DropForecast reads upcoming toy & sneaker drops, scores each on its odds of doing a hard multiple over retail, and cross-references live store stock so you know what to grab and where it's sitting — before the crowd.
Real portfolio P&L + today's drops · nightly refresh
Loading live data…
Your two live finds carry real sold comps; the pattern cards show the COP / HUNT / PASS logic.
Real 2026 USPS + marketplace numbers. On a $50 sale of a ~2 lb boxed figure, free-shipping model.
| Venue | All-in fee | Net @ $50 | Net @ $55 |
|---|---|---|---|
| Mercari cheapest | 10.0% | $35.00 | $39.50 |
| Whatnot | 11.5% | $34.25 | $38.70 |
| eBay deepest pool | 14.4% | $32.80 | $37.12 |
Gradient-boosted trees on fundamental data, calibrated so 80 literally means ~80% odds.
LightGBM classifier predicts P(net ≥ 2× within 6–12 mo), passed through isotonic calibration so the 0–100 is a real probability. A quantile regressor prints the P10 / P50 / P90 band. Brand-new drops with no price history get a k-NN cold-start prior from comparable past drops. Anchored to the Duke hedonic study — R² ≈ 0.65 on 349,556 StockX transactions.
Exclusivity tier, edition size / chase ratio, retirement & discontinuation. Restock −0.95 is the single biggest premium-killer in the hedonic data (verified, Duke Table 2).
Depreciating "hype stock" (δ ≈ 5%/day), bid/ask depth imbalance, watcher & #bids velocity, Google-Trends momentum, rolling franchise-heat index.
Celebrity collab +2.22, brand collab +1.54, but athlete signature −0.87 (annual re-release dilutes). Low retail multiplies easier — a $12 → $200 is 16×.
Sell-through rate (liquidity — good) separated from raw volume (abundance — bad). Price-trajectory slope, realized volatility, pre-release spread width.
Vault / retirement proximity, franchise events (film, show, album, license death), celebrity-wear moments — the time-varying triggers that re-score instantly.
A place that just exists, that you can pull up and breathe. Free for everyone, ad-free for a buck, loaded for the pros.
The hook nobody else has: Net-in-hand for this exact item, in your size, ranked across every venue's real fees — plus forward-looking restock risk (odds the maker re-drops and tanks your inventory). StockX/eBay will never show it; it cuts against them. Pure compute over free comps — that's the paid magnet.
Same muscle as BuildForecast + bb-watch. Scrape-disciplined, CLI-first, phone-alerted.
Trained on 99,956 free StockX sales. Walk-forward, no leakage. Duke collab sign reproduced (+0.63). Precision 73% @ COP, +107pt edge vs baseline. Live tonight.
Fleet scrapes free RSS (Toyark, McFarlane, JustFreshKicks, Sneaker News) + eBay Browse API (free 5k/day) + actionfigure411 + 130point sold comps. Zero paid APIs.
DuckDB feature store → LightGBM + isotonic calibration → quantile band → cold-start prior. Ships as dropcast score <sku>.
Target RedSky (free) returns per-store stock by DPCI. Photos: Wikimedia → /kie renders (owned, no logos). lilbro enriches every item on-device at $0.
bb-watch's ntfy pattern fires a push on a Flip Score ≥ 70 near you. Stripe subscription flips the ad-free flag for Pro.
This board on dropforecast.com — today's ranked drops, your portfolio, velocity, and the aging buckets that keep capital turning.
The math (COGS ≈ $0, so it's all traffic): at 250K monthly pageviews / ~40K active → ads ~$1.2k + payers ~$3.5k ≈ $4.3k/mo. At 1M PV / 150K active → $17–26k/mo (~$200–310k/yr). Payments are ~90% margin — every bit of SEO traffic drops near-straight to the bottom line. That's the family money.