Do Our Trading Signals Beat Random Timing?

First, what is being tested?

Charlie publishes a simple rule. Each day it looks at six market readings, each squeezed onto a fixed scale from -1 to +1 using rules set in advance: whether Bitcoin sits above its own average price of the last 200 days (Bitcoin trend), how technology shares moved over the last month (Technology shares), how the US dollar moved (US dollar, flipped so a weaker dollar counts as positive), how nervous the stock market looks (Equity stress, flipped), how fast the supply of stablecoins is growing (Stablecoin supply; stablecoins are digital dollars, coins built to stay worth one dollar), and how much traders are paying to hold Bitcoin futures (Futures cost, flipped; futures are contracts that bet on the price without owning the coin). The six are averaged. If the average is above zero the rule holds Bitcoin the next day; otherwise it sits in cash. A small fee is charged every time it switches. The chart follows $100 through the test period, the later 40% of the history, which was kept aside while the rule was built.

How many real decisions was that?

The test period has 183 trading days, and the rule changed its mind only 10 times in that stretch. Separately, each of the six readings is judged on what Bitcoin did over the following 30 days. Those 30-day windows overlap like a sliding door: two windows one day apart share 29 of their 30 days, so they are almost the same bet counted twice. The whole history holds 427 such windows, which sounds like a lot. But each 30-day window shares 29 days with the next one, so divide by 30: that is about 14 windows that do not overlap. The test period holds 153 windows, which is about 5 that do not overlap. Those smaller numbers are the honest ones.

If 1,000 fake versions picked their days at random, how many would beat holding by as much?

This is the test that matters most, and here is why it is fair. The rule only ever chooses one thing: when to be in the market. So freeze everything else and scramble only that one choice. If randomly chosen days earn the same money, then the choosing was never what earned it. Keep the exact same prices. Keep the exact same number of days in and out of the market. Only shuffle when those in-and-out stretches happen, in blocks of 30 days so the pretend versions still look like real rules. Do that 1,000 times and see where the real result lands.

Would a different mix of the same six ingredients also have worked?

If a recipe only works with exactly these six ingredients and no others, that is a warning sign: it may have been tuned to the past. So every possible mix was tried, from one reading alone up to all six together. That is 63 recipes.

Does each ingredient have a real edge on its own?

For each reading, look at what Bitcoin did over the next 30 days when the reading was positive, compared with when it was negative, using only the training part of the history. The dot is that difference. The bar is the range you get when the history is rebuilt from its own 30-day chunks 600 times (resampling). If the bar crosses zero, the edge could easily be nothing. A reading also fails if it spent fewer than 40 days on one side, because then one side is barely measured.

Are the Alpha Tracker’s quick calls better than random guessing?

The Alpha Tracker is Charlie’s public list of short-term trading calls, each saying a small coin will rise or fall over the next hours or days. It has 493 finished calls, published between 2026-09-05 and 2026-10-09, and 57% of them ended in the direction they predicted (a bet on a fall counts as a win if the price fell). That sounds strong. But the calls were published on only 35 separate days, and calls made on the same day tend to win or lose together. So two 50-50 guesses are shown: one where the coin is flipped for every call, and one where it is flipped once per day.

Do the famous Bitcoin signals beat luck over a longer history?

Charlie’s own rule only has a few years of history. Four well-known Bitcoin readings go back further, so each was turned into a simple rule fixed in advance, run from 2018 or from the first day the rule could be computed (each needs a run-up of past readings before its first day), with the same 0.10% switching fee, and put through the same shuffle test: keep the prices, shuffle when the rule was in and out, 1,000 times. The rules are as simple as they sound: for the three readings that measure how cheap or how crowded Bitcoin is, hold Bitcoin only when the reading is below its own middle value of the past year; for exchange flows, hold only when more coins left exchanges than arrived over the last 30 days. The bar is the share of pretend versions (the shuffled ones) that did as well as the real rule. Short bars are good. Each rule’s start year is shown in its label.

What did $100 do under “Coins leaving exchanges”?

The rule’s path against simply holding Bitcoin from 2018-01-01 to 2026-10-05. The rule was in the market 48% of the days and switched 169 times.

Did each signal work every year, or only in one?

For each calendar year, the rule’s return (how much money grew or shrank) minus the return from just holding Bitcoin, in percentage points. Where a cell says the rule made no move all year, it never switched position from January to December, so there was no decision to judge that year. Where it says there is not enough of the year, we did not have enough trading days to score it. Positive years are years the rule was ahead.

What should you do with this?

Two kinds of reader use this page. One runs money for other people and has rules to follow. The other is deciding about their own savings. The same evidence leads to different actions.

The conclusion

Words used on this page

Every technical word above is explained again here, in plain English.