This is too important to disappear under a long investigation, so here is the short version for everyone who did not want to read the full thing. For 5 days, I recorded almost every Solana token launch I could capture. The system collected 145,228 launches, roughly 29,000 new tokens per day. Then I stopped looking at the tokens and started following the wallets behind them. The first thing that stood out was concentration. Just 10 addresses created 8,895 tokens in 7 days. The largest created 1,621, roughly 1 launch every 6 minutes, around the clock. These are normal keypair wallets, not Pumpfun service addresses. One of them repeatedly creates a token and then sells its own token within seconds or minutes. The economics explain why this is possible. A launch requires roughly 0.0089 SOL of rent-funded account creation, around $1 at the observed price. Much of that rent can later be reclaimed when accounts are closed. In other words, producing another token is effectively close to free. Once the cost of manufacturing the asset approaches zero, the scarce resource is no longer capital. It is attention. Then it gets more interesting: I followed the funding layer behind one creator and found another wallet that had funded 22,258 token accounts, repeatedly with the exact same 0.00148844 SOL amount. In one transaction, it distributed the same token in the exact same quantity to 11 different wallets. It shows something important: holder counts can be manufactured mechanically and cheaply. The wallet had spent about 33 SOL just funding the accounts behind those distributions. I followed one of those recipient wallets further. Its recent activity consisted of DFlow swaps between roughly 0.0551 and 0.0567 SOL, repeating every few minutes with almost no variation. Based on the observed cadence, that single wallet would generate roughly 17 SOL of trading volume per day. Extrapolated across the 11 wallets from that distribution transaction, the observed rate would be around 190 SOL per day. That second figure is an estimate, not a measured daily total, but the transaction pattern itself is directly visible on-chain. So the structure I found looks like this: -Layer 1 creates the asset. -Layer 2 can manufacture the holder count. -Layer 3 can manufacture activity and volume. Those are 3 of the main signals retail traders routinely use to decide whether something looks alive, distributed and actively traded. And this is not just 1 wallet cluster. I checked 6 of the top 10 creator wallets. They were funded from different sources, including Binance, Bybit, OKX and unrelated wallets, with wallet ages ranging from 10 days to more than 1 year. I could not establish a common operator between them. That actually makes the finding more interesting: this appears less like 1 operation and more like a repeatable business model that multiple independent operators have discovered. One of those creator wallets was also receiving Axiom Rewards, meaning the generated trading activity can potentially create an additional revenue stream through trading-terminal incentives. The metadata has the same factory-like fingerprints. In the portion I had processed, usepaid-app appeared on 569 different tokens, elonmusk was claimed as the X handle by 73, and individual image files were reused across more than 50 tokens. Only 13.8% of the relevant metadata had been processed at that point, so those are minimum counts, not estimates. The trading results: Across hundreds of thousands of measured entry points, only 8.0% were profitable after 30 seconds, 8.7% after 60 seconds and 12.8% after 120 seconds. Average returns were negative at every one of those horizons. A realistic bonding-curve round trip cost about 2.47% through fees, spread and impact before the trader had even made a directional mistake. My ultra-fast test entered within 12 seconds of launch and exited within 30 seconds across 1,771 different tokens. It still averaged -2.41% per trade, with only 108 winners. The underlying directional edge was about +0.06%. The friction was about 40 times larger. There was even a trap inside my own data. The 5-minute survivors showed a positive average return, but only 16.5% of the original observations still had a measurable exit price at that point. The apparently bullish result was survivor bias. Most of the sample had already disappeared from the calculation. The chain already shows that tokens can be produced at industrial scale for almost nothing, holder counts can be engineered, trading activity can be automated, metadata can be recycled, and the average retail trader entering these launches is fighting brutal execution costs and terrible base rates.