How coins are filtered
How coins are filtered
Every hour the same fixed rules check the same coins. Below is where each coin stopped today, and what happens to the ones that pass. No opinions: the rules decide.
Before the checks
Which coins we look at
We only check coins with a busy perpetual futures market on the exchanges we read, because the rules need whale positions, funding and liquidation data for every coin. Our tracked list: 32 coins today, each inside the top 150 by 30-day volume. A coin whose volume looks artificial waits for a review.
- still in
- stopped at this step
- stopped earlier
Step 1 · Gate 1 of 3
Tradable: is there enough trading?
Thin markets give prices we can't trust. A coin needs 24h volume at least $50M and open interest at least $20M (open interest is the value of open futures positions).
Step 1 · Gate 2 of 3
Market safety: is the whole market calm enough?
This check is about the market, not one coin. It closes when Bitcoin trades within 2% of the dealer gamma level, or from 4h before to 1h after CPI, jobs or Fed (major US releases). It also waits after a sudden Bitcoin move. When it closes, coins already on the list stay and are watched; no new coin is added.
Step 1 · Gate 3 of 3
Direction: do the weekly numbers lean one way?
Every week big funds report their futures positions to the US regulator (the COT report). We put that next to the coin's weekly trend: price against its 20-week average. When both point the same way, the coin gets a direction, up or down. When they don't, there is no view and the coin stops. Only Bitcoin, Ether, Solana and XRP have their own report; other coins use Bitcoin's.
This count also includes any coin missing today's price or daily range.
Step 2
Five kinds of evidence, one vote each
Several indicators often measure the same thing. Counting them one by one would let one piece of evidence look like three, so they are grouped and each group casts one vote: agrees, neutral or against.
- Money flow: is real money buying or selling? (CVD and two volume indicators)
- Positioning: are whales on the read's side while the crowd is not, with funding not overheated?
- Liquidity: is a large pool of forced orders waiting in the read's direction?
- Timing: is the move still fresh, with no sign it is worn out?
- Value: where is price against what holders paid? Not in the beta yet: the data is paid, so this group has no data for any coin.
A group without fresh data never counts as agreeing.
Step 3
The rule: at least 3 of 5 must agree
- 5 of 5 agree, or 4 of 5 with none against: Bullish or Bearish
- 3 of 5 agree, at most 1 against: Mostly bullish or Mostly bearish
- Fewer than 3 agree, or 2 or more against: does not pass.
The direction from Step 1 says whether the label reads bullish or bearish. At most 1 of the 5 groups may be without data.
Step 4
Vetoes: checks that can still block a coin
- Whales heavily on the other side: 70%+ of whales positioned against the read. Without whale data the coin can't pass either.
- A big pool of forced orders against the read: the price where the view would be wrong is too close (under 0.5 daily ATR) or too far (over 3), or a much larger liquidation cluster sits just beyond it.
This count also includes coins with no liquidation cluster to set that price, and coins that left the list less than 72 hours ago.
Invalidated if…
Every coin on the list comes with one price where its view is wrong, fixed the moment it passes. It sits just beyond the nearest large liquidation cluster against the read (0.25 daily ATR beyond the nearest large cluster), so a normal sweep of those orders does not end a good view.
Invalidated / too late
A coin leaves the list for one of these reasons and moves to “Invalidated / too late” with the reason. Nothing is deleted.
- Invalidated: price reached the price where the view is wrong.
- Too late: it already moved more than 1.5x daily ATR with the read.
- Filter stopped passing: it failed the checks 2 hours in a row.
- Time limit: 7 days on the list ended.
- Suspended: a hack or delisting the rules can't see, with a public reason.
How it is scored, in public
The record scores the read, not a trade. A hit = after the pass, the price moves with the read by at least its distance to invalidation (the 1:1 level) before it touches the invalidation level. Levels are judged on touch, not on candle closes: a level counts as reached as soon as the price touches it, wicks included. Prices: the Binance USD-M futures mark price, minute by minute, for 7 days from the pass. If both levels are touched in the same minute, it counts as invalidated. If 1-minute prices cover less than 90% of those 7 days, the coin is judged on hourly prices and its row is marked “Hourly prices”. Then both levels in the same hour count as invalidated. Neither level in 7 days = “No result”. Counted, not hidden.
