Bitcoin's Cluster Mempool: How a 15-Year Architecture Overhaul Is Rewriting the Fee Market
For fifteen years, Bitcoin's mempool — the waiting room where unconfirmed transactions sit before being mined into blocks — has operated on architecture designed when a single Bitcoin was worth pennies. That era is ending. On November 25, 2025, Bitcoin Core merged Pull Request #33629, a sweeping redesign called Cluster Mempool that replaces the legacy transaction-sorting engine with a unified, cluster-based framework. Targeted for Bitcoin Core 31.0 in the second half of 2026, this upgrade quietly ranks among the most consequential protocol-level changes Bitcoin has seen since SegWit.
No new opcodes. No token standard. No flashy narrative. Just a fundamental rethinking of how every Bitcoin node decides which transactions matter most — and why that decision has been subtly broken for years.
The Problem Nobody Talks About: Two Rankings That Disagree
Bitcoin's mempool has always maintained two separate orderings for unconfirmed transactions. Ancestor feerate ranks transactions by looking backward through parent dependencies. Descendant feerate ranks them by looking forward through child transactions. Mining software uses ancestor feerate to build blocks. Eviction logic — the process of removing low-priority transactions when the mempool is full — uses descendant feerate.
This dual-ranking system creates a quiet but persistent misalignment. What miners want to include in a block and what nodes consider most expendable are ranked on entirely different scales. The consequences ripple across the Bitcoin ecosystem:
- Wallets misjudge fees. Fee estimation relies on predicting which transactions miners will include. When the ranking that builds blocks differs from the ranking that manages mempool capacity, wallets frequently overpay or underpay — leading to either wasted sats or stuck transactions.
- Replace-By-Fee (RBF) behaves inconsistently. A transaction replacement that looks favorable under one ranking system might be rejected under the other, creating unpredictable UX for anyone bumping fees.
- Lightning Network penalty transactions face uncertainty. Layer 2 protocols depend on timely transaction propagation and confirmation. When mempool behavior diverges from miner behavior, enforcement mechanisms like Lightning penalty transactions can't guarantee propagation to miners — a security-critical problem.
- Block construction leaves money on the table. Miners building blocks from ancestor-feerate-sorted transactions sometimes miss more profitable arrangements, leaving an estimated 2–5% of potential revenue uncaptured.
These problems are not theoretical. In 2025, the average Bitcoin transaction fee hovered around $0.62, with occasional spikes above $2 during congestion events. During those spikes, the gap between what wallets estimated and what users actually needed to pay widened — sometimes by orders of magnitude.
Cluster Mempool: One Ranking to Rule Them All
Cluster Mempool, designed by Bitcoin Core contributors Suhas Daftuar and Pieter Wuille, eliminates the dual-ranking problem by introducing a single, unified framework.
How Clusters Work
A cluster is a maximal set of transactions connected through spending relationships. If transaction A creates an output that transaction B spends, they belong to the same cluster. If transaction C spends an output from B, all three form one cluster. The key insight is that these connected transactions should be evaluated together, not individually.
Each cluster is then linearized — sorted internally by feerate into monotonically decreasing "chunks." A chunk is one or more transactions within the cluster that should be included (or excluded) as a unit. This linearization process uses sophisticated algorithms that Wuille benchmarked extensively in 2025, including a linear programming-based spanning-forest approach that proved both more efficient and more practical than simpler candidate-set searches.
Unified Operations
With clusters and chunks in place, every mempool operation now uses the same ordering:
| Operation | Old Approach | Cluster Mempool Approach |
|---|---|---|
| Block construction | Ancestor feerate sort | Select highest feerate chunks across all clusters |
| Eviction | Descendant feerate sort | Remove lowest feerate chunks |
| RBF evaluation | Check absolute fee and feerate | Compare chunk feerates; require strict feerate diagram improvement |
| Fee estimation | Heuristic-based, often inaccurate | Directly derived from chunk ordering |
The alignment is elegant: what miners want to mine is exactly what nodes consider most valuable. What nodes evict is exactly what miners would leave out. No more misalignment.
Cluster Limits
The new system enforces a global limit of 64 transactions and 101 kvB per cluster. This constraint replaces the old ancestor and descendant limits, which separately capped how many parents or children a transaction could have. The cluster limit is more flexible — a transaction can have unlimited ancestors as long as the entire cluster stays within bounds — while remaining computationally tractable for constant re-sorting.
What This Means for the Bitcoin Ecosystem
For Wallet Developers: Reliable Fee Estimation
The single biggest UX win from Cluster Mempool is predictable fee estimation. When the mempool's internal ranking matches what miners actually do, wallets can observe the mempool state and directly calculate the fee needed for a target confirmation window. No more relying on historical confirmation data that lags behind real-time conditions.
This is particularly significant during congestion events. In late 2024, Bitcoin Core developers proposed updates to fee estimation methods, and Cluster Mempool provides the architectural foundation to make those methods work as intended.
For Miners: Better Block Revenue
Block template construction becomes a straightforward merge sort across cluster chunks. The highest-feerate chunks from every cluster get selected first, without the approximation errors inherent in the ancestor-feerate algorithm. While the revenue improvement is estimated at 2–5% per block, this compounds across thousands of blocks per month — meaningful for mining operations running on thin margins.
