Decent Newsroom
Sign In

Full Blocks, Empty Fee Market

By Roger November 11, 2026
Bitcoin's blocks are 99.8 percent full and fees pay 0.66 percent of miner revenue. Measured on a synced node today: a full block needs 312 sat/vB to replace the subsidy. Three papers in thirty hours attack that margin, and two payout layers failed on the way.
Full Blocks, Empty Fee Market

Full Blocks, Empty Fee Market

At 09:21 CET this morning, block 975,411 was mined into the Bitcoin chain. It was 99.8 percent full: 3,993,692 of the protocol’s four million weight units, with 2,474 transactions packed into it. Look back over the preceding 2,016 blocks, one complete difficulty period, and the same picture repeats: the median block was also 99.8 percent full.

Now the second number. Over those same 2,016 blocks, miners collected 6,300 BTC of subsidy and 41.57 BTC of transaction fees. Fees were 0.66 percent of miner revenue.

Both figures come from the same measurement, taken on a fully synced node at 09:30 CET today (block height 975,411, difficulty period 973,396–975,411). They describe a network whose blocks are always sold out and whose fee market is, for practical purposes, switched off. The first number is quoted constantly as evidence of demand. The second decides whether Bitcoin can pay for its own defence after 2140, or 2036, or, if you are impatient, next April.

Nothing mysterious separates the two. It is a price, and prices are set at the margin.

The measurement, in full

The numbers, so that anyone can check them against a node:

  • 2,016 blocks: 6,300.00 BTC subsidy, 41.57 BTC fees. Fee share 0.66 percent.
  • 8,242,944 transactions in that window, an average of 4,089 per block.
  • Block fullness: median 99.8 percent of maximum weight.
  • Fee rates: median of the block-median rate is 1 satoshi per virtual byte. The median of the tenth percentiles is 0 sat/vB; the median of the ninetieth percentiles is 3 sat/vB.
  • Mempool at 09:30: 44,669 unconfirmed transactions, 15.14 MB, more than fifteen blocks of backlog.
  • Bitcoin Core’s own estimator: 1.23 sat/vB to get into the next block, 1.08 sat/vB for a 144-block horizon, 1.0 sat/vB for 1,008 blocks.

Fee share of miner revenue, block by block, 2,016 blocks

Fee share of miner revenue for each of the 2,016 blocks in the last difficulty period. Own measurement, 11 November 2026.

A backlog of fifteen blocks, and priority space costs one satoshi per virtual byte. That is not a market clearing at a scarcity price. That is a market clearing at the floor while the queue does the work of rationing.

Price context matters here, because it tells you what the subsidy is currently worth. Bitcoin’s last daily close on Binance was $73,734.01, against the intraday high of $126,199.63 set on 6 October 2025, a drawdown of 41.6 percent. James Check’s note on 10 November opened with the sentence “Bitcoin is hanging onto $75k by its fingernails,” and flagged the $75k–$72k band as the deepest a healthy correction should go. A falling price cuts the dollar value of the subsidy one for one, while the fee rate does not move at all.

Why a full block is not an expensive block

Here is the arithmetic that nobody puts in a headline. A maximum-size block is 4,000,000 weight units, which is 1,000,000 virtual bytes. If every one of those bytes paid one satoshi, the block would yield 1,000,000 satoshis, or 0.01 BTC. The subsidy for that same block is 3.125 BTC. In other words, at today’s median fee rate, a completely full block earns about 0.3 percent of what the subsidy pays.

To match the subsidy with fees alone, a full block would have to average 312.5 sat/vB, against a median of 1.

Average feerate needed to replace the block subsidy

Own arithmetic: the average feerate a completely full block would need to earn the subsidy through fees, by halving epoch, against today’s median of 1 sat/vB.

A block-size cap is a quantity, not a price. When supply is capped and demand at the floor price exceeds supply, the block fills. It also fills when demand is nearly free. “Full” tells you that the queue is non-empty. It tells you nothing about what anyone would pay if the queue were longer. The fee rate is set by the marginal transaction that gets in, and the marginal transaction is not a payment.

Three consecutive blocks sampled this morning make that concrete. Block 975,412 contained 4,755 transactions, and 2,889 of its outputs were OP_RETURN, the “null data” output type used to write arbitrary bytes into the chain. 2,884 of those 4,754 transactions, 60.7 percent, carried one. They took 39.2 percent of the block’s weight, paid 7.2 percent of its fees, and cleared at a median of 0.12 sat/vB, while the ordinary payments in the same block paid 1.20 sat/vB. Its predecessor looked nothing like this: 6.5 percent of transactions, 4.8 percent of weight, 5.6 percent of fees, at 2.63 sat/vB. And the one before that was 88.1 percent data transactions, holding 73.4 percent of the weight for 21.5 percent of the fees. Across all three blocks, 8,614 of 13,551 transactions, 63.6 percent, carried an OP_RETURN.

Share of block taken by OP_RETURN transactions in three consecutive blocks

Three consecutive blocks, 11 November 2026: transactions carrying an OP_RETURN output, and how much of the block they took.

So the marginal buyer of block space this morning was a script writing bytes at one tenth of a satoshi per virtual byte, and a miner was happy to include it, because the alternative was empty space. This is what a residual market looks like: the subsidy pays for the block, and the fees are a rounding error that the cheapest willing bidder gets to define.

