When Bitcoin rises while the bot loses money, improving the machinery is not enough. The trading rules need scrutiny. We tested whether evidence of aggressive spot buying could help a simple long-or-cash strategy enter earlier without taking as many losing trades.

The comparison that matters is whether the extra information improves the same price rule, and whether the resulting strategy beats holding Bitcoin. Beating a struggling bot alone is a low bar.

What we actually measured

We used Binance’s public Bitcoin spot records: the value of trades initiated by buyers, divided by total traded value. A reading above 50% means buyers initiated more of the volume in that window. It does not mean that more money entered Bitcoin than left it: every trade still has a buyer and a seller.

The base rule enters after a breakout above the previous 20 days’ high and exits below the previous 10 days’ low. An early-entry shortcut can buy above the previous five days’ high when price is also above its 20-day average. We tested allowing that shortcut only when the buying-pressure filter agreed. The normal breakout and exit stayed the same.

We first measured one day of buying pressure. We then changed how old that information was allowed to be, and separately tested totals covering three, seven and fourteen days. All signals used completed candles.

These are historical tests

Each experiment’s rules were written down before its results were calculated. But the historical periods had already been explored in earlier research. That makes these follow-up tests, not untouched holdouts.

We retained annual and longer windows, trading costs of 0.06% and 0.30% per dollar traded, and four- and twelve-hour execution delays. Feature delays were additional. Each account started with $10,000; Bitcoin bought once and held. No interest was credited to cash. The original short strategy’s funding and borrowing costs were omitted.

The recent comparison

June 11, 2026 at 04:00 UTC through September 8 at 00:00 UTC. Returns after modeled 0.06% trading costs and a four-hour execution delay.
StrategyReturnLargest drawdown
Original reconstructed bot−8.12%12.35%
Price-only early entry+11.28%11.64%
One-day pressure, 24-hour-old data+23.92%7.59%
Seven-day pressure, four-hour-old data+23.92%7.59%
Fourteen-day pressure, four-hour-old data+20.11%7.59%
Buy and hold Bitcoin+26.19%13.36%

Drawdowns are measured at the observed four-hour closes; losses inside a candle could have been worse. These figures use a pinned research dataset and explicit execution assumptions. They are not the changing public scoreboard or a live trading record.

One avoided loss carried the promising result

The one-day filter using 24-hour-old data skipped the losing June 15–24 position. Its later entries and exits matched the price-only strategy. That single avoided position explains its entire recent improvement.

Using 20-hour-old data produced the same +23.92% result. Using 28-hour-old data restored the losing trade and cut the return to +11.28%. The immediate timing neighborhood failed nine of 36 registered comparisons against the price-only rule.

We also tried sixteen fixed coin-flip filters that randomly permitted the shortcut. Three did better than the selected flow filter in the recent window. Those controls used a 50% permission rate, rather than matching the flow filter’s exact rate, so this is a cautionary comparison, not a formal probability calculation.

Longer histories complicate the story

From 2023 through the frozen April 2026 endpoint, the 24-hour filter’s advantage over price-only entry came from several trades. It survived subtracting its largest positive contribution. But the strategy returned +251.83%, while holding Bitcoin returned +365.18%.

A deliberately week-old one-day flow control did better than Bitcoin from 2019 through that endpoint: +2,920.32% versus +1,975.43% at the primary settings, with a lower observed drawdown. Its advantage survived the four cost-and-execution combinations. Yet it still trailed Bitcoin from 2023 onward and in the recent window. Week-old information can reflect persistent conditions, so this control is a research lead; selecting the best-looking delay after seeing the results would not validate it.

Did sustained pressure solve the timing problem?

Not consistently. The three-, seven- and fourteen-day filters improved on price-only entry in 55 of 72 comparisons across the three main periods and execution assumptions. None of the six variants beat holding Bitcoin in the recent or 2023-onward tests under any of the tested cost-and-execution settings.

The seven-day filter using four-hour-old data returned +23.92% recently; using 24-hour-old data returned +11.28%. The fourteen-day filter was more stable at +20.11% for both feature delays, but still lagged Bitcoin’s +26.19%.

Both seven-day versions slightly beat Bitcoin from 2019 onward at the primary settings. Each beat Bitcoin in only two of the four cost-and-execution scenarios. The longer aggregation window did not establish a dependable higher-return replacement.

What happens next

We have started a separate paper journal for fourteen fixed long-or-cash strategies, including Bitcoin hold and every multi-day candidate. Its first intentions were recorded on September 8 at 16:25 UTC. No simulated purchases had occurred at this article’s evidence cutoff: a later observed price, at least four hours after recording, is required. Future results will come from those recorded intentions, with delays and missed observations left visible. This initial entry is not evidence of profitability.

Sources and complete results

This study uses one exchange’s BTC/USDT aggressor volume alongside BTC/USD account prices. It does not validate a multi-exchange dashboard or measure institutional money flows. Nineteen incomplete or off-boundary archive records were excluded under a documented data-only amendment; missing intervals were never filled. Historical archive revisions and original publication times remain unverified.