For Layer 2 Protocols: Security Guarantees
Lightning Network, Ark, and other Layer 2 protocols rely on the ability to broadcast time-sensitive transactions that will propagate through the network and reach miners. The old system's inconsistent treatment of transaction replacements created edge cases where penalty transactions might be accepted by some nodes but not propagated to miners.
Cluster Mempool's unified framework provides stronger guarantees. RBF rules now require that a replacement must strictly improve the mempool's overall feerate diagram — a mathematically precise condition rather than an ad-hoc fee threshold. This means protocol developers can reason about transaction propagation with greater confidence.
For Ordinals and Inscription Economics
Bitcoin's inscription ecosystem — Ordinals, BRC-20 tokens, and Runes — has introduced transaction patterns that stress the old mempool architecture. Complex CPFP (Child-Pays-For-Parent) chains, large transaction clusters, and high-feerate competition during inscription mints all benefit from the unified ranking system. Inscription transactions that were previously hard to fee-bump can now be managed more predictably through the chunk-based evaluation.
The Engineering Journey: From 1989 to 2026
Cluster Mempool's development history reveals the kind of deep, patient engineering that defines Bitcoin Core's approach.
The project traces back to a 2023 proposal (GitHub issue #27677) that identified the fundamental misalignment between mining and eviction rankings. But the algorithms powering the solution draw on research spanning decades. Early in 2025, Stefan Richter discovered that a 1989 paper on the maximum-ratio closure problem could be applied to cluster linearization — an academic breakthrough repurposed for cryptocurrency infrastructure nearly four decades later.
Wuille spent months in 2025 evaluating three linearization approaches: a simple candidate-set search, a linear programming-based spanning-forest algorithm, and a min-cut-based approach. His benchmarks showed the two advanced methods were "far more efficient" than the simple search, with the spanning-forest approach winning on practicality.
The actual implementation was merged as PR #33629 on November 25, 2025, following years of review, testing, and incremental pull requests. The Bitcoin Core 31.0 release, expected in H2 2026, will be the first version to ship Cluster Mempool to the broader network.
Comparison with Other Fee Market Designs
Bitcoin's Cluster Mempool upgrade arrives as other blockchains have already iterated on fee market design:
- Ethereum's EIP-1559 (August 2021) introduced a base fee that adjusts dynamically with demand, plus a priority tip for miners. It fundamentally changed how Ethereum users experience fees — more predictable, with automatic refunds of overpayment. However, EIP-1559 doesn't address internal mempool organization; it changes the fee mechanism itself.
- Solana's localized fee markets (introduced 2024) segment fee competition by the accounts a transaction touches. A high-demand NFT mint doesn't raise fees for a simple token transfer if they touch different state.
- Bitcoin's Cluster Mempool is different from both: it doesn't change the fee mechanism (Bitcoin still uses a first-price auction) or segment the fee market. Instead, it fixes the internal plumbing so the existing mechanism works correctly. Think of it as repairing a broken thermometer rather than redesigning the heating system.
This distinction matters. Bitcoin's fee market is arguably the simplest among major blockchains — users bid fees, miners pick the highest-paying transactions. Cluster Mempool doesn't add complexity to this model. It removes the internal inconsistencies that prevented the simple model from working as intended.
What Comes Next
Cluster Mempool enables several follow-on improvements that were previously impractical:
- Package relay: Broadcasting groups of related transactions together, rather than one at a time. With cluster-aware mempools, nodes can evaluate packages as units, improving CPFP reliability.
- Improved RBF policies: More precise replacement rules that can evaluate whether a replacement actually improves the mempool, not just whether it pays a higher absolute fee.
- Mempool-based fee estimation: Direct computation of optimal fees from the current mempool state, replacing historical confirmation-rate models.
These aren't speculative features. Bitcoin Core developers have been designing package relay and improved RBF around the cluster mempool architecture for years, waiting for the foundational change to merge before building on top of it.
The Quiet Revolution
In an ecosystem that gravitates toward loud narratives — ETFs, memecoins, Layer 2 launches — Cluster Mempool represents the kind of infrastructure upgrade that rarely makes headlines but shapes everything built on top of it. More reliable fees mean fewer stuck transactions. Better block construction means miners earn what they should. Stronger propagation guarantees mean Layer 2 protocols can trust the base layer.
Bitcoin Core 31.0 won't ship until H2 2026, and network-wide adoption will take months beyond that. But when it does, every Bitcoin transaction — from a Lightning channel close to an Ordinals inscription to a simple peer-to-peer payment — will pass through infrastructure that's been fundamentally rebuilt for the first time in Bitcoin's history.
Sometimes the most important upgrades are the ones you never notice.
BlockEden.xyz provides high-performance Bitcoin and multi-chain node infrastructure for developers building on the protocols that matter most. As Bitcoin's internal architecture evolves, reliable RPC access to well-maintained nodes becomes critical for wallets, exchanges, and Layer 2 protocols navigating the transition. Explore our API marketplace to build on infrastructure designed for Bitcoin's next era.