Methodology & FAQ

How Mihenk measures crypto KOLs · Last updated: 17 October 2026

1. What the Mihenk score measures

The Mihenk score runs from 1 to 10. It answers one question: if this account posts about something, how many real, relevant people will actually see it? It is built from four groups of signals:

• Real reach. The median number of views per post across the account's recent posts. We use the median, not the average, so one viral post doesn't inflate the picture.
• Audience quality. How much of the audience looks like real, active, crypto-relevant accounts. We screen for bots, engagement farms and dormant followers.
• Integrity. Signals of inflated engagement: sudden follower jumps, engagement coming from farms, repetitive or automated replies, and follower-to-view ratios that don't add up.
• Smart followers. How many already-measured, credible crypto accounts follow the KOL. It's a strong signal that the people who matter pay attention.

Scores place accounts into tiers: T1 Elite, T2 Strong, T3 Rising and T4 Watch. We don't publish the exact weights; that keeps the score hard to game.

2. Dormant accounts

An account's score is only as good as its recent activity. Our rule:

• An account silent for 60+ days (confirmed by a recent measurement and our post archive) is marked dormant. Its score becomes provisional: 80% of its last score.
• When it starts posting again, the provisional score stays until we see 10 new posts or 14 days of activity. Then the score is measured again from recent activity, up or down.
• Dormant and returning accounts are labelled on their KOL card, so you always know why a score looks the way it does.

3. How calls are detected

A call is a KOL's own post that takes a view on a coin's price. An AI referee reads every archived post against fixed written rules; a cashtag like $ZEC helps but isn't required.

• Bullish posts are long calls, bearish posts ("it's going lower", "stay away") are short calls. Sponsored posts about a coin count as long calls too.
• News, definitions, lists of four or more coins, and posts that only describe past price moves aren't calls. Stocks, gold and indices are never calls.
• We archive posts from the most influential KOLs every hour, and from the wider set of measured accounts on a rolling schedule.
• Several posts about the same coin by the same KOL on the same day count as one call, timed at the first post.
• Tickers that turn out to be stocks, or coins with no price data anywhere after 48 hours, are dropped.
• Deleted posts stay in the record and are marked as deleted. A call doesn't disappear because the post did.

4. How calls are measured

For every call we record the coin's price at the time of the post and again 5 minutes, 1 hour, 4 hours and 24 hours later. Measurements run every hour.

• Price data. Centralized exchange prices first (such as Binance, Bybit and OKX). For tokens that trade only on-chain, we use DEX pool data (GeckoTerminal / CoinGecko), choosing the pool that existed at the time of the post. When a post includes a contract address, we measure that exact token.
• Against BTC. A coin going up while the whole market goes up isn't much of a call. So we compare: vs BTC = the coin's 24h change minus BTC's change over the same 24 hours, in percentage points.

A call beats BTC at +2 points or more, lags BTC at −2 points or less, and is neutral in between.

5. Leverage reaction (futures)

Price shows what happened. Futures show how traders positioned. For every measured call on a coin that has a USDT perpetual contract (Binance first, then Bybit), we also record:

• Open interest at the time of the post and 24 hours later. We count it in coins, not dollars, so a rising price doesn't inflate it.
• Against BTC. Same idea as price: the coin's 24h open interest change minus BTC's open interest change over the same 24 hours, in percentage points.
• Funding rate and long/short account ratio at both moments, shown as context.

We mark +5 points or more as leverage up and −5 points or less as leverage down. Rising open interest means new leveraged positions were opened after the post. It tells you traders acted on it, not which way they bet or whether they were right.

A KOL's leverage reaction summary appears once 3 or more of their calls have futures data. Verified call records show it as a column. About half of measured calls are on coins with perpetual futures; the rest show "—".

6. Telegram and YouTube

Mihenk isn't tied to one platform. Public Telegram channels and YouTube channels are measured on the same principles as X: real reach instead of follower counts, audience authenticity, and every call checked against BTC.

• Telegram: real reach. The median views of a channel's posts that are 1 to 30 days old (newer posts are still collecting views). Channels that haven't posted in 30 days get no reach figure.
• Telegram: view rate. Median views divided by subscribers. Many subscribers but few views per post usually means inactive or inflated subscribers.
• Telegram: calls. A coin's cashtag or a token contract address in the channel's own post. A contract address identifies the exact token, so look-alike tickers aren't confused. Posts that list four or more coins count as lists, not calls.
• YouTube: real reach. The median views of long videos (over 60 seconds) published 2 to 30 days ago. Shorts are tracked separately.
• YouTube: engagement. Likes plus comments, divided by views.
• YouTube: calls. A coin named in a video title, with its cashtag or as a ticker that's actively discussed across the market. Descriptions are skipped because sponsor sections add noise, and titles listing several coins count as lists.

