At 11:09:39 UTC on 1 August 2025, a raffle contract on Base called RaffleSettle asked Chainlink VRF for one random number. It paid 10.670625544226876422 LINK for it, about $176 at the day’s price. Chainlink’s own coordinator billed 0.001058 LINK to deliver that number, a little under two cents. The raffle paid roughly 10,000 times what the oracle charged.
Nobody was robbed and nothing was hacked. The raffle paid exactly what it was told to pay, to the wei, by a Chainlink contract that worked as designed. The extra $176 is still sitting inside that contract today, along with every other surplus it has collected since July 2024, because in 870 days nobody has withdrawn a single wei of it.
The contract is VRFV2PlusWrapper_Optimism at 0xb0407dbe851f8318bd31404a49e658143c982f23, Chainlink’s “direct funding” door into VRF on Base. Games that do not want to run their own subscription pay it per draw instead. It is the fourth-biggest drawer on the chain’s all-time leaderboard, and until now nobody had compared what games pay it with what their randomness actually costs.
TL;DR
- Chainlink’s VRF wrapper on Base prices every draw from the requester’s own transaction gas price, wallet tip included, multiplied across the full callback gas limit and a 60% premium.
- From 27 September to 26 October 2026, 23 consumer contracts paid it 0.009202 ETH for 1,637 draws. The oracle billed 0.004490 ETH. Games paid 2.05 times the bill, and 41% of what they paid came from wallet tips.
- Over its lifetime the wrapper has taken 1.3621 ETH and 23.59 LINK from games (about $2,537 today). Its subscription spent 0.9752 ETH and 2.099 LINK delivering those draws (about $1,686).
- The worst single case: a raffle settlement with a 15 gwei tip was quoted 10.67 LINK (about $176) for a draw billed at 0.001058 LINK. At the block’s base fee the quote would have been 0.0038 LINK.
- The surplus has never been withdrawn. The wrapper’s ETH and LINK balances have not fallen once in 870 days.
Two ways to buy a random number
Chainlink VRF on Base offers two payment models. In the subscription model, which is what Satoshie uses, a game prepays a balance at the coordinator and every draw is debited at its actual cost after delivery. I wrote about that model in Provable Fairness Is Prepaid.
In the direct funding model, the game skips the subscription and pays the wrapper at the moment it asks. The wrapper then forwards the request to the coordinator through a subscription of its own, which Chainlink funds. When the number comes back, the coordinator debits the wrapper’s subscription at actual cost, exactly as it would for anyone else. The wrapper’s subscription is …f808c9, owned by a 4-of-9 Safe at 0xf0db7318a51a21c413cadd4abdc1e8a500fe5b1b. The wrapper itself is owned by 0xcEf2893e396d530bdECB1F1467e69F7B994eC5Ad, the same Safe that owns the coordinator.
So every direct-funded draw has two prices: what the game paid the wrapper up front, and what the coordinator billed the wrapper’s subscription afterwards. Both are on-chain. Nobody publishes the gap.
How the wrapper sets its price
The wrapper cannot know what the delivery will cost, because delivery happens in a later block at a later gas price. So it estimates, and the estimate is written into the verified source:
price = tx.gasprice × 13,400
+ (tx.gasprice × (callbackGasLimit + 128,500 + 435 × numWords)
+ L1 data fee) × 160 / 100
The 13,400 is the wrapper’s own overhead, 128,500 is its estimate of the coordinator’s overhead for a native-paid draw, 435 is per random word, and 160/100 is the coordinator’s 60% native premium. The configuration was read from getConfig() on the live contract today. For LINK payments the overhead is 150,400 and the premium 50%, and the result is converted at the LINK/ETH feed.
Two features of that formula matter. First, it charges the whole callbackGasLimit, not the gas the callback will actually use. A game that reserves 2,000,000 gas for a callback that uses 100,000 pays for 2,000,000. I measured that reservation habit across the chain in One Base Callback Left 115 Gas Spare. Second, and this one is bigger, tx.gasprice is the price the requester’s transaction paid. That is the block’s base fee plus whatever tip the requester’s wallet attached.
On Ethereum mainnet that distinction is small. On Base it is not. The base fee spent almost all of my measurement month on its floor of 0.005 gwei (median 0.005, maximum 0.0058). The typical wallet tip in the same month was 0.001 gwei, so a routine tip raised the gas price the wrapper used by 20%. And because the formula multiplies that price by the full reserved gas plus overheads plus premium, every fraction of a gwei a wallet adds is charged hundreds of thousands of times over.
One month, every draw
I pulled every log the coordinator emitted for the wrapper’s subscription between Base blocks 51,837,824 and 53,133,824 (27 September to 26 October 2026), in 1,296 slices with zero failed reads: 1,637 RandomWordsRequested matched one-to-one with 1,637 RandomWordsFulfilled. Every request in the window was paid in ETH, and every delivery succeeded. The wrapper emitted no events at all in that month, so no callback failed and nothing was withdrawn.
