Anatomy of a crypto influencee: what RealVision did to my friend

The zip landed in my downloads on a Thursday evening: 5,061 messages, 360 photos, screenshots and voice notes, December 2023 to this week. Somewhere in it is the whole 2024 bull run, the August 2024 crash, the Trump coin weekend, a paid RealVision subscription, a presale scam, a pump.fun memecoin, and a coin called SUI going from one dollar to five and back to seventy cents. I call the friend BJ4 here.

The point is a question I could not answer from memory: if I had bought every coin he told me to buy, on the day he said it, where would I be today? And the question behind that one: what makes a careful, numerate, successful person say "prends 10k de sui, tu le regretteras pas" and mean it?

1. Turning a chat into data

A WhatsApp export is a text file with one line per message. I parsed it, then read the whole thing and coded by hand every statement about a coin: date, time, coin, what kind of statement (a buy call, a sell call, a price target, a prediction, a comment, his own executed trade, a scam warning, a sarcasm), sentiment on a five-point scale from very bearish to very bullish, how sure he sounded, the horizon he had in mind, the price target if any, the money he suggested, and the source he named. 326 rows in total, 256 of them asset-level statements by BJ4 on 41 distinct coins. Every row keeps the quote, so anyone can dispute the coding.

Prices are daily closes from Binance, with Bybit filling the coins Binance does not list. Two "coins" have no market at all: a presale he paid $500 into in March 2024 and a pump.fun token that went to nothing in May 2025. Both count as a total loss, because that is what the chat says happened.

The timeline below is the raw material. Every dot is one message, teal when bullish, red when bearish, bigger when he sounded sure. The triangles are the moments he said, in one form or another, put money here.

Three things jump out before any arithmetic. The coins arrive in waves that match the narrative of the month: an ETF story brings BTC, a "Solana killer" story brings SUI, an AI-agent story brings Virtuals and ai16z, a Binance listing brings Aster. Almost nothing is ever sold; the down-triangles cluster in one month, August 2024, and they are all Bitcoin. And the bearish red is rare, late, and lands on coins after they have already fallen.

2. The BJ4 Index

The rule is mechanical. Every time he says buy, whether as advice to me or as a report of his own purchase, the index buys $1,000 of that coin at that day's close and never sells, because he never said sell. Fifty-four such calls between March 2024 and November 2025. For a benchmark, the same $1,000 goes into Bitcoin on the same day, and, separately, into Solana on the same day. The chart divides what the basket is worth by what has been put in so far, so a rising line means the calls are paying and 1.0x means break-even.

For one month, following him was the best trade in the chat. Between mid-November 2024 and mid-January 2025 the basket was worth twice what had gone in, and I remember the tone of the messages: "tu vas te faire 1k par minute ce weekend", "Cetus top le 14 décembre". Nobody said sell on the 14th of December, or on any other day. The index then rode SUI, Cetus, Deep, Virtuals and ai16z down, and today sits at 0.59x, nose to nose with a pure Solana drip and forty points behind a pure Bitcoin drip.

Call by call, the picture is not one bad decision but a distribution. The Bitcoin calls of summer 2024 all made money. The alt calls of autumn 2024 all lost most of it. The one call that beats Bitcoin over its whole life is XRP on 19 June 2024, a coin he mentioned once, never bought, and dismissed in November as one of the "projets pourris" that rise while SUI does not.

"To today" is a harsh judge, because today is a bear phase: Bitcoin at 77,000, Solana at 101, SUI at 0.73. So the fair question is not only where the calls are now but whether they ever beat the benchmark at any reasonable holding period. The answer is no at every horizon I tested. Against Bitcoin, the share of calls ahead after 30 days is 31 percent, after 90 days 21 percent, after a year 17 percent. A coin flip would give 50. Against Solana he is a coin flip, which is another way of saying his alts and Solana move together and he adds no information to "own SOL".

One nuance the hit rate hides: the mean excess return over Bitcoin at 90 days is positive, plus 11 percent, with a bootstrap interval of minus 11 to plus 37. Four or five entries into SUI and Cetus in September to November 2024 doubled or tripled within three months and drag the average up while four calls in five lose. That is the shape of his edge, if he has one: rare, large, short-lived, and never cashed.

3. The promises

Twenty-nine times he gave a number. Solana to 600, then 660, then 1,000. SUI to 8, to 18, to 25, to 60. NEAR to 57. Bitcoin to 150,000, 250,000, 450,000, one million, eight million by 2034. Cetus times ten. Ethereum times two by the end of summer 2024. The chart puts each promise next to the best price the market ever paid afterwards.

The targets are not his. They are RealVision's, Raoul Pal's, Peter Brandt's, a YouTuber's, "tout le net". He relays them with the confidence of the source and none of the source's caveats, and the further the target sits from the current price, the more certain the message sounds. That is the first clue to the mechanism.

4. Anatomy

Behavioural finance has spent forty years naming the ways a smart person loses money in a liquid market. The chat is a field sample of five of them, and each one leaves a measurable trace.

