Polymarket has moved its primary blockchain indexing infrastructure in-house, cutting the time needed to surface some onchain events by as much as 14 blocks, or roughly 28 seconds, in a shift that quietly reshapes the data backbone of the world’s largest crypto prediction market.
Josh Stevens, Polymarket’s Vice President of Engineering, said the platform built the new system internally using rindexer, an open-source indexing tool written in Rust. The infrastructure can index data within 0 to 10 milliseconds of a block being broadcast, according to Stevens, turning what was once a third-party dependency into a core in-house capability.
“We now see onchain data events up to 14 blocks (28 seconds) faster, and we index within 0 to 10ms of block broadcasting, making it nearly instant,” Stevens said. That speed matters for a trading venue where the gap between a blockchain event and its appearance in the user interface can be the difference between a filled order and a missed one.
## What actually changed under the hood
Indexers are the plumbing of onchain applications. They organize raw blockchain activity into data that apps can retrieve efficiently. For Polymarket, the indexed information includes everything behind trades, user positions and balances, which makes the latency between an onchain event and its appearance in the application a core part of the product experience rather than a background technicality.
Until now, much of that work ran through Goldsky, a data infrastructure provider that handled important parts of Polymarket’s real-time blockchain data pipeline. According to a Goldsky case study on the partnership, its subgraphs supplied real-time indexing while its Mirror product connected onchain activity with offchain information stored by the prediction market platform.
That combination let Polymarket join trades with offchain context such as usernames and comments, while dashboard features including recent activity, top holders and leaderboards relied on the resulting data pipelines.
Stevens’ announcement changes that arrangement. Polymarket’s internally operated rindexer system now sits on the primary data path, with Goldsky retained as a backup. For a platform whose trading volumes have made it a household name in crypto, moving such a critical layer in-house signals both confidence in its engineering team and a desire to control latency at the millisecond level.
## Why rindexer
Rindexer is an open-source EVM indexing tool built in Rust. Its documentation describes the software as compatible with EVM chains and designed to let developers index blockchain events through a YAML configuration, without requiring additional code for basic deployments. Developers can extend the framework for more complex indexing pipelines, while the tool provides a GraphQL API for querying indexed data.
The project remains under active development, and its own documentation warns that changes and bugs should be expected, an unusually candid caveat for software now powering production infrastructure at one of crypto’s most heavily used applications. The choice of Rust also fits a broader industry pattern: performance-critical blockchain tooling, from execution clients to indexers, has increasingly converged on the language for its memory safety and speed.
For Polymarket, the payoff is measurable. Seeing events up to 14 blocks earlier means trades, positions and balances update faster as blockchain activity reaches the platform’s systems, tightening the loop between what happens onchain and what traders see on their screens.
## Part of a broader infrastructure buildout
The indexing change does not exist in isolation. Polymarket has spent months building out more of the infrastructure supporting its trading products, extending the platform well beyond the event contracts that made it famous.
Earlier in September, the company launched perpetual futures for eligible international users across an initial group of 10 markets covering crypto assets, equity indices and commodities. The service offers leverage of up to 20 times, while U.S. users remain excluded from the international perps product. The initial lineup includes Bitcoin, Ether, Solana and Hyperliquid alongside gold, silver, oil and U.S. equity indices.
The platform has also hardened its core prediction market business. In April, Polymarket introduced onchain surveillance through a partnership with Chainalysis, applying blockchain analytics to trades, positions and settlements to flag possible insider trading and market manipulation. By late August, the company said its surveillance systems were prepared for increased activity around the U.S. midterm elections, using machine learning and other monitoring tools to identify anomalous behavior, and that it had referred more than 100 cases to law enforcement, including cases involving suspected use of sensitive government information.
User-facing rails have evolved as well. In July, the platform enabled Bitcoin Lightning deposits through Spark, letting users fund accounts through the Lightning Network instead of waiting for conventional Bitcoin onchain confirmations.
