Skip to main content

Every Chainlink VRF subscription on Base carries a counter called reqCount. The coordinator adds one to it each time it delivers a random word, and nobody can reset it. Read all of them and you have the chain’s all-time leaderboard for provable fairness: who has drawn the most random numbers, ever.

So I read all of them. At block 60,478,264 (18:33 UTC today, 21 November 2026) there were 1,193 live subscriptions holding 2,238,645 lifetime draws between them. The one at the top has 625,314. It belongs to a suite of eight casino contracts called Wheel, Keno, Dice, CoinToss, CoinTossBattle, RussianRoulette, Roulette and LeaderboardHub. Its last draw landed at 17:17:13 UTC on 11 May 2026. It has not drawn once in the 194 days since, and it still holds 0.7539 ETH that its owner could take back in a single transaction.

That is not an outlier. Six of the all-time top ten have not drawn in 90 days. Four have not drawn in six months. If you judge a game by its lifetime count, you are reading a trophy cabinet, and most of the people who won those trophies have left the building.

TL;DR

  • Base’s 1,193 live Chainlink VRF subscriptions hold 2,238,645 lifetime draws. Only 360 subscriptions have ever drawn a single random word.
  • The all-time number one (625,314 draws, a suite of eight casino contracts) last drew on 11 May 2026. Six of the lifetime top ten drew nothing in the last 90 days.
  • 186 subscriptions that once drew randomness have been silent for six months. They hold 994,122 lifetime draws, 44.4% of the whole ledger, and still sit on 473.90 LINK and 3.6847 ETH (about $10,250).
  • Only 51 subscriptions drew in the last 30 days. One of them made 67.75% of the 58,430 draws, and three made more than 90%.
  • 120 of the 186 silent subscriptions could be cancelled and refunded today (about $8,100). 57 cannot, because a request nobody answered is still pending.

How the ledger is kept

The VRF 2.5 coordinator on Base lives at 0xd5d517abe5cf79b7e95ec98db0f0277788aff634. Two of its view functions are enough for everything in this post. getActiveSubscriptionIds(0, 0) returns every live subscription in one call. getSubscription(id) returns each one’s LINK balance, native ETH balance, owner, consumer contracts and reqCount.

The counter is incremented in exactly one place, inside fulfillRandomWords, immediately after the coordinator hands the random words to the game:

bool success = _deliverRandomness(output.requestId, rc, randomWords);

// Increment the req count for the subscription.
++s_subscriptions[rc.subId].reqCount;

Despite the name, it is a count of fulfilments, not requests. A request that is never answered never reaches it. That makes it the honest number: randomness that actually arrived.

A lifetime counter only tells you the total, not when it was earned. So I read the whole set four times: now, and at the blocks closest to 30, 90 and 180 days ago (52,952,265 on 22 October, 50,360,265 on 23 August, 46,472,265 on 25 May). Subtract one snapshot from another and you get exactly how many draws each subscription made in that window, without pulling a single event log. Every read returned cleanly: 1,193 subscriptions today, 1,176, 1,106 and 945 at the three earlier blocks, zero failed calls, and no subscription whose counter went backwards. For the top ten I then found the exact block of the last draw by binary search on the counter itself, which is the first block at which it reached its current value.

The all-time top ten

Here is the leaderboard. Subscription IDs on VRF 2.5 are 77-digit numbers, so I show the last eight digits. Contract names are as verified on Blockscout.

