BIP-110: 20% of nodes, zero blocks and a stalled fork

On August 14, 2026, at 20:17 UTC, 5,429 of the 27,148 reachable nodes on the Bitcoin network —20.00%— were running Bitcoin Knots, the alternative client supporting BIP-110. Scanning the chain block by block that same day, 0 of the 841 blocks mined within the mandatory signaling window declared bit 4 of the proposal. One in five nodes and not a single block. That gap is the entire story. Bitcoin lacks a governance mechanism that aggregates will. What it has are three censuses —nodes, hashrate (computing capacity dedicated to mining), and exchanges— that measure different things and only align as long as no one subjects them to a cost. BIP-110 misaligned them simultaneously on August 8, 2026, making it the most significant governance data point for the network since 2017. This article recounts the signaling using a declared methodology, explains why the 0 out of 841 does not mean what it seems, calculates how long the minority chain would actually take to adjust its difficulty, and dates the closing of a window that almost everyone considered closed on August 10.

What does each of the three Bitcoin censuses measure?

Public discussion about who decides in Bitcoin suffers from a confusion of units. When someone says "the community rejected the proposal," they are mixing three populations that are counted differently, overlap very little, and have entry prices that differ by several orders of magnitude.

Setting up a node costs a disk and a connection. Signing a block costs hardware, energy, and the risk that the reward is paid in a currency with no buyer. The three censuses look similar as long as the question is rhetorical; the moment someone puts a price on the answer, they diverge.

The node census is the best-publicized and the most misinterpreted. The 27,148 from the bitnodes snapshot are reachable nodes: those that listen for incoming connections and respond. It excludes those behind NAT and some of those that only communicate via Tor, so it functions as a proxy for client distribution, not as a full network census. With that caution stated, the series remains informative: Knots went from around 22% of reachable nodes on July 27, 2026, according to counts published by the specialized press at the time, to 20.00% measured on August 14. The failure of the fork cost it about two percentage points of share in two weeks—a modest erosion given the magnitude of the episode.

Why does our count show 0 out of 841 when the press repeats 2.53%?

The two figures circulate together and are not comparable. The 2.53% —51 blocks out of 2,016— corresponds to the adjustment period prior to the mandatory window, when supporters and detractors were still mining on the same chain and signaling was voluntary. The 0 out of 841 corresponds to period #477, already under mandatory signaling, with the two populations separated.

That zero is by design, and it is worth stating before using it: under mandatory signaling, nodes enforcing BIP-110 reject blocks that do not signal, so those supporting the proposal began building on a different chain tip. Their blocks —the two that eventually existed— are not on the main chain and therefore do not appear in the count. The zero does not measure an evaporation of support. It measures that the separation of populations was complete.

Methodology matters because it makes the figure defensible. The window 961,632-963,647 coincides exactly with difficulty adjustment period #477, because 961,632 = 477 × 2,016, and this coincidence allows the count to be compared against the proposal's threshold without cherry-picking the segment. The bit declared by the proposal is bit 4, according to the deployment parameters published in the BIP itself, with a threshold of 1,109 blocks out of 2,016 (55.01%). And there is a detail that prevents a classic false positive: bit 4 falls outside the standard version-rolling mask (0x1fffe000, bits 13 to 28), the range of bits that equipment can freely alter for ASICBoost, a hardware optimization that reorders part of the block header to save computation. By counting only that bit, mining optimization does not contaminate the count. Of the 841 blocks read between 961,632 and 962,472, none had it active.

What exactly did BIP-110 propose and why was it forcibly activated?

BIP-110, titled Reduced Data Temporary Softfork, was assigned in the BIPs repository on December 3, 2025, and currently appears there with a "Closed" status. The published header declares a single author, "Dathon Ohm," a pseudonym with an associated email address; the attribution of the original draft to Luke Dashjr, a veteran Bitcoin developer, comes from event coverage, and Dashjr publicly backed the proposal's subsequent direction. Both should be taken for what they are: a declared authorship in a public document and a journalistic attribution.

The technical content is a set of temporary restrictions on arbitrary data: new output scripts limited to 34 bytes, OP_RETURN outputs to 83 bytes, data pushes and witness elements to 256 bytes, and several Taproot functions disabled for the year the deployment would last. UTXOs created before activation were exempt. All this comes as a reaction to a previous policy change: Bitcoin Core v30 increased the default datacarriersize value from about 83 bytes to 100,000 and allowed multiple OP_RETURN outputs per transaction, while Bitcoin Knots maintained the strict limit. If you lack context on what a fork is and what types exist, the corresponding entry in the crypto glossary covers it, and Bitcoin from zero explains why it matters who runs which software.

