About TripleCoin
The manual: what TripleCoin is, how the simulation works, every trading strategy and every AI model in it, and how this differs from real institutional trading.
Everything on this page describes a simulation on paper money. Nothing here is advice, an offer or a recommendation.
1. What TripleCoin is
TripleCoin is a free, experimental coin curating service. It follows the world’s best traded coins, ranks them by fixed rules, and uses the ranking as raw material for an open experiment: do the data and the algorithms built on it hold up when they are tested against each other, hour after hour, on live prices?
It has three parts.
- Curation. Every coin is scored from 0 to 100 by the same published rules. Chains, sectors, launchpads, stablecoins, central bank digital currencies and new DeFi protocols have their own pages, with holder analysis, security checks and charts.
- A TV-format information screen. A ready-made presentation for digital signage, for anyone who wants running updates on the best traded coins on a display.
- A trading experiment. Ten simulated traders, from simple rules to AI models and a multi-strategy desk, trade paper money against live prices. The AI models are not re-trained; they are tested continuously on curated baskets, under identical conditions, to find out which approaches deserve attention. Algorithms that work are kept, those that do not are retired, and new ones are added over time.
The findings are published openly. Every closed trade gets a short report in the blog, with a chart, what the situation was, what the strategy intended and, for losses, which events lined up with the result.
TripleCoin shows dynamics, it does not predict them. The ranking is not absolute, it changes continuously, and nobody can buy or move a position in it.
2. The information screen
The information screen turns any display, TV or browser into a live market wall: a rotating programme of featured coins with charts, side-by-side comparisons of the best coins per chain and sector, the lead currencies, the top-10 ranking board, the six best simulated traders and, where available, sponsored clips. Two tickers run along the top and bottom edges: prices on top, a wire of headlines and data lines below.
- Four looks. TripleCoin (dark), NY Blue, TV (broadcast style) and PRO White/Grey (light, for bright rooms).
- Charts. Area, line or candles, over 7, 30 or 90 days. Each featured coin carries a small analysis box: rank, strongest and weakest score part, distance from its all-time high, turnover and holder concentration.
- Control. Press S for settings, F for full screen and the arrow keys to move between scenes. Every setting can also be put into the address, for example `/en/screen?theme=tv&chart=candles&views=feature,ranking,traders`, so a signage player can be configured by link alone. The scene length of each view is adjustable.
- Channels. A signed-in member can save a screen as a channel with its own address, narrowed to one sector, one chain or the coins they follow, public or private.
- Languages. The screen runs in all eleven languages of the site.
Sponsored scenes are always labelled “Promoted”, appear at most once in six scenes and never influence the ranking. Pro channel owners can switch them off.
3. Data and ranking
All data is gathered from open sources (OSINT) and none of it is altered: market data from CoinGecko and Binance, DeFi and chain data from DefiLlama, contract security checks and the largest wallets from GoPlus and Blockscout, headlines from public news feeds. Written texts, such as coin portraits and sector stories, are drafted by AI models from these numbers; the models explain, they do not invent figures.
The score has six parts: momentum 35%, liquidity 20%, stability 15%, holder structure 15%, size 10% and attention 5%. A coin is ranked only above 5 million dollars of market value and 250,000 dollars of daily volume, without signs of wash trading and without a contract that blocks selling. Stablecoins, wrapped, bridged and staked tokens are tracked but not ranked, because they follow another asset. The full rules are on the methodology page.
Prices for the simulation come from Binance, every 15 minutes, for all coins with a USDT pair, plus CoinGecko for the rest. Funding rates come from Binance perpetual futures. Real-time data will be improved over time; delays of seconds to minutes are normal and are part of what the simulation accounts for.
4. How the experiment works
Every trader starts with 100,000 dollars of paper money and trades live prices. No real money moves, nothing is offered to anyone, and every order is shown in a public log with its reason, the venue, how long the fill took and what it cost.
- The daily basket. Every day at 00:00 UTC the engine takes the ten best-ranked coins as they stand, puts them into a basket at equal weights and follows it for seven days next to Bitcoin. It is the plainest possible test of the ranking: does the top of the list beat the market?
- The traders. Each trader is an algorithm with its own rhythm: rule traders look every four hours, the AI traders every twelve hours, the aggressive trader and the Quant Desk every hour. Stops, targets and trailing stops are checked every hour, for the aggressive trader every 15 minutes.
- The benchmark. Every trader is compared with holding Bitcoin over the same time. Beating the market after costs is the only result that counts.
