283 altcoin pumps, replayed: what helped, what it cost, and the 3 that can still liquidate

Two-year pump timeline

Log scale. The dashed lines are the liquidation lines for two entry sizes: 1/300 and 1/500 of balance per entry.

• Under 7×
• 7× or more
• 1/300 line · 11.4×
• 1/500 line · 18.5×

Pumps that could have liquidated, by setting

For the 16 pumps of 7× or more, every 5-minute moment at which an entry was possible was replayed with the same rules. A pump counts if any entry moment reached liquidation.

• Two positions open
• One position open

Replay-implied liquidations = the sum, over the 16 pumps, of the share of entry moments that reached liquidation (table below), with two positions open. It is an average over entry timing in a replay, not a forecast.

Rows 3 and 4 add the two safeguards V19.7 ships on by default, so the last row is the V19.7 default configuration in this model. In the four pumps where the emergency close fired, the account still ended down 52% on average (76% at worst); in the model those would otherwise have been liquidations. Three extreme pumps that ran tens of times within minutes (COAI, RAVE, AKE) can still reach liquidation with two positions open.

What “underwater” looked like

The eight cycles with the deepest unrealised loss in the sample (loss as a share of balance). Seven passed the 20% line; the eighth, AKE on 17 Jul at 19%, closed at −601 USDT.

Why a position can stay open so long: by default the exit rules act on the profit side. The only loss-side action is the emergency partial close, which never fired in this sample, and the stop-loss is off. A position that never trades back below its average entry price has no other exit rule. That is why the loss side needs its own safeguards.

A stop-loss alone

The stop distance runs from tight (level 1) to wide (level 6); everything else is the baseline. Sample: 932 real cycles, three months. Level 3 was tested only together with the cap (next table).

USDT. Monthly columns count realised results; the three-month total also includes positions still open when the window ended.

The pattern suggests that wide stops lock in large losses on the biggest positions, and tight ones cut positions that would have recovered. No level alone beats the baseline. The emergency partial close never fired in this sample (it is a last resort).

A re-entry cap first, then the stop-loss

Same sample, same sizing.

USDT. Monthly columns count realised results; the three-month total also includes positions still open when the window ended.

Why the combination did better than a stop-loss alone: with the cap, the same coin is not re-entered after a stop, so level 2 fired 51 times instead of 66, and the three-month result rose from +371 to +841 USDT. Return is still lower than with the cap alone (+2,085); the stop-loss buys the drawdown cut (60.8% → 6.8%). The cap on its own lifted the old rules' +1,599 to +2,085 (+486). The 314 entries it blocked netted −171 (262 would have closed in profit, +684; 52 in a loss or still open, −855), which explains +171 of that +486; the rest follows from changes in the replay (for example margin and slots freed), which are not broken out here.

The 16 largest pumps, with the combination

Each 1-hour signal entered (re-entries capped, waiting while a position is held), staged entry on pump detection, stop-loss at ** %, 14-day window, real 5-minute prices. P&L in % of balance.

All 16 pumps: −0.1% (1/300) and −0.0% (1/500), none reached liquidation, worst single loss 0.9% (1/300) and 0.6% (1/500). This run had the stop-loss on, which is not the V19.7 default. With V19.7 default settings, 3 of these 16 pumps could still reach liquidation (chart above). On the same 16 pumps the old rules had 8 events that could liquidate and a 157% maximum drawdown. No 5-minute candle opened beyond a stop trigger, so no gap fills were needed.

LSK, 12–13 September 2026

The same pump replayed with the combination. Entries are the real account's; the shading is its real holding period.

• Real entry (number)
• Blocked in the replay
• Replay: second leg filled
• Replay: stop-loss fill
• Closed in profit (as real)

Five coins on the program's own code

This is not an approximation of the rules: the replay runs the released V18.4.1 engine, its three strategy modules, staged entry, re-entry cap, stop-loss and exit code on past 1-minute candles. The harness only feeds candles, replaces the clock and fills orders in shadow mode (stop fills assumed 1.5–2% worse than the trigger). Balance 4,000 USDT, 1/300, 10×. The coins are the four pumps that crossed the 1/500 line (RAVE, COAI, AKE, ALPACA) plus LSK as a check: its replayed entries land on the real record's times (09:41 vs real 09:41:58, 11:11 vs 11:09:51, 13th 09:49 vs 09:50:27). The method differs from the 5-minute model above, so the same pump can show a different sign (AKE −0.7% there, +3.4% here).

Read the numbers carefully. The loss side holds up: every stop fired as designed. The profit side is partly overstated: some trades (AKE +192 USDT, COAI +59 and +41, RAVE +21) opened and closed inside single one-minute wicks that a real order book may not have filled. Without them AKE and COAI end slightly negative and RAVE and ALPACA slightly positive. So the conclusion is narrow: in this replay, with the stop-loss on, the account stayed solvent through these five pumps. It does not show that pumps make money, or that V19.7 would have done the same. On the July AKE pump the real account closed a −601 USDT position (about 15% of balance); the replay's worst single loss was −35 USDT.

This engine run used V18.4.1, not V19.7. V19.7 contains these safeguards; it has not been re-run on V19.7.