What we changed and why

Version 3, part 4 (ET-28378, 2026-09-29): the Degen meter and the coin detail.

• Degen column. Every risk flag beside an app’s name became one meter: the level as a number and ten segments, dashed where the level is not certain. Click the header to sort by it: high to low, low to high, then back to the table’s own order. The level sorts only there and changes nothing else.
• Chart column. The last 90 daily closes of the coin’s DEX pool and the Chart Analyzer’s score for them, read once a day after the daily close, on the last closed candle, matched by contract and chain. The column cannot be sorted.
• Coin detail, rebuilt for apps: level, Chart card, monthly revenue with the usual month (the median of the previous 12 full months), the 18 checks, supply, six numbers, trajectory, then the Degen report. Chains keep their detail.
• Where the flags went. Micro is the Market cap check, Thin liquidity is Slippage, Cliff, % in 90d and low float are Unlocks, Revenue bought with emissions is Real yield, Post-crash, Declining and No trend are Trend, No holder share is Reaches you; New and Stale data widen the level’s range.
• Copy buttons in the detail for what an AI agent needs: the coin’s API address, its contract, the Chart Analyzer link and a ready prompt.
• Wording. Where the request says grey, the page uses EventTrader’s muted blue-grey: plain grey is off our palette.

Version 2 (ET-27671, 2026-09-27), every line of the follow-up request:

• Built for degens who will not read. Paragraphs that do not change a buy decision moved here, into About; every tooltip is 20 plain words or fewer with the maths behind “How it’s calculated”.
• First load. The page carries the ranked tables and the Top-10 chart’s 10 weekly series; logos follow right after first paint, and a row’s details load when you open it.
• Header and status line. the Gem Screener logo (“Revenue-driven AI Research”, “by Cymetica”; a click opens cymetica.com), then only “Updated … · apps · chains · Live”.
• One benchmark line and a Legend next to the tabs; clicking a flag next to a name opens the Legend on that flag.
• One colour rule in the table, cards, detail and charts: green good for a buyer, light blue caution, red bad, grey neutral or missing.
• Strength 3M (three months instead of six) in the collector, the table and the detail.
• Real or printed. Each app’s 30-day revenue minus the tokens it paid out as incentives (DeFiLlama). A red “Revenue bought with emissions” flag in the table and the detail. A flag only: the pre-registered gates are unchanged. Since replaced: revenue is shown as revenue, and new tokens are their own figure, Dilution (the larger of last-30-day emissions and next-30-day unlocks against market cap, per week, gross), flagged against the benchmark app.
• Plain labels and badges in the detail, a big Trade on EventTrader button when the token has a pair on our exchange, and a small “tradeable here” icon in the table.
• Top Picks tab: altseason index, the Gem Screener Top Picks fund, then the Degen picks; the near misses waiting in line, and the backtest in one line.
• Degen report: one button in each coin’s detail pre-fills NEXUS with the report prompt; you press Enter.
• Incomplete data never shows a URL or a log line: prices carried over from the previous update are counted in one plain line under the header.
• For AI agents: small endpoints for the picks, a ranked list and one coin, MCP tools, a markdown view of /gem-screener and a Gem Screener section in llms.txt.

Version 1 (ET-26608):

• 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.
• Altseason panel (ET-28378, part 1). The old Altseason Index card and strip are replaced by one panel, the same on Top Picks and on Sectors: five tiles (Altseason cycle, OTHERS.D, BTC.D, Retail, Volume), a slider that is always visible with a red tick at the top zone, and one tall chart per tile. Built from the customer’s specification at commit ec7e5ff of github.com/zakutana/gem-screener; the calculation is their open-source method, re-run by us on data we fetched ourselves.
• Our review of the Altseason method. Kept as specified: the index formula, the top zone at 75, the six cycle words, the trend-line rules, the spike cleaner and the retail score. Changed: every code and value in the data is English; every word and its colour is decided by the back end; the “AI questions on crypto” row says “not available”, because no public feed with history exists that we can verify; Coinbase’s best traffic rank comes from the ranking history we store, not from a typed-in number. Weak points you should know: three past tops is a small sample, and the rules were tuned on the same history they are tested on; OTHERS.D here follows CoinMarketCap’s ranked list and reads lower than TradingView’s; two sources are public web endpoints without a service promise, so a tile can go stale, and the panel then says so. A second, separate program recalculates every week of the index and blocks the update when any number, word or trend line disagrees.