4.1 He gets bullish after the move

For every asset-level message I computed how the coin had moved in the 30 days before he sent it, and how it moved in the 90 days after. Sentiment tracks the past: the correlation between his tone and the trailing 30-day return is 0.31, with a p-value below one in a hundred thousand. Twenty-eight percent of his buy calls came within a week of a plus 10 percent move; nine percent came after a minus 10 percent move. His sentiment also predicts the next 30 days a little, because momentum persists for a few weeks, and predicts nothing at 90 days.

4.2 He herds on the paid voice

In October 2024 he subscribed to RealVision and started calling its analysts "les SME", subject matter experts, the term we use at work for the people whose sign-off ends an argument. From then on the big allocation calls trace back to a flash update: 30 percent SOL, 30 percent SUI, 30 percent BTC; buy Deep; switch part of SOL into SUI, "x5 voire x10"; Solana to 1,000; Bitcoin 450,000 in Q1 2026. Sixteen percent of his asset statements name an outside voice explicitly, and those are the ones with the money behind them.

The calls he attributed to an outside voice returned, on average, 17 percent less than Bitcoin over the next 90 days. The calls where he named no source returned 26 percent more. The calls from his own chart reading sat in between. The paid signal was the worst signal in the chat. The SUI arc is the whole story in one coin.

He was right about SUI, early and for a while. His first message on it, 23 August 2024 at 0.84, promised a double in two weeks; it did 4x in four months. Then the paid voice said 25, and 25 became the anchor. At 4.2 in January he was bearish because SUI had slipped to fifteenth by market cap. At 3 in February he doubled his position. At 0.98 in May 2026 he was still waiting for the alt season. The coin trades at 0.73.

4.3 He is sure

Overconfidence is the best-documented bias in retail trading, and it is not a personality flaw so much as a feature of how confidence is produced: the more recently a view has been confirmed, the more certain it feels. In the chat the pattern is exact. After a win the next call carries an amount ("prends 10k", "tu mets 5k de Cetus"), a deadline ("dans les deux semaines", "Cetus top le 14 décembre") and a promise ("ça va payer tes safaris"). After a loss the tone drops to "à mon avis, mais je ne suis pas conseiller". Four of twenty-nine targets were met, and the four that were met were the modest ones.

4.4 He panics, then he holds

The single most expensive week in the chat is 5 to 24 August 2024. Bitcoin fell 25 percent in six days on a Japanese rate shock. He had 3.1 BTC bought at an average of 39,000 euros and a paper gain he had been watching every hour. On the morning of the fifth, at the low, he sold everything in three orders. That evening the market recovered. He bought back the next day, 0.36 BTC lighter. Then he did it again, and again, thirteen trips in twenty days, "pour tester ma stratégie", "trop le fomo quand je suis en euros". On 24 August he owned 2.59 BTC. The half-coin he lost to those twenty days is worth $39,000 today and was worth $64,000 at the cycle top.

The mirror of the panic is the hold. Cetus, bought on his advice in November 2024 at 0.37, went to 0.02. His counsel in February 2025, with the position at minus 80 percent: "garde-le, tu vas te refaire, 80 percent c'est normal sur des jetons comme le Cetus." This is the disposition effect, selling winners early and riding losers down, in its purest textbook form, and it is asymmetric in the chat: the coin he could not bear to hold was the one that recovered, the coins he could not bear to sell were the ones that did not.

4.5 The one real skill

I wanted to find something he does well, and the data obliged. For every call I ranked the coin's forward return against twenty peer coins on the same date. His picks land, on average, at the 60th percentile over the next 180 days, and the test rejects "no skill" at p = 0.0001. He does find the coins that are about to run. The same test on timing, the actual entry date against a thousand random dates for the same coin, gives the 48th percentile. And his conviction is informative: the calls he made with the most certainty beat Bitcoin by 37 percent over 90 days, the tentative ones lost by 23.

So there is a signal in the chat. It is a selection signal, it decays within a quarter, and it is destroyed by three habits: entering after the move, never writing a sell, and scaling up after the win.

5. The profile

I scored eight traits from the same numbers. Each bar has its basis in the workbook behind this page, and each is a description, not a diagnosis.

In one line: a narrative-driven momentum follower with a sound long-term Bitcoin core, a high risk appetite on the satellite, a paid influencer for conviction, no exit rule, and one documented panic that cost him half a coin. The textbook names are Barber and Odean's overconfident active trader for August 2024, Shiller's herding for the SUI months, Shefrin and Statman's disposition effect for Cetus. He is not unusual. He is the median retail participant, with a better memory and a worse chat log.

6. What RealVision did to my friend

Nothing that he did not ask for. The subscription did not make him wrong about SUI; he was right about SUI before he paid. What it did was give his confidence a number (25) and a source he could not argue with, so that the exit he would have improvised on his own never came. An expert's target is a ceiling on your thinking, not a floor under your money.

The four rules I took from the exercise, for anyone with a BJ4 in their chat:

Use his selection, not his timing. When he names a coin for the first time, he is early more often than not. Buy on the next red week, not on the message.
Write the sell before the buy. He never once did. A basket that doubled and gave it all back is a basket with no sell rule.
Weight by his certainty, discount by his source. "Je suis sûr" from his own reading is the best signal in the chat; "Raoul dit" is the worst.
Do not talk to him on a red day. His words, 6 August 2024: "on a paniqué et on s'est entraîné dans la panique. On aurait mieux fait de pas se parler."