🎲 Up-and-Coming Forecaster

An agent that forecasts Roblox's Up-and-Coming chart top 10 — a Gaussian rank-diffusion model, running live in your browser on real archive reads.

How it works

• Drift by rank bucket. Each game's rank is a random walk. The drift (rank change per 6h) is estimated from every pair of consecutive reads, separately for rank buckets 1–3, 4–10, 11–20, 21–40, 41–200 — top games cool off, mid-chart games climb.
• Own velocity. The drift is blended 70/30 with the game's own rank velocity over the last 40 hours (clamped to ±15 ranks per 6h).
• Diffusion. Uncertainty widens with the horizon (σ² = 2.0² + 0.7·hours). Per-place probabilities are the time-average of each game's Gaussian mass on that place over the 72h window.
• Unnamed mass. 30% of every place is left unnamed: about a sixth of top-10 slots in the archive go to games never seen before.
• Eligibility taper. Games whose 28-day Up-and-Coming eligibility ends inside the window are tapered by the fraction they survive.

The forecast above is computed live in this page by forecaster.js, a faithful port of the Python agent (forecaster.py in the repo). The port is verified against the Python original — same inputs give the same top 10 and probabilities within 1.1e-6 (run node verify.cjs in demo/). No server, no tracking, no data leaves your browser.

Data: 34 hourly chart reads Oct 30 – Nov 7, 2026 plus 10 dev reads Nov 8–10, from Roblox's public chart API via the MIT-licensed Kenny-Dhub/roblox-snapshots archive (see data/NOTICE.txt). Source code: github.com/Goatleaf12/roblox-up-and-coming-forecaster