The governance aspect is more interesting than the technical one. BIP-110 was framed as a UASF —User Activated Soft Fork—: nodes enforcing it impose the rule even if miners do not sign it. With 55% of blocks signaling, the proposal locks in and the chain remains united. Below that, nodes enforcing the rule reject blocks that do not signal and split off. The mechanism is designed to force a choice with consequences, and in August 2026, it did exactly that: BIP-110 was resolved against those who had convened it.

What happened between August 8 and 10, 2026?

The chronology, with the source where each point is verified, is short and quite conclusive.

A methodological warning regarding the last row of the minority chain: as of August 14, 2026, we have not located a public explorer serving that chain, so its height is not independently verified in this article. The last height documented by the specialized press is 961,633, and secondary sources fluctuate between "one block" and "two blocks" and between 26, 48, 88, and 111 blocks of lag depending on when they took the snapshot. This fluctuation is itself a data point on the quality of the event's coverage.

Why is a pool's signal not equivalent to a miner's vote?

On August 10, 2026, CoinDesk documented the case that dismantles the reading of hashrate as a count of will. The operation Simple Mining mined block 961,634 —two blocks after the split point, on the main chain— through Ocean, the pool that since July had signaled BIP-110 by default, and without activating the bit. It was able to do so because DATUM, the pool's own protocol, allows each miner to build their own block template with their software and decide what to signal while still contributing computation and receiving a share of the payout. The phrase left by the operation —"hashrate is a vote that cannot be forged"— well describes the mechanism and, applied to the case, proves the opposite of what is usually cited: the block signature belongs to the pool, the machine belongs to someone else, and the two positions can move in opposite directions.

The map of public positions of large pools points in the same direction. Ocean produced practically all the blocks that eventually signaled. F2Pool explicitly declined. AntPool did not comment and mined the first non-signaling block that BIP-110 nodes rejected, which is the event that materialized the split. And Foundry USA, with about a third of global hashrate, conducted the closest thing to a formal vote seen on the network: a ballot weighted by the average hashrate of each miner over a ten-day window, a 51% threshold of participating computation, and non-responses counted as "no." That design —silence counts as a negative— is an implicit recognition that participation in these processes is low and that the absent cannot be treated as supporters.

From this comes the 99.85% of hashrate that, according to the most cited estimate during the episode, remained on the original chain. This figure should be read for what it is: a distribution of computing capacity between two chain tips, and not confused with the security metrics used when discussing 51% attacks and other consensus vectors. What occurred here was an unattended chain, with no one attempting to rewrite it and almost no one willing to sustain it.

How long would the minority chain actually take to adjust its difficulty?

This is the number that summarizes the episode, and it has circulated in three incompatible versions in four days. On August 9, 2026, several reports published about 350 days until the minority chain's first difficulty adjustment. On August 11, CoinDesk reported that a live monitor had raised that estimate to 6.3 years. Our own calculation, using the difficulty inherited from the August 14 extraction, yields even higher figures.

The BIP-110 minority chain inherited the difficulty of the network that no longer supports it: 127,479,855,693,691 in the August 14, 2026 extraction, i.e., about 127.5 trillion (1.27 × 10¹⁴). That difficulty implies an equilibrium hashrate of about 912 EH/s to produce a block every ten minutes. With the split at 961,633 and the end of the adjustment period at 963,647, there are 2,014 blocks left to mine before the difficulty can drop.

None of the three minority chain scenarios come close to 350 days; the most benevolent is twenty-five times higher. The gap with the 6.3 years from the monitor cited by CoinDesk comes entirely from the hashrate assumption: 6.3 years requires the minority chain to retain around 0.6% of the network's computing capacity, four times the 0.15% estimate that circulated during the episode. This is the reason for publishing scenarios with the assumption stated upfront instead of a single figure: the answer depends entirely on a parameter that no one is measuring. And all this assumes continuous mining, when the verified fact is that it stopped. Inherited difficulty is the structural punishment for splitting off: for 2,014 blocks, the small chain competes against an adjustment calibrated for a network a thousand times larger, and can only pay rewards at a rate a thousand times lower. It is the same mechanism that squeezes miner margins on the main chain when difficulty rises faster than price, analyzed in detail in the hashprice squeeze, but taken to the extreme.

Why did the census that didn't vote end up deciding?

In 2017, with the Bitcoin Cash split, exchanges published guides before the event: when they would stop deposits, what ticker they would use, how they would credit balances. In August 2026, with BIP-110, no major platform or custodian commented on the minority chain before the split, and none listed it afterward. Coinbase and Kraken limited themselves to reporting normal operations after the event had already occurred. The result was produced without a veto and without a statement, by pure absence of decision.