## The takeaway
The migration to an in-house rindexer stack is the kind of change most users will never notice, and exactly the kind that compounds. Faster indexing means faster positions, faster balances and a trading experience that feels closer to a centralized exchange than to a typical onchain application.
It also reflects a maturing posture among large crypto platforms: as volume and scrutiny grow, the infrastructure layers once outsourced to specialists are being pulled back inside. Goldsky’s demotion to backup is not a judgment on its technology so much as a statement about where Polymarket believes its competitive edge now lies, in owning the milliseconds between the chain and the trader.
For a prediction market that lives and dies by the accuracy and immediacy of information, 28 seconds is not a rounding error. It is the product.
0 to 10ms indexing with rindexer is wild. anyone who has waited on a lagging subgraph during a volatile market knows exactly why they did this in-house
0 to 10ms sounds great until the first reorg forces a reindex sprint. Hope they kept Goldsky warm for exactly that scenario.
goldsky staying warm as fallback is the tell. stevens knows the first deep reorg during a live market is the real test, not the 10ms brag
first deep reorg during a live market is the exam nobody wants to sit. at least with goldsky warm they can diff both states before resolving anything
and the fallback only helps if they can dual write during a live market. switching sources mid resolution would be its own dispute generator
dual write during live markets is the whole migration risk. if rindexer and goldsky ever disagree on a state mid resolution, which one is truth
reorgwatch_rl nailed the dual write problem. whoever is declared truth mid resolution is the actual product decision here, 10ms is just the marketing layer
keeping goldsky warm is smart but the real exposure isnt indexing speed, its a market resolving off a state that later gets reorged. no fallback saves you from a disputed fill there
fills print off the displayed state though, so faster indexing does shrink the disputed fill window. reorgs are the tail risk, latency is the daily tax
28 seconds earlier on resolution data can literally change a trade. infra is alpha and Polymarket gets it
props to them for using open source instead of some closed vendor black box. rindexer in rust, good taste
cutting 28 seconds off indexing on a site where resolution timing decides fills, whoever was front-running the slow UI just lost their edge lol
The real story is they kept Goldsky as backup. Migrating a live pipeline carrying positions and balances without downtime is no joke.
Agreed with Kwame. Josh Stevens basically admitted they were renting their core product experience until now.
keeping goldsky warm as fallback during the cutover was the smart part. one bad reorg mid-migration and every open position disputes at once
the front-run edge just moves to whoever runs their own node instead. speed is an arms race, polymarket only closed one gap
exactly, the edge just relocates. same people who arbed slow subgraphs will arb fast rindexer, they only need everyone else to stay slow
exactly, whoever colocates and runs their own rindexer config still wins. speed gaps between participants are what matter, the absolute 10ms is marketing
exactly, 10ms only matters relative to other participants. the day someone colocates their own rindexer config next to polymarket the edge resets again
colocating an open source indexer does nothing unless you also see orders before they match. the displayed state was always the moat, not the pipe
colocation point is right. 28 seconds was never an edge, it was a refresh rate. the second one hft shop stands up their own rindexer config this whole in house build is table stakes again
rindexer in rust indexing within 10ms of a block, then your order still waits on their matching ui. still, cool plumbing
14 blocks earlier on resolution data is fine until you remember the real bottleneck for most users is the tab freezing during a headline dump
0 to 10ms off block broadcast is nice plumbing but the user facing win is fills printing 28 seconds sooner on resolution. that is where the money actually is
14 blocks matters most on resolution timing, thats where fills actually print. the rest of the 28 seconds is just watching the chart sooner
everyone is arguing about the 28 seconds but 0 to 10ms of block broadcast is the wild part. that basically turns polymarket into a direct read model of the chain, no indexer queue in between
direct read model of the chain is exactly right. the interesting question is whether they open that feed to api users or keep it internal for their own books
indexing in 10ms is nice but resolution is still the chokepoint when a market disputes. speed to display just means you watch the bad fill happen in real time
goldsky kept warm plus rindexer in house means they can diff two indexers live before resolving a dispute. that redundancy is worth more than the 28 seconds honestly