Reading the results. The overview shows return, return against Bitcoin, maximum drawdown (the deepest fall from a peak), the Sharpe ratio (return per unit of risk, annualised), beta to Bitcoin (1 means it moves like Bitcoin, 0 means independent), the number of trades, the win rate and the costs paid. A single week proves nothing. Sharpe and beta only become meaningful after several weeks, and a strategy needs many market phases (rising, falling, sideways) before it can be judged.
Reports. Each fully closed trade is written up in the blog within minutes: what the setup was, what happened to the price and to Bitcoin in the same window, how the trade ended and, if it lost, which headlines and market moves lined up with it. Partial sales during rebalancing do not get a report. The reports state only what the data shows; where no event can be found, they say so.
Paper trading is always kinder than reality, and the simulation says where: see “What the simulation leaves out” below.
5. The strategies: five rule traders
The first five traders use a single transparent rule each. They are deliberately simple: they show what the site’s data can do on its own. They go long only, never use leverage and always buy or sell whole positions. After selling a coin, a trader waits 24 hours before buying it again, and never holds a coin and trades against it at the same time.
Momentum Rider
Idea: what rose strongly in the last week tends to keep rising for a while.
Buys the strongest 7-day performers among the top 100 that are also up over 24 hours and above their 7-day average, up to 5 positions of 18% each.
Sells a coin that falls more than 6% in 24 hours or drops out of the 15 strongest. A trailing stop sits 12% below the highest price since entry.
Weak spot: reversals. It buys after the move has started, so sharp turns hurt.
Breakout Hunter
Idea: a coin that breaks through its recent high with real trading activity often continues.
Buys coins in the top 150 trading within 1.5% of their 7-day high, up more than 3% in 24 hours and with daily turnover of at least 5% of market value, up to 6 positions of 15%.
Exits at a stop 8% below entry or a target 20% above, and after 72 hours if the gain is under 2%, to free the capital.
Weak spot: false breakouts and fast reversals right after entry.
Dip Buyer
Idea: quality coins that fall sharply in one day often bounce.
Buys top-50 coins after a fall of 6% or more in 24 hours, provided the 30-day change is better than −25% (a dip, not a collapse), up to 6 positions of 12%.
Exits at a target 8% above entry, a stop 12% below, or after five days.
Weak spot: a dip in a falling market can simply keep falling. It trades against the trend by design.
Score Follower
Idea: the ranking itself is the signal. Hold what the ranking puts on top.
Buys a coin when it enters the top 8 and sells it only when it falls below rank 20. No stop and no target, up to 8 positions of 12%. The wide gap between entry and exit keeps trading low.
Weak spot: it holds through crashes until the rank drops, which can lag the price.
Distribution Skeptic
Idea: coins whose supply is spread across many private wallets are harder for a few holders to move.
Buys top-60 coins where the ten largest private wallets hold less than 15% of supply, best score first, up to 6 positions of 15%.
Sells when that share rises above 25%, when the coin falls below rank 80, or at a stop 15% below entry.
Weak spot: wallet labels are statistical hints, and good distribution does not protect against a market-wide fall.
6. The strategies: Aggressive Long/Short
This trader works the way a fast discretionary day trader would, with fixed rules. It looks at the whole liquid market, not only the ranking: every coin with at least 5 million dollars of daily volume, minus stablecoins and wrapped, bridged or staked tokens, up to 500 coins. It goes long and short.
Rhythm. Prices every 15 minutes, stops and trailing stops every 15 minutes, a decision every hour, at most 48 hours of holding, 6 hours of cooldown after a closed coin.
Entry. The 4-hour trend must agree with the direction (EMA 9 against EMA 21). On top of that it needs a break of the 24-hour high (long) or low (short) or a fresh 15-minute EMA 8/21 crossover, an RSI in a healthy band (55–78 for longs, 22–45 for shorts) and a push in the last hour. It trades a coin only if the target would pay the expected round-trip costs at least five times: many coins are simply too expensive to trade.
Sizing and exits. The stop sits 1.5 times the coin’s average 4-hour range away (between 2.5% and 10%), the target three times the stop, and a trailing stop follows at 1.5 times the stop distance. It risks 1.5% of equity per trade, at most 20% of equity in one position and at most 0.5% of the coin’s daily volume. At most 8 positions, 6 per side. It also exits when the trend turns, when momentum fades while in profit, or when RSI becomes extreme.
Shorts are simulated as perpetual futures: 0.05% fee and the real funding rate, paid or received every 8 hours.
Weak spot: choppy, directionless markets. Short-term signals produce many small losses there, and costs eat the rest.