Rank Sub (…last 8) Lifetime draws Last draw (UTC) Consumer(s) Still holds
1 …88259559 625,314 11 May 2026 8 casino contracts (Wheel, Keno, Dice, Roulette…) 0.7539 ETH
2 …05577593 337,668 today GridMining 0.1194 ETH
3 …19499365 187,944 today GameManager (proxy) 0.4898 ETH
4 …21938889 164,606 today VRFV2PlusWrapper_Optimism 227.91 LINK, 1.5149 ETH
5 …13595378 119,344 30 Mar 2026 15 contracts, unverified 6.67 LINK
6 …83284993 107,738 20 Jul 2026 Jackpot + a UUPS proxy 2.99 LINK
7 …45212668 103,116 1 Jun 2026 HashHorse (two deployments) 10.40 LINK
8 …74540828 100,034 13 Dec 2025 an ERC-1967 proxy 0.0098 ETH
9 …64352807 70,359 19 Nov 2026 CoinFlip + 3 others 0.0928 ETH
10 …91636747 70,308 4 Jan 2026 GameManager (proxy) 0.0563 ETH

Four of the ten drew today or this week. The other six are a cemetery with good numbers on the headstones. Between them, the four that have been silent for six months (ranks 1, 5, 8 and 10) account for 915,000 draws, 40.9% of every random word ever delivered to a live Base subscription.

None of the six moved to a new subscription on this coordinator, either. I checked every consumer contract of every top-ten subscription against all 1,193 live subscriptions. Only rank 5 has a consumer registered anywhere else, on a sibling subscription owned by the same key, and that sibling has never drawn. Whatever happened to these games, they did not simply open a fresh account next door.

How the leaders went quiet

A silent subscription can mean a game that faded or a game that was switched off. The counter can tell those apart. For each of the six, I found the first block at which it ever drew and counted its draws in each of the three 30-day stretches before its last one.

Rank First draw Last draw 60 to 90 days before 30 to 60 days before Final 30 days
1 5 Nov 2024 11 May 2026 8,819 2,678 30
5 28 May 2025 30 Mar 2026 9,028 5,226 2,304
6 19 May 2025 20 Jul 2026 490 30 10
7 2 Sep 2025 1 Jun 2026 5,436 5,490 2,670
8 9 Jan 2025 13 Dec 2025 10 18 1
10 19 Nov 2025 4 Jan 2026 n/a 25,346 44,962

Three different endings are in there. The number one faded: 8,819 draws, then 2,678, then 30 in its final month, a game whose players left before its operator did. Ranks 6 and 8 were already trickling along at a handful of draws a month long before they stopped, so their “last draw” date is close to arbitrary. Ranks 5 and 7 were still busy, at roughly 75 to 90 draws a day, when they went silent.

Rank 7 is worth a second look. It is one of the two silent leaderboard subscriptions with a request still pending, and it stopped while it was still doing several thousand draws a month. I cannot date that pending request from a storage read, so I am not claiming it is what ended the game. But a game whose randomness request is never answered cannot finish the round it is on, and this one never drew again.

Rank 10 is the strangest line in the table. Its very first draw was on 19 November 2025. Forty-six days later it had made 70,308 draws, about 1,530 a day, and then it never made another. Every draw it will ever have was made in a month and a half, and that was enough to put it in the all-time top ten of a chain with 1,193 subscriptions. That tells you how thin the rest of the table is.

Who is actually drawing

Now the same ledger by window, counting only draws made inside it:

Window Draws Subs that drew Top 1 share Top 5 share Subs to reach 90%
Lifetime 2,238,645 360 27.93% 64.10% 14
Last 180 days 481,434 174 46.86% 88.41% 6
Last 90 days 191,361 93 59.09% 88.31% 6
Last 30 days 58,430 51 67.75% 95.84% 3

The closer you get to the present, the fewer names there are. Lifetime, it takes 14 subscriptions to reach 90% of all draws. In the last 30 days it takes three. The Herfindahl index, the standard measure of market concentration where 10,000 means a monopoly, goes from 1,260 on the lifetime ledger to 4,870 on the last month. The current US merger guidelines call anything above 1,800 highly concentrated. The lifetime ledger sits below that line; the last month is nearly three times over it.

The game holding 67.75% of the last month is GridMining at rank 2, with 39,584 draws. It is the same game that has carried roughly two thirds of the draws in every window this blog has sampled since October, and its dominance is the reason the prepaid-fairness post found so few top-ups paying for so much of the chain’s randomness. Behind it come the GameManager proxy at rank 3 with 6,762, an unranked subscription ending …75933495 with 6,657, and Chainlink’s own wrapper at 1,949. After those four, the remaining 47 subscriptions made 3,478 draws between them in a month. That is about 74 each, or two and a half a day.