Returns at 24 hours, 3 days and 7 days are measured in the direction of the read, for every coin that reached that point, and compared with a random top-50 coin over the same window. The comparison coin comes from the top 50 by market cap at the pass time, but only from the coins we track, so the comparison covers fewer than 50 coins. The record shows how many. The hit rate is compared too: each of those coins gets the read's direction and the same distance to invalidation in its own daily ATR, from the pass time, scored by the same rule. Each day's new rows are locked with a fingerprint anyone can check: the pass price, the levels, the method version and how the coin left. Scores are added later and written once, but they are not part of the fingerprint yet.
No backtest yet. The v0.1 thresholds are starting values, so this live record is the only evidence for them so far.
The rules in full
Everything the story above leaves out, in the order the checks are described in our plan. Fixed rules decide, and the same rules write every sentence you read here.
This page describes Method v0.1, in force (built-in default).
1. Which coins we look at
Our tracked list: 32 coins today, each inside the top 150 by 30-day volume.
Each coin needs a busy perpetual futures market on the exchanges we read (Binance, Bybit and Hyperliquid), so whale positions, funding and liquidation data exist, and about 45 days of history. Stablecoins, wrapped coins, leveraged tokens and tokenised gold are left out. A coin whose volume or open interest looks artificial waits for a review.
The check runs every hour, when each hourly candle closes (UTC).
2. Step 1: three gates
Three gates must all pass before a coin is looked at in detail.
- Tradable. Enough trading that prices are real: 24h volume at least $50M and open interest at least $20M.
- Market safety. Bitcoin at least 2% away from the dealer gamma level, and no major US release window: from 4h before to 1h after CPI, jobs or Fed. The build also waits after a sudden Bitcoin move (the volatility breaker). When this fails, coins already listed stay and are watched, and new ones wait. Our system runs this check before Tradable.
- Direction. The weekly read is up or down: the COT report and the weekly trend (price against its 20-week average) must point the same way. Only BTC, ETH, SOL, XRP have their own COT line; other coins use Bitcoin's. The plan also names MarketLens; it is not defined yet and is left out. A coin missing today's price or daily range stops here too.
3. Step 2: five evidence groups
Each group casts one vote: agrees, neutral or against, by the majority of the indicators inside it that have data (a tie is neutral). A group without fresh data is "not available" and never counts as agreeing.
Money flow
What's inside: Chaikin money flow (CMF), money flow index (MFI), CVD divergence.
Agrees means: Real buying (or selling) volume points the same way as the read.
The exchange whale ratio in the plan comes from a paid source and is not in the beta.
Positioning
What's inside: Whales vs crowd on Hyperliquid, open interest and funding.
Agrees means: Large traders lean with the read while the crowd does not, and funding is not overheated.
Wintermute's positions in the plan come from a paid source and are not in the beta.
Liquidity
What's inside: Liquidation clusters above and below the price (model estimate from open interest).
Agrees means: A large cluster sits in the direction of the read, and nothing large sits just on the other side.
Timing
What's inside: TD Sequential count on 3-day candles.
Agrees means: No completed exhaustion count of 9 against the read.
The Markov model in the plan is not defined yet and is not used.
Value
What's inside: Holder cost basis (short- and long-term holders).
Agrees means: Price sits on the right side of what holders paid.
Not in the beta: holder cost-basis data comes from a paid source, so Value is not available for any coin yet.
4. Step 3: the decision rule and labels
| Votes | Label |
|---|---|
| 5 of 5 agree, or 4 of 5 with none against | Bullish or Bearish |
| 3 of 5 agree, at most 1 against | Mostly bullish or Mostly bearish |
| Fewer than 3 agree, or 2 or more against | Doesn't pass. The list never shows "Neutral". |
4 groups agreeing with 1 against reads as Mostly, not the full label.
Groups without data, and the two method versions
A group without fresh data counts as neutral, never as agreeing. How many groups may be without data is set by the method version. Every coin card says how many of the 5 groups had data and which version the coin passed under.
| Version | Groups without data | What it means |
|---|---|---|
| v0.1 (in force) | At most 1 of 5 | The rulebook rule. At least 4 groups must have data; the labels above apply as written. |
| v0.1 beta | Up to 2 of 5 | For the beta, while Value has no data source and Liquidity covers only some coins. With 2 groups without data, all 3 groups with data must agree and the label is at most Mostly, never the full Bullish or Bearish. With 0 or 1 without data, nothing changes. |
A switch applies to new passes only. Coins already on the list keep the version they were listed under until they leave, and the record shows the version of every row, so the two can always be told apart. Weights start equal and change only between named versions.
