The Lab

How we tested

• Bitcoin, Ether and Solana, on daily, 4-hour, 1-hour and 15-minute charts (5-minute for one trigger), with no peeking ahead.
• Two years, run apart:
  • 8 October 2025 to 8 October 2026, a falling year: BTC ×0.69, ETH ×0.58, SOL ×0.53;
  • the year before, a rising one: BTC ×1.96, ETH ×1.85, SOL ×1.54.
  • The setting to re-check was chosen on the first; the year before was the test.
• The same trade shape for both placements:
  • a first entry and two adds;
  • a stop;
  • three take-profit steps, with the stop moved to the entry price after the first.
  • The only difference is where the levels sit: at Fibonacci zones, or at fixed percentages.
• Two shapes:
  • swing: trend from the daily, held for days;
  • day: trend from the 4-hour, levels from the 1-hour, held at most a day.
• The random-timing control: each of the setting's own trades, kept as a shape, opened at random moments of the year in the trend's direction, 200 times. The test, fixed in advance: the setting must be flat or better on all three coins, and beat 90% of its random draws on two of them.

The maths, step by step

1. Fibonacci levels. The ratios come from the Fibonacci numbers (1, 1, 2, 3, 5, 8, ...). Their use as trading levels is old trading lore, with no clear inventor. Take a move from a low of 100 up to a high of 120. The levels mark how far a pullback has gone, as a share of that 20:

For example, 120 − 0.618 × 20 = 107.64.

2. Confluence. Each chart has its own move and its own levels. Where levels from two or more charts sit close together, that zone counts as stronger. A score threshold sets how many must agree.

3. The volume-weighted price (VWAP), long used by traders and studied as a cost benchmark by Berkowitz, Logue and Noser (1988), is the average price, weighted by how much traded at each.

• Example: 2 coins traded at 100 and 1 at 110 gives (2 × 100 + 1 × 110) ÷ 3 = 103.33.
• One filter only takes a zone that the 1-hour move's VWAP supports.

4. Reading the results.

• Equity ×0.92 means the money ended at 92% of where it started, an 8% loss.
• Profit factor is total winnings divided by total losses. Wins of 6 against losses of 4 is 1.5. Under 1 loses.
• "Beat 8% of random draws" means 92% of the random-timing versions did better. By chance alone, a setting beats 90% of its draws only 1 time in 10.

What we found

1. Fibonacci zones placed trades no better than fixed distances.

Equity over each year:

• Neither placement won steadily.
• The day shape can't be compared fairly. Fibonacci made only 5–13 trades a year against fixed's 246–411, so the two say little about each other.
• The fixed day ladder traded 20–34 times a month and lost to its own fees and stops (×0.65 to ×0.87).

2. The one promising setting didn't hold. After the first look, one setting stood out: the day shape with the VWAP filter, a loose threshold and a hold of up to a day. It was flat or better on all three coins in the falling year.

In brackets: the share of random-timing draws it beat.

• In the year before, it lost on BTC and ETH, and did worse than random timing there.
• Where it beat chance changed from coin to coin and year to year. ETH beat chance in the first year and not the second, and SOL the other way round. That is what noise looks like.

3. The direction came from the year, not the levels. In the falling year, every long total was negative and every short total positive, for both placements. "With the daily trend" simply meant short that year.