During the Ordinals fee spikes (May 2023, Dec 2023), fees occasionally exceeded the block subsidy. Carlsten et al. (2016) predict that fee-dominated mining introduces incentives to fork the chain tip (undercutting) rather than extend it, which would show up as elevated same-height competition rather than deep reorgs.
Is there any published dataset or measurement of stale block rates broken down by period, that would allow checking whether high-fee blocks correlate with increased forking? forkmonitor.info seems to have historical data but I'm not sure how far back it goes or whether anyone has analyzed it against fee data.
There are datasets that make this test possible, although stale-block data has some important limitations.
The most useful open source I know is the Bitcoin Stale Block Dataset maintained by bitcoin-data. It currently contains more than 2,000 observed stale blocks, with headers available for a large subset and complete block data for many of them. It also provides a rolling stale-block-rate chart and a downloadable CSV, so the observations can be matched by height or time against block fees, subsidy, mempool conditions and miner identity.
ForkMonitor is useful for inspecting individual competing blocks and recent chain-tip conflicts, but it should not be treated as a complete historical census. Stale blocks are not preserved in Bitcoin’s canonical blockchain, so every historical dataset depends on one or more monitoring nodes having seen and retained the losing block. A stale block that never reached the observer is effectively invisible. This makes the data suitable for event studies, but less suitable for claiming an exact network-wide stale rate without correcting for changes in monitoring coverage.
There is also an older academic dataset from Neudecker and Hartenstein’s An Empirical Analysis of Blockchain Forks in Bitcoin. It includes raw fork observations, first-seen timing, propagation measurements and information about which competing block eventually won. That dataset is particularly useful for separating ordinary propagation races from potentially strategic behaviour, although it mainly covers an earlier period rather than the 2023 Ordinals episodes.
One major warning applies to May 2023. The open stale-block dataset specifically annotates a Bitcoin Core block-relay bug around height 789,000 that caused roughly ten stale blocks within one week. That overlaps almost exactly with the first major Ordinals and BRC-20 fee spike. So an apparent May 2023 increase in stale blocks cannot automatically be attributed to fee-driven undercutting; the relay incident is a very strong confounder that would need to be excluded or controlled for.
More importantly, Carlsten-style undercutting is not predicted simply by a high absolute fee level. The relevant incentive is the reward available from replacing the latest block relative to the reward from honestly extending it. A miner may try to recreate some of the previous block’s transactions while deliberately leaving enough fees in the mempool to attract other miners onto the competing branch. The empirical variable should therefore be something like the difference or ratio between the previous block’s fee reward and the expected fees available in the next block, rather than merely whether fees exceeded the subsidy.
Interestingly, a new July 2026 NBER working paper by Fabian Schär, Dario Thürkauf and David Yermack, Unruly by Design: Fee Volatility and Strategic Attacks in Bitcoin Mining, appears to provide the closest published measurement so far. The authors use mempool data covering 2017–2025 and report empirical evidence that sharp differences in fee rewards between adjacent blocks are associated with behaviour consistent with attacks on recent blocks. They describe it as, to their knowledge, the first empirical evidence that fee volatility already affects miner behaviour in the direction predicted by the theoretical literature. It is a new working paper rather than settled consensus, but it directly addresses the gap raised in the question.
For a clean independent test of the Ordinals periods, I would:
- Merge the bitcoin-data stale-block CSV with canonical block data by height.
- Calculate fees, subsidy, total reward and the fee difference between adjacent blocks.
- Use the actual arrival or first-seen time where available, rather than trusting miner-supplied block timestamps.
- Control for block propagation improvements, unusually short block intervals, block size or weight, mining-pool identity and known relay incidents.
- Exclude or separately model the May 2023 Bitcoin Core relay-bug window.
- Compare May and December 2023 not only with quiet periods, but with earlier high-fee episodes such as late 2017 and 2021.
- Inspect the contents of the losing block. A genuine undercutting candidate should look economically unusual—for example, a competing block built on the previous height with a transaction selection designed to leave attractive fees for miners who might follow it.
My prior would be that the Ordinals spikes may have increased the economic incentive for tip competition, but the raw number of stale blocks is probably too small and too heavily affected by propagation and observation bias to make a strong claim from those two episodes alone. The December 2023 window may be cleaner than May because it does not share the same obvious relay-bug confounder.
So the answer is: the data exists, the test is feasible, and there is now early empirical research supporting the broad mechanism—but proving deliberate undercutting requires more than showing that stale blocks rose when fees were high.