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.
• It shows caution in light blue, not amber. EventTrader’s palette has no amber, orange or yellow, so the caution colour is the platform’s light blue; green, red and grey are as requested.
• It cannot say “real or printed” for every app. DeFiLlama publishes a token-emission schedule for about half of the listed apps; the rest read “unknown”, never zero.
• It does not write the Degen report with a language model on the page. The card prints what the report service returns; when there is no report it says so, and NEXUS can write one in the chat from the pre-filled prompt.
• The Chart score is a reading of a chart, not advice and not a forecast. It is read from DEX pool candles, which the analyzer’s study did not measure. The table shows the daily score only; a weekly score needs 60 weeks of candles, so few coins have one. A coin without a pool, a contract or enough candles shows a dashed line and says why in its detail.
• It cannot show who holds the tokens that are not out yet (insiders or others), how many tokens are minted, or what the mint is worth at today’s price. The sources give the float and the mint in dollars only, so those lines read “not available”.
• It does not let the Degen meter or the Chart score change anything: not the order (except a click on the Degen header), not Top Picks, not Degen picks, not a fund.
• It does not predict an altseason. Alt sensitivity is measured on a year that had none, from small swings; the “give or take” next to it says how unsure it is.
• It does not fix survivorship. A theme’s year-ago basket is picked from the coins CoinGecko lists in the category today, so coins that dropped out were never candidates.
• It does not hear every conversation. “Talked about” covers the crypto chats EventTrader’s Belief Networks read, counts people not messages, and shows “too few” under 20 people.
• It has no themes for single ecosystems or BTCfi yet, and DEXTools links exist only for coins with a DEX pair on a chain DEXTools covers.
• It does not price a chain on the money parked on it, and it does not use market cap ÷ TVL. Stablecoins and DEX volume are shown as context only.
• It does not count every fee. Lending interest, staking yield, resold chain resources and most plain-transfer fees are left out of a chain’s activity, so a chain’s number here is lower than the raw fee total you see elsewhere.
• It cannot tell a real trade from a wash trade inside one app. It limits the damage instead: one app can count for at most as much as all the other apps on the chain together.
• It cannot put a Trade button on every Degen pick. The button shows only while a token has a listed pair on EventTrader with a real ask on its order book; PUMP, KNC, FXN, MNDE and EDGE (Definitive) have none right now. UP and NEST cannot be listed, because their only pools are on chains the exchange does not route orders to. SOLV’s deepest reachable pool holds about $20K, under the exchange’s $100K listing minimum. The Trade button appears the moment a pick has a pair with an ask.
• The Altseason panel does not forecast. It compares this week with the 2018 and 2021 altseason ends; it cannot know whether the next top looks like them. It updates with the screener’s refresh: weekly closes once a week, the newest day on each refresh.
• The Altseason panel does not score AI questions on crypto, Hyperliquid volume, memecoin fees, crypto app ranks, web traffic or new stablecoins: they are shown as facts. YouTube views and X, Reddit or Google Trends attention are not shown: their terms do not let us store that history.
• The Altseason panel does not trade, place orders or move funds. The exit sentence is a reading of past cycles, not advice for your position.

What this tool can't see

Altseason, and the picks log

What is not listed, and how sectors are built

Data: DeFiLlama, CoinGecko and DexScreener public APIs; prices from CoinGecko, or from DeFiLlama's coin prices when CoinGecko does not answer. The screening engine is MIT-licensed open-source code, collected and audited on EventTrader servers. Not investment advice.