For each request I read the wrapper’s s_callbacks entry at the request block, which names the consumer contract, its callback gas limit and the gas price it paid. Then I asked the wrapper itself, at that same block, what it would charge, using its own estimateRequestPriceNative function. The sum of those 1,637 quotes is 9,201,770,307,835,166 wei. The wrapper’s ETH balance rose over the window by 9,201,770,307,835,166 wei. Same number to the last digit, so every consumer paid exactly the quote and not a wei more.
| One month, 1,637 draws | ETH | Share of what games paid |
|---|---|---|
| Paid to the wrapper | 0.009202 | 100% |
| Billed by the coordinator | 0.004490 | 48.8% |
| Surplus caused by wallet tips | 0.003774 | 41.0% |
| Surplus from the formula itself | 0.000937 | 10.2% |
The split comes from asking the wrapper a second question for every draw: what would you have charged if the requester had paid only the block’s base fee? That counterfactual sums to 0.005427 ETH. The difference between what games paid and that figure is the part caused purely by tips. The difference between that figure and the bill is what the formula’s reservations and overhead estimates add on their own.
In other words, games paid 2.05 times their oracle bill, and the larger part of the excess was not the formula being cautious. It was wallets tipping. Every one of the 1,637 requests carried a tip.
Who paid, and how much
Twenty-three contracts drew through the wrapper in the month. One dominates the volume:
| Consumer | Draws | Paid (ETH) | Billed (ETH) | Paid ÷ billed |
|---|---|---|---|---|
LuchamonBattleWrapper (verified) |
1,422 | 0.003927 | 0.003130 | 1.25 |
RandomnessOracle (verified) |
84 | 0.003386 | 0.001009 | 3.35 |
0xb2ce8675… (ERC1967 proxy) |
12 | 0.001285 | 0.000070 | 18.35 |
0xa6c882c9… (unverified) |
73 | 0.000191 | 0.000134 | 1.43 |
| Other 19 contracts | 46 | 0.000413 | 0.000147 | 2.81 |
The battle game made 86.9% of the draws, sends a tight 150,000 gas limit, and pays 1.25 times its bill. That is roughly what a careful direct-funded consumer looks like.
The proxy at 0xb2ce8675… is the instructive one. It drew once a day, at 00:06:41 UTC every day, which reads like a scheduled job. Each of its transactions paid a gas price of 0.105 gwei on a 0.005 gwei base fee, a tip twenty times the base fee. On 27 September it paid 0.000107 ETH for a draw billed at 0.0000056 ETH, 19 times the cost. At base fee the quote would have been 0.0000051 ETH. Its 12 draws were 0.7% of the month’s volume and paid 14% of the month’s money.
The surplus is lumpy. The 16 draws with the biggest tip surcharge (1% of the total) account for 40.8% of all the tip-driven surplus. Most players pay a little extra. A few pay a lot.
The raffle that paid $176
The month above is small money, because Base is cheap. The wrapper’s history shows what happens when a wallet is configured for a different chain’s habits.
Transaction 0x737fb881…3368414 on 1 August 2025 settled a raffle through RaffleSettle at 0xa02df55c636ce6ac54afe6cace5603b0780cfdba, which paid the wrapper in LINK. The sending wallet set a priority fee of 15 gwei on a block whose base fee was 0.0053 gwei, so the transaction’s gas price was 2,816 times the base fee. The raffle reserved 2,000,000 gas for its callback.
I asked the wrapper, at the block before, to price that request at that gas price. It answered 10,670,625,544,226,876,422 juels (10.6706 LINK), the exact amount the raffle transferred, to the last juel. At the block’s base fee it would have answered 0.0038 LINK. The coordinator’s actual bill for delivering the number, 18 blocks later, was 0.001058 LINK. With LINK at $16.50 on Chainlink’s own Base feed that day, the raffle paid $176.08 for randomness billed at $0.017. The same transaction also paid about $41 in ordinary gas to get mined, which is the cost the wallet thought it was choosing.
It was not alone. A second RaffleSettle deployment paid 1.316 and 1.300 LINK for two draws a week later, each with a 2 gwei tip, against bills of 0.000812 and 0.005291 LINK. Those three draws are most of why the wrapper’s LINK side looks the way it does over its lifetime.