7. The strategies: Quant Desk
The Quant Desk is built the way systematic funds work: not one rule, but several independent sources of return under one shared risk layer. It was added because single-rule traders showed the usual weakness: they are really bets on the direction of the whole market.
Book 1: funding carry
A perpetual future pays or charges a “funding rate” every 8 hours to keep its price near spot. When longs outnumber shorts, shorts are paid. The desk buys the coin on the spot market and sells the perpetual in exactly the same size. The price cancels out; what remains is the funding. It enters when the smoothed funding rate is at least 10% a year and two weeks of funding would cover the round-trip costs 1.5 times, holds up to 5 pairs of at most 12% each (45% in total), and unwinds when funding falls below 3% a year. Bitcoin and Ethereum are excluded; they belong to the trend book.
Weak spot: funding can flip negative, and in calm falling markets the rate is only the standard 10–11% a year.
Book 2: market-neutral momentum
Every day the desk ranks the 150 most traded coins by their 7-day return divided by their volatility. It goes long the five strongest and short the five weakest, weighted by inverse volatility, so calm coins get more capital. Positions are only changed if their weight drifts more than 25% from the target, which avoids needless trades; they are resized, not only opened and closed. Each has a stop at three days of typical volatility. Then it measures the beta of the book to Bitcoin and hedges it with a Bitcoin position, so that what is left is the quality of the selection, not the market direction.
Weak spot: momentum crashes, when yesterday’s losers suddenly rally.
Book 3: trend
Bitcoin and Ethereum are held long, flat or short depending on their own 4-hour trend and their 7-day return, with the size set by their volatility (at most 12% each).
Weak spot: sideways markets, where trend signals flip back and forth.
The risk layer
- Volatility target. The whole portfolio aims at about 23% volatility a year; positions shrink when the market gets wilder and grow when it is calm.
- Stress cut. If Bitcoin’s volatility exceeds 90% a year, everything directional is halved.
- Drawdown brake. 8% below the peak: half size. 15% below: the directional books are closed for 24 hours.
- Gross limit. At most 95% of equity invested at any time.
Execution
Orders are worked over 30 minutes in slices, mostly as resting limit orders (TWAP). That costs a fraction of the spread and lower maker fees, as real desks do. The panel on the trader’s card shows the current regime, the size of each book and the carry pairs earning funding.
8. The AI models and their strengths
Three discretionary traders are run by AI models. They receive exactly the same instructions, the same data and the same answer format, and they are bound by the same engine rules, so a difference in results is a difference in judgement. Every twelve hours each model sees the 40 best-ranked coins plus its own positions, with score parts, price changes over 1 hour to 30 days, position in the 7-day range, turnover, holder concentration and the spread. It may hold up to 8 long positions of at most 20% each, must set a stop of 3–25% on every purchase, and must explain every order in a plain sentence. Holding everything unchanged is a valid decision.
| Model | Maker | What it is known for |
|---|---|---|
| Claude Opus 5.5 | Anthropic | Careful reasoning over long rule sets, precise instruction following, clear written explanations. It also writes our coin portraits, sector stories, translations and trade reports. |
| GPT-6.1 Sol | OpenAI | Broad general reasoning, strong structured-output handling, wide knowledge across markets. |
| Gemini 4 Hydrogen | Fast, efficient inference and a very large context window, with strength in multimodal and tool-heavy work. Runs on the Flash-class tier here. |
These are general characteristics of the models, not claims about trading skill. Whether any of them trades well is exactly what the experiment measures, and the leaderboard is the evidence. More models will be added over time, and the long-term plan is a monthly score of leading AI models built from these results.
Two things the models cannot do: they cannot see the future, and they only know what is in the data block they are given. They are not allowed to use outside information.
9. How a trade is executed
Every simulated order goes through the same execution model, so costs are not wished away:
- Venue. Coins listed on a central exchange fill there in about a quarter of a second at a 0.1% taker fee. Tokens that only trade on-chain fill on a decentralised exchange after the next block of their chain, plus a pool fee and gas. Shorts fill as perpetual futures.
- Spread. Buys fill above the mid price and sells below it, by half the bid-ask spread.
- Market impact. A larger order moves the price more. The model uses the square-root law: impact grows with the square root of the order size divided by daily volume.
- Drift. The price keeps moving while the order is underway. The slower the venue, the more adverse movement is expected.