5. Step 4: vetoes
Checks that block a coin even when the votes pass.
- Whales against the read: 70%+ of whales positioned against.
- Whale data missing or stale, so the whale check can't run: the coin does not pass.
- The price where the view is wrong is less than 0.5 or more than 3 daily ATR away.
- A liquidation cluster at least 2× the one that sets that price sits within 0.5 daily ATR beyond it. Price is likely to sweep it and end the view anyway.
Two more rules stop a coin at this point: when no large liquidation cluster against the read exists to set the price where the view is wrong, and for 72 hours after a coin left the list.
6. Invalidated if…
Every coin gets one price where the view is wrong, fixed when it passes. It sits just beyond the nearest large liquidation cluster against the read: 0.25 daily ATR beyond the nearest large cluster. Liquidation clusters do double duty: they vote in the Liquidity group and they set this price. The same clusters are shown on each card as “Key liquidity levels”.
7. Invalidated / too late
- Invalidated: price reached the price where the view is wrong.
- Too late: it moved more than 1.5x daily ATR with the read, so it's no longer early. Still scored.
- Filter stopped passing: it failed 2 hourly checks in a row (fewer groups agree, a gate fails or a veto applies).
- Time limit: 7 days on the list ended.
- Suspended: removed for a reason the rules can't see (a hack, a delisting), with a public reason. Still scored.
Nothing is deleted. Every coin that leaves lands in "Invalidated / too late" and in the record.
8. How it is scored
A hit = after the pass, the price moves with the read by at least its distance to invalidation (the 1:1 level) before it touches the invalidation level. Levels are judged on touch, not on candle closes: a level counts as reached as soon as the price touches it, wicks included. Prices: the Binance USD-M futures mark price, minute by minute, for 7 days from the pass. If both levels are touched in the same minute, it counts as invalidated. If 1-minute prices cover less than 90% of those 7 days, the coin is judged on hourly prices and its row is marked “Hourly prices”. Then both levels in the same hour count as invalidated. Neither level in 7 days = “No result”. Counted, not hidden.
Returns at 24 hours, 3 days and 7 days are measured in the direction of the read, for every coin that reached that point, and compared with a random top-50 coin over the same window. The comparison coin comes from the top 50 by market cap at the pass time, but only from the coins we track, so the comparison covers fewer than 50 coins. The record shows how many. The hit rate is compared too: each of those coins gets the read's direction and the same distance to invalidation in its own daily ATR, from the pass time, scored by the same rule. Each day's new rows are locked with a fingerprint anyone can check: the pass price, the levels, the method version and how the coin left. Scores are added later and written once, but they are not part of the fingerprint yet.
No backtest yet. The v0.1 thresholds are starting values, so this live record is the only evidence for them so far.
9. What we never do
- Tell you what to trade, where, or with how much.
- Add, force or re-score a coin by hand.
- Hide or edit a record row after it is sealed.
- Show data we don't have: missing data says "not available", never a guess.
- Let a person or an AI pick the coins. Fixed rules decide, and they write every sentence.
10. Glossary
- Perpetual futures
- Futures contracts with no end date, traded around the clock. Most crypto trading happens here, so this is where we read positions.
- Open interest
- The total value of open futures positions. Rising open interest means new money is entering positions.
- Funding
- A small payment between long and short futures holders every 8 hours. Positive funding means longs pay shorts: the crowd leans long.
- Whales
- Traders with very large positions. We read their positions on Hyperliquid, where positions are public.
- Liquidation cluster
- A price where many leveraged positions would be force-closed at once. Price is often drawn to these levels. Our clusters are a model estimate from open interest.
- Daily ATR
- Average true range: how far a coin usually moves in a day. We measure distances in daily ATR so quiet and wild coins compare fairly.
- CVD
- Cumulative volume delta: buying volume minus selling volume over time. CVD falling while price is flat suggests real selling.
- COT report
- The weekly Commitments of Traders report from the US CFTC. It shows how large funds are positioned in bitcoin and other crypto futures.
- Dealer gamma level
- Options dealers hedge as price moves. Near this level their hedging can speed moves up either way, so the market check waits.