web products. Includes the sampling method, the four I threw out, and two assumptions that would invalidate the whole ranking.
demand researchRedditproduct opportunities
Part one was all B2B process. This round took 34 different subreddits into caregiving, health insurance, education, creative work and household logistics. The most valuable thing that came out of it was not the ten openings — it was several practitioners turning up to explain why the obvious product will not work.
demand researchRedditproduct opportunities
Response headers, sitemaps, JS bundles, whatever API docs they left public. What the marketing page claims is one more input to verify, not a source.
Wayback snapshots, version against version. A number that has not moved in three months tells you more than the number itself.
Each teardown ships with a monitoring script that re-runs weekly. Time answers questions a website never will.
Read-only collection and public documents only. No write operations, no management endpoints. What follows are observations; the judgement is yours.