Measures
What we replayed, what it implies at pump.fun's scale, and what it does not prove.
What was measured, and where
The measures come from Pons, a launchpad on Robinhood Chain, not from pump.fun: 348,897 trades across 9,521 launches over four days, with each coin's reserves rebuilt exactly, trade by trade, on its bonding curve. A buyer puts in a small amount at the 10th (or 40th) trade of a coin and holds the tokens for three hours. We run that on each coin with and without a defence, the other trades unchanged, and compare the buyer's return. The defence only buys.
The result there
| Setup on Pons (2 % creator tax) | 10th trade | 40th trade |
|---|---|---|
| One coin, all its fees as defence | +1.3 pt | +1.0 pt |
| One coin, half its fees as defence | +0.6 pt | +0.4 pt |
| Fees of about 1,500 coins pooled, half as defence | +2.2 pt | +1.7 pt |
| Fees of about 1,500 coins pooled, all as defence | +4.6 pt | +3.4 pt |
Points of return gained compared with the same coin paying the same tax with no defence.
What it implies on pump.fun
The result scales with how much money the defence has. On Pons the creator tax was 2 %. On pump.fun the creator fee of a new coin is 0.30 %, about seven times less. A single Conatus coin therefore has about a seventh of the defence money of the single coin above, and the gain should be a fraction of a point. This is an estimate drawn from the replay, not a measurement on pump.fun.
The dangerous-dip zone
The replay above used a single trigger, a fall of 40 %. The live rule reacts from 30 % and harder as the dip deepens (15 %, 25 % and 35 % of the reserve at 30, 40 and 50 %). On the same replay the tiered version gained +4.4 and +3.3 points against +4.6 and +3.4 for the single trigger: the same result within noise, with a gentler first reaction. Other detectors we tried (speed of the fall, share of sells, waiting for a rebound, and combinations) did not beat the single trigger, and several did clearly worse.
What we tried and dropped
We first tested a tournament in which several spending strategies compete and the best one is followed. The best strategy was the same on both halves of the data and the tournament added at most 1.8 points even with perfect hindsight, so we kept a single fixed rule.
What this does not prove
- It is four days on another chain, not a market cycle on pump.fun.
- Other traders are assumed not to change their behaviour because a defence exists. They might, either way.
- Only a small share of coins earn enough in fees for any defence money to exist at all.
The scripts are in the project under research/. They only read data already downloaded and send nothing.