Telegram and YouTube calls are measured exactly like posts on X: 5 minutes, 1 hour, 4 hours and 24 hours after publishing, against BTC. Scores use the same scale and the same impact tiers as X.

7. Verified call records and maps

• Verified call record. Shows a KOL's calls from the last 90 days whose 24 hours have passed. Every call is listed with a link to the original post; none are picked or removed. Record links are signed, so a record can only be issued by Mihenk and can't be faked for another account.
• "Who's talking" maps. Show the KOLs who posted about a coin in the last 14 days, sized by real reach. Project teams, exchanges and non-crypto accounts are left out. The green percentage under a KOL is the peak: the highest price the coin reached after that KOL's first post, compared with the price at the post.
• Weekly heatmap. Shows every coin KOLs called that week, sized by number of calls and coloured by the average result against BTC. It shows coins, not people.

8. Trader scorecard

For KOLs who post full trade setups, we track each setup the way a follower would have traded it. No black-box score: the record says how many trades succeeded, out of how many.

• Who is listed. KOLs who posted at least 3 complete setups in the last 90 days: a stop, or a target together with a clear long or short. Only measured setups count.
• Levels. An AI model reads the entry, targets (TP1, TP2…), stop and any trigger level from the post. Exchange position cards in images are read too; their entry and exit prices are checked against the exchange price at the same hour, and cards that don't match aren't counted.
• Tracking. Hourly exchange candles from the moment of the post, for 7 days. A conditional setup ("long if it breaks 100") starts only after the trigger level is reached.
• Win rate. Trades that reached TP1 before the stop, plus trades the trader closed in profit before TP1, divided by all trades with an outcome. If TP1 and the stop fall in the same hourly candle, it counts as a stop.
• No stop given. A 10% move against the entry counts as the stop (a full loss at 10x leverage), so a trader can't skip the stop and wait for the price to come back.
• The trader's own close. If the trader says they closed, or shares a close card, before TP1 or the stop, we take the exchange price at that moment: in profit against the entry is a success, otherwise a failure. A claim that contradicts the price isn't counted.
• 7 days and 30 days. The main record uses 7 days. Trades that reach neither TP1 nor the stop in 7 days are followed to day 30 and shown separately.
• Return. Unleveraged, as if you had followed the trade exactly: the position is split equally across the targets, a part closes at each target reached, the rest at the stop or at the end of day 7.
• If you had followed. What an account would be after putting 1% or 10% of it into each trade, unleveraged, in call order, with gains and losses compounding.
• Ranking. By the lower bound of the win rate's confidence interval (Wilson), so 3 out of 3 doesn't outrank 40 out of 50.
• Deleted posts. A setup stays in the record when its post is deleted. The trader page shows how many checked setups were deleted, and how many of the stopped-out ones.
• Shared positions. Exchange position cards and "opened / closed" posts, checked against exchange prices. We also show how many shared openings later got a close shared, because traders mostly post their winners.
• Telegram signal channels. Measured from the first 15-minute candle after the signal. A signal posted after the price had already passed its first target is counted as late and isn't measured.
• Alerts. When you follow a trader, Telegram alerts arrive when they post a setup, when the price reaches the entry, a target or the stop (the same rules as the record), and when they close a position.

9. How often data updates

• Post archive and call measurements: every hour.
• Trader setups, levels and position cards: every hour. Targets and stops of followed traders: every 5 minutes.
• Telegram channels: every 2 hours. YouTube channels: once a day.
• Futures data (open interest, funding, long/short ratio): every hour, once a call's 24 hours have passed.
• KOL scores and the dormant-account rule: every night.
• Deep measurements of individual accounts: on a rolling schedule, and on request.

10. What we can't measure (yet)

• Correlation, not causation. A price move after a post doesn't prove the post caused it. Markets move for many reasons.
• Reading posts. The AI referee and the level reader follow written rules, but they can misread a post, especially charts and long multi-coin threads. Corrections are welcome (see the FAQ).
• Small samples. A record built on a handful of calls says less than one built on dozens. We always show the number of calls.
• Timing. We measure from the post's own timestamp, not from when we found it.
• Audience overlap. Real reach totals add up each KOL's median views; people who follow several KOLs are counted more than once.
• Thin markets. For very small on-chain tokens, a few trades can move the price a lot.
• Futures coverage. Leverage data exists only for coins with perpetual futures on Binance or Bybit, and exchange history goes back about 30 days, so older calls have none. Liquidations aren't included yet.
• Platform coverage. We cover public Telegram channels (not private or paid VIP channels or groups) and public YouTube channels. TikTok and Discord aren't measured yet.
• Hourly candles. Trader records use hourly candles, so a wick inside the hour counts, and the order of two levels touched in the same hour is unknown (we count the stop).

11. FAQ

Try free for 5 days →