The newcomers are thinner still. Twenty subscriptions were created in the last 30 days. Ten of them have drawn anything, and their combined total is 226 draws.

The flow is shrinking

Divide each window by its length and you get a monthly rate for the subscriptions that are still alive today:

  • 25 May to 23 August: 290,073 draws over 90 days, about 96,700 per 30 days.
  • 23 August to 22 October: 132,931 draws over 60 days, about 66,500 per 30 days.
  • 22 October to 21 November: 58,430 draws.

That is a fall of roughly 40% in six months. One caveat is built into the method and it cuts the other way: subscriptions cancelled during a window take their draws with them, because a cancelled subscription is deleted from storage. Twenty were cancelled since May, and they held 169,077 draws at the start of it. Some of that activity happened inside the window and is missing from the May figure, so the true decline is, if anything, steeper than the one above.

I am not going to pretend that is good news for the sector Satoshie is in. Fewer games on Base are asking for verifiable randomness than six months ago, and the market that remains is dominated by one title. But it is a measurable fact about the fairness layer, and you will not find it on any dashboard, because the dashboards show the lifetime number.

The money left on the shelf

Here is the whole live population, grouped by when each subscription last drew:

Group Subs Lifetime draws LINK held ETH held
Drew in the last 30 days 51 869,613 364.23 4.0408
Last drew 30 to 90 days ago 42 134,482 120.68 0.7478
Last drew 90 to 180 days ago 81 240,428 351.42 0.8973
Silent for over 180 days 186 994,122 473.90 3.6847
Never drew at all 833 0 182.08 0.8861

The 186 retired subscriptions hold more lifetime draws than the 51 active ones, and more money too. Priced with Chainlink’s own Base feeds at the time of the snapshot (ETH/USD 1,679.32, LINK/USD 8.58), their 473.90 LINK and 3.6847 ETH come to about $10,250 sitting in subscriptions that have not used a penny of it since May at the latest.

Can the owners get it back? I simulated cancelSubscription(subId, owner) from each of the 186 owners against the live chain. 120 succeed. Those 120 would refund 386.13 LINK and 2.8631 ETH, about $8,100, to whoever calls it. The number one subscription is among them: its owner, a contract, can close it today and collect 0.7539 ETH, roughly $1,266.

The other 66 are stuck. 57 revert with PendingRequestExists(), the trap described in the post on subscriptions that cannot be closed: a request went out, was never answered, and now blocks the exit forever. Ranks 7 and 8 on the leaderboard are both in that group, holding 10.40 LINK and 0.0098 ETH respectively. The last 9 revert with FailedToSendNative(), because the owner is a contract that cannot receive ETH, even a zero-value refund.

Nobody is obliged to close a subscription. But a game that walked away without doing so left its trophy on the shelf and its deposit in the till, which tells you something about how it ended.

Why the lifetime number misleads players

“Over 600,000 provably fair draws” is a sentence any of these games could still put on a landing page, and it would be true. It is also the least useful number a player can be given.

A lifetime count tells you that randomness was delivered in the past. It does not tell you whether the subscription is funded now, whether the consumer contract on the page is still the one registered, whether there is a request stuck in the queue, or whether the operator is still around to answer when something goes wrong. Every one of those is a question about the present, and every one has a free on-chain answer. The trophy only has one about the past.

It is also, quietly, a measure of size rather than fairness. A game can rack up a six-figure count with a VRF integration that delivers perfect randomness to a contract that then does something unfair with it. This blog has spent two months showing the gaps that live between the random word and the payout. A big counter covers none of them.

One subscription is not one game

The leaderboard has a structural quirk worth knowing. A subscription is a billing account, not a game, and two of the top four are many games each.

