starling
Measurements
The data
We replayed every Pons V2 launch seen on Robinhood Chain from block 82,486,873 to 85,808,992 (4 days, snapshot of 2026-10-11): 8,135 launches with the ETH pair and the default configuration, with every buy and sell on each curve. Reserves were rebuilt from the logs (buy: x += q − fee − tax, y −= tokens; sell the reverse). They satisfy the constant-product invariant to a relative error of 6.89e-14.
What each second costs
Premium paid above the first price by buys of 0.005 to 0.05 ETH, by seconds after the launch (fees included):
| arrival | buys | median | mean | 90th pct |
|---|---|---|---|---|
| +0s | 7,380 | +6% | +50% | +39% |
| +1s | 2,925 | +18% | +38% | +88% |
| +2s | 1,605 | +24% | +59% | +138% |
| +3s | 1,139 | +81% | +118% | +213% |
| +4s | 1,916 | +73% | +109% | +212% |
| +6s | 2,288 | +40% | +83% | +192% |
| +10s | 4,304 | +41% | +92% | +200% |
| +20s | 3,589 | +50% | +82% | +185% |
| +30s | 7,500 | +73% | +118% | +255% |
| +60s | 8,207 | +104% | +141% | +287% |
| +120s | 11,329 | +138% | +203% | +453% |
| +300s | 13,706 | +185% | +237% | +475% |
| +900s | 10,683 | +211% | +282% | +611% |
| +3600s | 13,115 | +285% | +323% | +605% |
These are buyers who got in on launches that traded, so the premium also reflects that those launches went up. It is a description of the cost of arriving late, not a forecast of what a late buyer will lose.
Who is there first
On 5,173 launches with a buy in the first 30 seconds (excluding the launcher): median 3 wallets, median 2.86% of the supply bought; the biggest wallet takes 71% of those tokens on average, the top three 90%; 48% of them are bought in the first 3 seconds.
Entering at the start, or later
A buyer of 0.01 ETH entering at the very start of the curve (flock) versus entering later, exiting on the curve, fees paid both ways. Mean return, with the share of entries losing more than half:
| entry | exit +10min | exit +1h | exit +3h | lose >50% (1 h) |
|---|---|---|---|---|
| start (no size effect) | +8.6% | +4.5% | +4.1% | 0.0% |
| +3s | -5.1% | -7.9% | -8.1% | 2.6% |
| +10s | -5.7% | -8.3% | -8.5% | 2.8% |
| +30s | -7.3% | -9.6% | -9.8% | 3.5% |
| +2min | -7.4% | -9.4% | -9.6% | 3.6% |
By flock size
A flock of the given size buys at the start; at +1 h the whole flock sells (the group pays its own impact on the way out). Fees: 2.75% on each leg (the average over the snapshot).
| flock | price paid | mean | median | lose >20% | lose >50% |
|---|---|---|---|---|---|
| 0.02 ETH | +4.0% | +4.2% | -5.4% | 0.0% | 0.0% |
| 0.05 ETH | +5.7% | +3.7% | -5.4% | 0.0% | 0.0% |
| 0.10 ETH | +8.6% | +2.9% | -5.4% | 0.0% | 0.0% |
| 0.15 ETH | +11.5% | +2.3% | -5.4% | 0.0% | 0.0% |
| 0.20 ETH | +14.4% | +1.8% | -5.4% | 0.0% | 0.0% |
| 0.25 ETH | +17.3% | +1.3% | -5.4% | 0.0% | 0.0% |
For reference, a buyer arriving 10 seconds late: mean -8.3%, median -7.2%, 2.8% losing more than half.
Rerun it
node research/measure.mjs # arrival premium, first 30 seconds
node research/measure2.mjs # entry at the start vs later, three exits
node research/measure3.mjs # flock sizes, invariant check
node research/graduation.mjs # graduation count
node research/export-site-data.mjsThe source trades are a snapshot of Pons logs too large to commit (70 MB): the format and the way to rebuild it from the chain are in research/data/README.md, and PONS_DATA points the scripts at it. Every result file is committed in research/.