The chain of events is mechanical. A chain without a listing has no reference price. Without a price, there is no predictable income for the miner. Without income, the hashrate leaves, and with 0.15% of computing capacity, blocks stop arriving. The computation supporting BIP-110 withdrew following that accounting chain. That is the third census, the one that does not signal, does not appear in any time series, and in practice determines if a fork has a market.

And it is the census with the highest cost asymmetry of the three: for a regulated platform, listing a new chain means technical integration, a distribution policy among customers, and legal liability for duplicate balances, while not listing it costs nothing and requires no justification to anyone.

Is the BIP-110 window still open and what remains afterward?

Almost all coverage buried BIP-110 on August 10, 2026. Formally, the BIP-110 mandatory signaling window had not closed on August 14 and will not close until block 963,647. At the actual pace of the last 1,008 blocks measured that day —10.20 minutes per block, with the tip at 962,472 at 20:15 UTC— 1,175 blocks were remaining, placing the close around August 23, 2026, at 04:04 UTC. Nine more days of formal life for a proposal considered dead a week earlier.

Arithmetically, locking in is not impossible either, and that isolated statement is misleading. The honest way to say it is with the requirement: 1,109 signaling blocks are needed and 1,176 remain to be mined, so 94.3% of all remaining blocks would have to declare bit 4, starting from zero and with the hashrate that was doing so now stopped. The BIP also sets a maximum activation height at block 965,664, around September 1, 2026. In the calendar, two deadlines remain to expire; the threshold, however, fell out of reach on August 9, when the tip signaling it stopped producing blocks.

A technical risk remains that did not materialize this time. BIP-110 did not incorporate replay protection: if the minority chain had gained traction, a transaction signed on one of the two chains would have been valid on the other, resulting in moving balances without having decided to do so. With two blocks and no market, the risk remained theoretical. The precedent matters because the next event in the series arrives shortly: the eCash hard fork is set for block 964,000 and, at the same actual pace measured on August 14, 2026, that block falls toward August 25 at 16:06 UTC, not the August 21 at 15:00 UTC published by the press —that date would require 6.39 minutes per block, 36.1% below the ten-minute target. And there, replay partially reappears: Sergio Demian Lerner, co-founder of Rootstock Labs, published on May 8, 2026, that eCash protection "is only partially implemented" because it "does not clearly separate transaction signature domains." The technique to do it correctly has been documented since 2017 in the SIGHASH_FORKID mechanism adopted by Bitcoin Cash, so in both August episodes, it was available and left not fully applied. That a change in consensus rules forces a coordinated jump without a safety net is the same lesson left by the Orchard counterfeiting bug in Zcash, from the other extreme: there the hard fork was mandatory and here it was optional, and in both cases, the cost falls on those who are unaware.

A new proposal also remains. After the halt, the author of BIP-110 announced that its supporters are working on a change to the proof-of-work algorithm with the stated goal of "firing the miners," and Dashjr and Roughnecks —the mining operation that provided most of the minority chain's computation— publicly backed that direction. The arithmetic consequence is direct: changing from SHA-256 renders the equipment of the very miners who supported the minority chain useless —starting with those who provided those over 15 EH/s— and turns the project into a different currency, with its own market to build.

And the framework remains, which is what survives the event. Any future proposal on Bitcoin will pass through the same three filters in the same order: first the cheap census, which gives generous and flashy signals; then the expensive census, which corrects them; and finally the silent census, which publishes nothing and decides if a market exists to get paid. The 20.00% of nodes on August 14, 2026, and the 0 out of 841 blocks on the same day are the same population portrayed with and without the bill in front of them.

Sources and links: BIPs Repository — BIP-110, header and deployment parameters · Bitnodes — reachable nodes snapshot (Aug-14-2026, 20:17 UTC) · CoinDesk — a miner rejects BIP-110 mining in a pool that signaled it (Aug-10-2026) · CoinDesk — two blocks and a stop (Aug-9-2026) · CoinDesk — live monitor raises estimate from 350 days to 6.3 years (Aug-11-2026) · crypto.news — the gap widens and signaling for period #477 · CryptoSlate — resumption and proof-of-work change plan · Bitcoin.com News — drop in signaled hashrate · Crypto Briefing — Foundry USA's hashrate-weighted vote · AMINA Bank — pool positions, custodians, and August fork calendar · Sergio Demian Lerner (Rootstock Labs) — partial replay protection in eCash (May-8-2026). Signaling count, node share, and time estimates: proprietary calculation based on the August 14, 2026 extraction.