Method
Where every number comes from, how a product “crosses a line”, how briefs are written and graded, and what we refuse to do.
Sources
| Source | Role | Label | History | Last run |
|---|---|---|---|---|
| Appark charts appark.ai/api/charts/top-charts | discovery | verified verified rank | today only for guests: snapshot daily since Nov 18 | ● Nov 18 · 46,297 rows |
| Google Play detail play.google.com/store/apps/details | measurement: exact installs | verified verified | our daily snapshots since Nov 18 | ● Nov 18 · 595 rows |
| iTunes Lookup itunes.apple.com/lookup | measurement: rating count | verified verified count / estimated proxy | our daily snapshots since Nov 18 | ● Nov 18 · 15,050 rows |
| Gravity Engine mini-games rank.gravity-engine.com/rank-data/{wechat,douyin}/{day,week}/ | discovery + measurement | verified verified rank | today only: snapshot daily (public = top 20 per board) since Nov 18 | ● Nov 18 · 60 rows |
| Appark drama appark.ai/api/advertiser/{tiktok-series-list, playlet-list} | discovery | verified verified rank / estimated impressions | drama-ad chart has daily history (we backfill 30 days); series chart today only since Nov 18 | ● Nov 18 · 100 rows |
| TrustMRR trustmrr.com/api/ai | measurement: payment-verified revenue | verified verified | API + our snapshots since Nov 17 | ● Nov 18 · 45 rows |
| GitHub Trending github.com/trending | discovery: stars today, 8 languages | verified verified | our daily snapshots since Nov 17 | ● Nov 18 · 119 rows |
| Tranco tranco-list.eu | measurement: web traffic rank | verified verified | daily lists by date since Nov 17 | ● Nov 18 · 28 rows |
| Reddit (Arctic Shift) arctic-shift.photon-reddit.com | Demand Echo: buyer-intent threads | verified verified counts | API since Nov 17 | ● Nov 18 failed |
| Xiaohongshu via TinyFish TinyFish search | Demand Echo (CN): public 小工具 posts | inferred inferred | CI only, when TinyFish is configured since Nov 18 | ● Nov 18 · 4 rows |
Appark and Gravity Engine endpoints are public but undocumented. We pace requests, publish derived ranks and links, and never re-host raw chart dumps. Appark gives guests today's charts only, so history starts the day we start collecting.
The line a product must cross
A product qualifies when today is at least 3× its own average over the previous 7 days, above an absolute floor. New sources have no history yet: until a product has 3 days in our data, we use the change figure the source itself reports and label the card source delta or cold start instead of own baseline.
| Rule | Metric | Threshold |
|---|---|---|
| Install spike | Daily installs | ≥ 3.0× own baseline and ≥ 300/day |
| Ratings spike | Daily new ratings | ≥ 3.0× own baseline and ≥ 20/day |
| Star spike | Daily new stars | ≥ 3.0× own baseline and ≥ 100/day; cold start ≥ 400 |
| Chart jump | Chart rank | up ≥ 30 places into the top 50 |
| New on chart | Newly on chart | new or improved inside the top 100 |
| Mini-game climb | Mini-game chart rank | up ≥ 5 places into the top 20 |
| Ad surge | Daily ad creatives | ≥ 2.0× own baseline and ≥ 300/day; or ≥ 50% of creatives new today |
| Drama chart | TikTok drama chart | new or improved inside the top 10 |
| Revenue climb | MRR | v2 logic: MRR ×1.3 in 30 days (≥ $100 gain) / Tranco rank ×1.3 in 7 days |
| Traffic climb | Web traffic rank | v2 logic: MRR ×1.3 in 30 days (≥ $100 gain) / Tranco rank ×1.3 in 7 days |
Picking what to brief
score = strength × publisher penalty × freshness × (0.5 + direction fit) × (1 + demand echo) × novelty. Grossing-chart moves weigh 1.4×, paid-chart moves 0.45× (a few sales move a thin paid chart); products launched in the last year weigh 1.3×, older than three years 0.55×. Big platforms are penalised, products briefed in the last 14 days are skipped, each surface gets at least 2 briefs and at most 40% of the 24 daily slots.
Two modes of brief
Data-only (always)
Every field is computed from collected data: badge, curve, charts, measurements, dated events, peers, Demand Echo, discounts, grade, market-gap wedge. The headline describes the move; the trigger reads “not checked” unless evidence exists.
Agent-checked (CI with keys)
TinyFish searches the web for the trigger; an LLM writes the lesson headline, positioning, differentiators, moat, wedge and Build Kit from the gathered evidence only. A verify step downgrades any “fact” whose link we didn't gather to “inferred”. Badges and grades stay computed by code.
Labels
fact linked to a source · verified measured by us or payment-verified · estimated a third party's estimate or a proxy · inferred our reading of the evidence · not found we looked and didn't find it.