EventTrader[Gem Screener]

What we changed and why

• Rules and backtest: unchanged, on purpose (same gates_version as upstream). The pre-registered backtest is inconclusive and the gate-passing rows did worse than the rest over six months. Without a new pre-registered hypothesis, any "improved" gate would be fitted to the same six months of data, so we kept the ranking exactly as published and the forward picks ledger stays comparable.
• Collection moved to our servers and made continuous. Upstream's 6-hour CI never ran (GitHub Pages needs a public repo). Here the collector runs every 6 hours in its own process under a host-wide lock, and any reader that finds a stale snapshot starts a refresh.
• A snapshot that fails an audit is never published, and upstream's known gap is closed: a failing snapshot used to become the next run's baseline; now the last published snapshot is copied back as the baseline.
• Live push instead of reloads. A socket delivers each new audited snapshot and the page re-renders in place. No polling.
• Presentation. English only, standards mode (upstream ran in quirks mode), the platform's dark palette with the chart series recoloured to it, wide charts that scroll inside their box on phones, and a Chains table that fits its box (stacked Trajectory cell, one row of chips per project).
• Data. The same DeFiLlama, CoinGecko and DexScreener endpoints and the same universe rules. Nothing added, nothing dropped.
• Source revision. Rebuilt from github.com/zakutana/gem-screener at commit 2c365a7 (2026-09-23 06:35 UTC, the last commit before the request was filed). The one later upstream commit, 69d7253 (2026-09-24), changes only the picks-ledger data file, so every source file the build started from is that revision's; ARCHITECTURE.md is carried verbatim. Repeat visits revalidate with an ETag and get a 304 until a new snapshot is published.
• Flags next to the name became icons with a tooltip, so the Project / Chain column stays narrow — the trust, no-holder-share and risk flags beside a project or chain name are now small icons; hover or Tab to one for the reason. Category, market share and theme stay text chips in one row, so the sector reads at a glance.

What this version does not do

• It does not predict prices or tell you to buy. The pre-registered backtest on the Start tab came out inconclusive, and the gate-passing rows did worse than the rest over 6 months. Read the shortlist as a research starting point, not a signal.
• It does not trade, hold funds or connect a wallet. Nothing on this page places an order.
• It does not show tick-by-tick prices. Market caps and revenue come from the last audited snapshot. A new snapshot is collected every 6 hours and pushed to this page the moment it passes its audits.
• It never publishes a snapshot that fails an audit. The page keeps the last one that passed and shows its time in the header.
• It does not rank by trust, float or unlocks. Those are tags you can read, never part of the order.
• It covers only tokenized projects that DeFiLlama tracks. Apps without a token are counted in the sector totals but are not listed.

Data: DeFiLlama, CoinGecko and DexScreener public APIs. The screening engine is MIT-licensed open-source code, collected and audited on EventTrader servers. Not investment advice.