- Fees, gas and funding. Exchange fees, pool fees, gas in dollars, and funding on perpetual positions.
| Chain | Typical wait for the next block | Pool fee | Gas per swap |
|---|---|---|---|
| Arbitrum | ~0.25 s | 0.30% | ~$0.03 |
| Solana | ~0.4 s | 0.25% | ~$0.01 |
| BNB Chain | ~0.75 s | 0.25% | ~$0.05 |
| Sui | ~0.5 s | 0.30% | ~$0.01 |
| Base, Optimism, Polygon, Avalanche | ~2 s | 0.30% | $0.01–0.05 |
| Tron | ~3 s | 0.30% | ~$1 |
| TON | ~5 s | 0.30% | ~$0.05 |
| Ethereum | ~12 s | 0.30% | ~$3 |
On top of the wait for the block come about 0.3 seconds for signing and sending. For a volatile small coin, those seconds are a real cost; for a large coin on a central exchange they are almost nothing. The trader cards show the venue, the fill time and the cost breakdown of every order.
10. How this differs from institutional trading
The simulation behaves like a well-run retail account with fast execution. Large institutions trade differently, and mostly more cheaply:
- Size and discounts. Larger platforms and funds do not just hit the public order book. They deal with market makers and block-trade desks directly (over the counter), buy large amounts at negotiated prices without moving the market, and earn fee discounts by volume. Maker rebates can even turn a trading fee into income. A retail taker pays 0.1%; a large market maker may pay nothing or be paid.
- Spot and derivatives together. Institutions run spot, futures, options and lending in one account, with prime brokers, cross-margining and cheaper financing. Carry trades like the one in our Quant Desk are usually done at a scale and cost that individuals cannot reach.
- On-chain costs are a different world. A trade on a blockchain does not pay an exchange fee, it pays gas to the network and a fee to the liquidity pool, and loses value to price impact in the pool and to others who trade ahead of it (MEV). Costs vary by chain: Ethereum mainnet is the most expensive and slowest; side chains and layer-2 networks such as Base, Arbitrum and Optimism, and alternative layer-1 chains such as Solana, BNB Chain or Sui, are much cheaper and faster, but their pools are often thinner. Moving value between chains adds a bridge fee and waiting time.
- Settlement and response time vary by chain. On a central exchange, a trade settles instantly on the exchange’s own books. On a blockchain, the order waits for the next block, is “included” quickly but only final after more blocks (on Ethereum, about 13 minutes), and can fail or be reordered. Response times differ from about a quarter of a second to twelve seconds, and the exchange or the chain can be congested precisely when the market moves.
- Infrastructure. Professional desks sit physically close to the exchange servers, use several venues at once and split an order across them. They route around outages.
All of this means real results would be better for large professional players, and worse for anyone who trades small caps on slow chains. Our simulation uses the costs of an ordinary account on purpose: it shows what an algorithm earns after the costs that most people actually face.
11. What the simulation leaves out
To keep the experiment honest, here is what is not modelled:
- Over-the-counter discounts, volume tiers and maker rebates (except the Quant Desk’s limit-order approximation).
- The real depth of every order book: market impact is a statistical model, not a replay.
- Borrowing costs for shorts beyond the funding rate, margin calls and liquidations.
- Failed or reordered on-chain transactions and MEV losses.
- Exchange outages, API limits and withdrawal limits.
- Taxes, legal limits and the rules of any jurisdiction.
- The wait before a price is reported: prices are refreshed every 15 minutes, not tick by tick.
Past results, including every result on this site, say nothing about the future. Crypto assets are highly volatile and can lose their entire value.
12. Support us
TripleCoin is free and stays free to read. It is an experimental service by Flipside Coin FZCO, RAK, United Arab Emirates. If it is useful to you, there are four ways to help it grow:
- Sponsor the data. The data can be retrieved through an API key against sponsorship. See the API documentation. Real-time data will be improved over time.
- Promote a token. Verified issuers can book labelled promotion in the information screen. Promotions are always marked “Promoted” and never change the ranking. See Promotion.
- Use and share the screen. Put the information screen on a display, a shop window or a trading room, and share what you find in the blog.
- Tell us what is wrong. The best way to improve a benchmark is to find its mistakes. If you spot a wrong number, a missing coin or a strategy you would like to see tested, we want to hear about it.
Thank you for supporting an open experiment.
TripleCoin is a free curation and benchmarking service: it benchmarks AI trading algorithms against each other, and the result is a monthly score of leading AI models. More models are added over time. All data is gathered from open sources (OSINT) and none of it has been altered; the research is carried out entirely by leading AI models. No guarantee is given for the accuracy, completeness or timeliness of the data. TripleCoin is a service of Flipside Coin FZCO, RAK, United Arab Emirates. Data can be retrieved by API key against sponsorship; real-time data will be improved over time.