Rank 1 is eight contracts sharing one balance, so its 625,314 draws were spread across roulette, dice, keno and the rest. Rank 4 is stranger: its only consumer is VRFV2PlusWrapper_Optimism, Chainlink’s own direct-funding wrapper. Games that pay per request through the wrapper never open a subscription of their own; the wrapper draws from this one on their behalf. So the 164,606 draws at rank 4 belong to an unknown number of separate games, and the 1,949 the wrapper made this month might be one busy title or twenty quiet ones. The subscription-level view cannot tell, and I have not attempted to unpick it here.

That cuts both ways for the concentration figures. The wrapper makes the market look slightly more concentrated than it is. The single-owner suites at rank 1 and rank 9 make it look slightly less concentrated than it is by owner. Neither moves the headline: the top of the lifetime table is mostly silent, and the current market is mostly one game.

What this means if you play

Treat a lifetime count the way you would treat a restaurant’s “established 1998” sign. Nice to know. Not evidence of anything happening in the kitchen tonight.

The live questions are cheap to answer. Ask the coordinator for the subscription behind the game you are about to play. Check that your game’s contract is on its consumer list, that it holds enough to pay for your draw, that pendingRequestExists is false or at least recent, and that the counter has moved this week. If a game’s randomness account has been silent for months while its front end still takes deposits, something is wrong, and you found it before you paid rather than after.

That is the standard Satoshie should be held to as well. Our raffles and coinflip draw from Chainlink VRF on Base, our subscription is on the same public coordinator as every one in this post, and every number above can be read for ours with the same free calls. If our counter stops moving while our site is open, that is a fair thing to hold against us. Provable fairness is something a game does every day, not something it did once.

For a sense of how short most of these lives are, the cancelled-subscriptions post found a median of 21 hours among the ones that were properly closed. The ones in this post are the other kind: they ran for months, drew hundreds of thousands of numbers, and then simply stopped, without closing the account.

Honest limits

  • Silent is not dead. A subscription with no draws may belong to a game that moved to another chain, another oracle, another coordinator version, or a fresh contract I cannot link to it. I checked only that no top-ten consumer reappears on another live Base VRF 2.5 subscription. What the operators are doing now, I do not know.
  • The ledger excludes the cancelled. A cancelled subscription is deleted, counter and all. Lifetime totals here cover live subscriptions only, and window totals miss draws by anything cancelled during the window (3 subscriptions in the last 30 days, 10 in 90, 20 in 180). The decline is therefore understated, not overstated.
  • A fulfilment is counted whether or not the game accepted it. The counter increments even when the consumer’s callback fails, so a draw here means randomness delivered, not a game resolved.
  • Snapshots are four points, not a series. The monthly rates are averages across each window. A game that ran hot for a week and stopped looks the same as one that ticked along slowly.
  • Cancel simulations are a snapshot. The 120 subscriptions that can be closed today were tested from each owner’s address at one block. A pending request arriving tomorrow would change the answer, and owners that are contracts may have no function that actually calls cancel, which a simulation from their address does not test.
  • Dollar figures use one price. Both feeds were read at the snapshot. LINK and ETH both move, so treat $10,250 and $8,100 as orders of magnitude for today.

Three questions for any game you play

  1. When did the subscription behind this game last receive a random word?
  2. Is the contract I am about to pay on that subscription’s consumer list, and is anything stuck in its queue?
  3. If the operator walked away tomorrow, could anyone close the account, or would my request sit there forever?

Everything above is re-checkable on Base for free. Call getSubscription(93271074453882857048982544254580571499844752099945152824703671195862788259559) on 0xd5d517abe5cf79b7e95ec98db0f0277788aff634 and the third field is 625,314. Call it again at block 45,865,242 and it is one lower; at 45,865,243 it is not. That is the last time the all-time number one asked for anything.

📷 Photo by breizhography on Unsplash

Valentina Ní Críonna

Author Valentina Ní Críonna

More posts by Valentina Ní Críonna