Fee Gap Between Blocks Drives More Hashrate Into Rete mining Races, Research Finds

Block 964,120 carried 0.0077 BTC in transaction fees. Block 964,121, mined immediately after, carried 0.0536 BTC, nearly seven times more. On a network where miners earn a fixed 3.125 BTC subsidy per block, that fee swing looks like noise. A July 2026 NBER working paper argues it is a security signal. Fabian Schär of the University of Basel, his PhD student Dario Thürkauf, and NYU's David Yermack analyzed eight years of mempool data captured directly from the Bitcoin network between 2017 and 2025. Their finding: when the fee gap between one block and the next grows wide, competing blocks at the same height, known as block races, become more frequent. The next block also becomes less likely to appear in its first seconds. The paper is observational. It identifies a network-level association, not miner intent behind any individual race. But it reframes a question the security-budget debate has mostly ignored: Bitcoin's mining incentives do not only depend on how much the network pays in fees over a month or a year. They respond to how fees arrive from block to block.

The Mechanics Behind the Race

Bitcoin's consensus rule is simple. A miner who finds a valid block extends the chain and collects the subsidy plus included transaction fees. Rational miners should always build on the newest block. But when the just-found block contains an unusually large fee payload and the mempool left over is thin, the calculus shifts. A miner can attempt to recreate the valuable prior block, claim its fees, and extend an alternative chain. That attempt starts behind the accepted tip. Its economic appeal rises when fees in the prior block greatly exceed the expected fees in a new tip-extending block. Hash-rate share, propagation speed, and whether other miners extend the accepted tip all affect the odds. Thürkauf defines a block race as two or more competing blocks at the same height. His co-author explanation links these races directly to the fee-gap incentive: "If a block that was just found captured an unusually rich set of fees, and the transactions left over for the next block are thin, a miner can earn more by trying to re-mine that rich block than by building on top of it. Today the effect is muted. Glassnode data from Aug. 26 places transaction fees at approximately 0.70% of miner revenue. BTC.network's weekly report covering Aug. 14, 21 calculated a 0.67% fee share, with miners collecting $232.6M in total revenue across 1,009 blocks, of which only 21.25 BTC, or roughly $1.68M, came from fees. Those aggregate numbers hide the block-level variation. A Glassnode daily snapshot can show a fee-revenue share below 1%, while an individual block carries 10 or 20 times the fees of the one before it. The paper's contribution is that this variance itself creates incentive distortion, even when the network-wide fee total looks benign. | Difficulty | 127.48 T | BTC.network, Aug. 14, 21 | The next subsidy reduction arrives with the 2028 halving, when the block reward drops to 1.5625 BTC. As the subsidy shrinks, fee variance occupies a larger share of total mining income. The incentive to contest high-fee blocks grows in proportion.

What Lock Fields Do — and Why Coverage Gaps Matter

Wallets can reduce the prize available to a fee-sniping miner by setting transaction lock fields. A wallet can configure a transaction to become mineable only in the block after the current tip, which excludes it from a replacement of the current tip. Widespread lock-field use changes the economics of re-mining because a miner rebuilding the earlier height loses access to some of the newest pending transactions. BIP 326 describes anti-fee-sniping behavior for Taproot transactions through `nLockTime` or `nSequence`. It remains a draft informational proposal, and wallet adoption is voluntary. An open Bitcoin Core issue from April 2026 documents one concrete inconsistency: the `send` RPC and GUI wallet set `nLockTime` near the current block height, while `createrawtransaction` and `walletcreatefundedpsbt` default to zero. The proposed fix remains open. The gap means Bitcoin Core's anti-fee-sniping default is applied unevenly across transaction-creation paths. Estimating the network-wide effect would require data on how much volume each path represents. The issue does not provide that breakdown.

The Observation Window

CoinGecko data shows schär, Thürkauf, and Yermack's dataset spans 2017 through 2025, a period that includes the 2020 and 2024 halvings, the 2024 Runes fee spike (where fees briefly exceeded 75% of miner revenue), and long stretches where fees sat below 2%. The association between fee gaps and block races persisted across those regimes. Four signals provide a forward-looking view. Adjacent-block fee gaps mark periods when a prior block carries the largest extra prize. Competing-block frequency shows whether hash rate is contesting prior heights more often. The first seconds after a high-fee block are the timing window highlighted by the study. Lock-field coverage across wallet implementations describes the available defense. The paper's title captures the structural tension: Unruly by Design. Bitcoin's issuance schedule guarantees that fee dominance grows with each halving. The instability the authors document is baked into the protocol's own reward curve. As the subsidy halves again in 2028, the block-level fee incentives that produced measurable hash-rate contestation in 2017, 2025 data will carry proportionally more weight. Bitcoin last changed hands at $77,983 per CoinGecko at 14:00 UTC Aug. 26, down 1.60% in 24 hours. The 7-day close series from CoinGecko shows the move from $69,254 on Aug. per CoinGecko, 20 to $79,097 on Aug. 26, a $9,843 range that itself produced the kind of block-to-block fee variation the NBER paper flags as a security signal.