870 days of it
To see the whole history without pulling 44 million blocks of logs, I read the wrapper’s two balances and its subscription’s balances and draw counter at 1,025 points, one every 43,200 blocks, from its deployment at block 16,689,438 (5 July 2024) to today. Then I located every top-up of the subscription to the exact transaction: 2.49 ETH and 230.01 LINK in total, the last on 16 June 2026, sent from three plain wallets, none of them a game.
| Period | Draws | Games paid (ETH) | Oracle billed (ETH) | Ratio | Games paid (LINK) | Oracle billed (LINK) |
|---|---|---|---|---|---|---|
| Jul 2024 to Jan 2025 | 100 | 0.0030 | 0.0007 | 4.36 | 0.134 | 0.044 |
| Jan to Jul 2025 | 5,555 | 0.0095 | 0.0077 | 1.23 | 8.581 | 1.728 |
| Jul 2025 to Jan 2026 | 12,316 | 0.1503 | 0.0262 | 5.73 | 14.751 | 0.246 |
| Jan to Jun 2026 | 116,776 | 1.0338 | 0.8020 | 1.29 | 0.002 | 0.002 |
| Jun to Nov 2026 | 29,881 | 0.1655 | 0.1385 | 1.19 | 0.122 | 0.079 |
| Lifetime | 164,628 | 1.3621 | 0.9752 | 1.40 | 23.591 | 2.099 |
In ETH, games have paid 1.40 times their oracle bill over the wrapper’s life. In LINK they have paid 11.2 times, and more than half of the LINK the wrapper holds (13.29 of 23.59) came from those three tipped raffle draws. At today’s Chainlink feed prices ($1,710.15 per ETH, $8.81 per LINK) games have paid about $2,537 and the oracle has billed about $1,686, which leaves roughly $851 of surplus.
Neither of the wrapper’s balances has gone down at any of the 1,025 points. That matters because both of its withdrawal functions, withdraw and withdrawNative, send the entire balance in one go, so any withdrawal would show up as a collapse to near zero. None appears. The wrapper holds 1.3621 ETH and 23.5905 LINK today, which is every wei it has ever been paid. Meanwhile its own subscription was topped up five times in May and June 2026 from a separate address. The surplus that could have funded those top-ups sat untouched beside it.
One methods note, because it nearly fooled me. One of the 1,025 readings, from mainnet.base.org, showed the ETH balance dipping by 0.000007 ETH. Re-reading that block on a second endpoint gave a different answer: the public endpoint had returned today’s balance for a historical block. I re-checked 148 of the readings, including every one since block 60,000,000, against Tenderly’s archive node. That was the only mismatch. With it corrected, the series never falls.
What this is, and what it is not
I want to be careful here, because “Chainlink overcharges games” is the lazy headline and it is not quite true.
- The quote is honest about its method. The source code says that fees are “paid based on current gas fees”, and the wrapper has no way to know the delivery block’s gas price in advance. Pricing from the request is a reasonable design. The problem is narrower:
tx.gaspriceincludes the tip, and on a chain where the base fee sits at 0.005 gwei, the tip is the part that varies most. - A base-fee quote would still have covered the bill. Priced at base fee, the month’s quotes sum to 0.005427 ETH against a bill of 0.004490 ETH, so the wrapper would still have come out ahead overall. It would have under-quoted 284 of the 1,637 individual draws, but over-quoted the rest by more. That is an inference from one month, not a proof for every month.
- The surplus is not all profit. The subscription that pays the oracle is funded separately, from wallets that are not games, and the 60% premium in each bill is collected by the coordinator. Who ends up with what is a question about Chainlink’s books, not something the chain shows.
- The sums are small. $851 over 870 days is not a scandal. The $176 raffle draw is the scandal-sized number, and it came from a wallet tipping 2,816 times the base fee. That wallet is not Chainlink’s fault, but the formula is what turned its mistake into a 10,000-fold price.
- A subscription does not make you immune to waste. Subscription games also reserve too much callback gas. They just get billed for what they used, after the fact, instead of being quoted for what they reserved.
Where Satoshie sits
Satoshie pays for randomness through a prepaid VRF subscription, so each draw is billed at what it actually cost to deliver, and nothing is priced from a player’s wallet settings. That is a choice with its own costs, because we have to keep the balance funded (I covered what happens when games do not in 131 Base Subscriptions Cannot Be Closed). But it means the price of fairness on our side is a number we can show you, not a number your wallet accidentally sets.
Provable fairness covers whether the number was rigged. It says nothing about whether the price of the number was sensible, and a raffle that pays $176 for a random word is fair in the first sense and absurd in the second. Both are on-chain. Only one usually gets checked.
Three questions for any game you play
- Does the game pay for randomness per draw through the wrapper, or through its own subscription? If the wrapper, your transaction’s gas price may be setting the price of the draw.
- What callback gas limit does it reserve? Through the wrapper, every reserved unit is charged whether it is used or not.
- Who pays the randomness fee, the game or you? If the fee is passed on to players, your wallet’s tip settings have just become part of the game’s pricing.
All of this can be checked on Base for free. Call estimateRequestPrice(2000000, 1, 15005327781) on 0xb0407dbe851f8318bd31404a49e658143c982f23 at block 33,628,615 and the answer is 10,670,625,544,226,876,422 juels. Change the last argument to the block’s base fee, 5,327,781, and it falls to 3,791,678,193,177,526. The raffle paid the first number.
📷 Photo by viktor rejent on Unsplash