An Austrian would put it differently and more sharply. The block-size cap is a price control that takes the form of a quantity. Everything that cannot be bought with money gets allocated by queue position, and queue position is bought with fees, but only at the last byte, where the competition is thinnest. Rationing by cap does not create value. It creates a line, and then it measures the length of the line and calls it demand.

The arithmetic of the halvings

None of this is an emergency, and the reason is the schedule. The subsidy falls by half roughly every four years: 3.125 BTC now, 1.5625 after the 2028 halving, 0.78125 in 2032, 0.390625 in 2036. The fee rate needed to replace the subsidy in a full block falls with it: 312.5 sat/vB today, 156.25 in 2028, 78.1 in 2032, 39.1 in 2036, 19.5 in 2040.

That sounds like relief until you place the two series next to each other. That requirement halves every four years. Market price has been sitting at 1 sat/vB for as long as the chain has had a soft floor. Two orders of magnitude separate the two lines, and a decade of halvings only closes one of them. The security budget is not a problem for the 2030s. It is a problem that has already been priced, at one satoshi per byte, and simply has not been invoiced yet.

This is my own arithmetic on measured inputs, and it rests on one assumption worth stating: that blocks continue to sell out at 1,000,000 virtual bytes. If the cap is raised, the required rate rises proportionally, because the subsidy would then be spread over more sellable space. If demand for block space grows faster than the cap, the required rate falls. Neither of those is a prediction; together they are the shape of the constraint.

Three answers, published within thirty hours

What makes this week interesting is not the number. It is that three separate groups published work addressing exactly this margin, in the same thirty-hour window, from three different directions.

Who gets the coinbase. A discussion document posted to Delving Bitcoin on 11 November at 03:36 CET, “Decentralized Coinbase Payout Protocols (DCPP)”, asks how a pool can pay its miners without being a pool. The history it starts from is P2Pool, which worked and then failed, not on consensus but on payouts. The document names three constraints: paying every miner in a distinct coinbase output does not scale; the payout layer is an attack surface; and Bitcoin Script is too limited to build sophisticated settlement logic.

One attack it describes deserves the name it was given. Call it the Rich List Attack. A coinbase can fit N payout outputs. When more than N miners want one, the pool needs a rule, and the obvious rule is “pay the top N by shares over some window”, which is a rich list. A large miner can split its hashrate across many identities, each still outranking the smallest legitimate miners, take several slots, and push the small miners out. There is no identity system in a decentralized pool to stop this, and splitting hashrate costs nothing. From there the paper walks through the two ways out. If the excluded miners’ share is redistributed to the big miners, the attack is profitable and the pool is a cartel with extra steps. If instead the unpaid share becomes a credit carried forward, the pool is holding a liability, and the whole point of a decentralized pool, that nobody holds your money, is gone.

The proposals it reviews are real: Chris Belcher’s 2017 hub design with payment channels and a six-month OP_CSV escape hatch for the block finder; Braidpool’s rolling coinbase aggregation, where a 66-percent-plus-one threshold of recent block finders signs a payout transaction and then deletes its key shares so the funds cannot be spent any other way, batching settlement across a difficulty window; P2Poolv2’s market makers who buy sharechain outputs for Lightning before the pool finds a block. Each one is vulnerable to some form of payout settlement griefing. And each fix that is not a threshold introduces a fee or a minimum payout, which the document notes is “not free” because it makes the scheme uncompetitive with a custodial pool.

That is the honest core of it: a rank-based rationing rule is a price control, and the arbitrage is identity. Take away the price, and you get a line of pseudonyms. Put the price back, and you are less competitive than the custodian you were trying to replace.

How to keep a fee-only chain stable. A paper on arXiv the same morning (2611.13252, “Splitting for Stability”) attacks the other end. When fees dominate miner revenue, the winner of a block takes everything, and a miner with a large share of hashrate can profit by undercutting the fee level or by building on an older block with a richer reward. That is not a hypothetical about malicious actors; it is what a winner-take-all auction does when the prize is a flow rather than a stock.

The proposed fix is a split: the winning block producer keeps a fraction α of the block reward, and the remainder is distributed among the other miners. That changes the payoff of a fork from “everything, if you win” to “α of everything, and a share of the honest chain if you lose”, which is precisely the shape of incentive you want if you believe miners respond to relative reward rather than sunk capital. The authors derive a local condition for one-step deviations, run simulations, and report parameter regions where honest behaviour dominates, with the caveat that in the regions where forking does dominate, the learned strategy is the mildest variant studied, and that the scheme does not appear to worsen selfish-mining incentives.

It is a model and a simulation. There is no deployment, no field data, and no evidence that hashrate behaves the way the simulation’s agents behave. Treat it as a design space being mapped, not as a fix that exists.

Why “useful work” does not save energy. The third paper is the one I would hand to anyone who still argues that Bitcoin’s electricity is wasted and that the fix is to make mining compute something useful. arXiv 2611.13131, “Useful Work Can Increase Energy Use”, takes proof-of-useful-work seriously, in the Ball/Rosen/Sabin/Vasudevan construction from 2017 and the more recent matrix-multiplication instantiation by Komargodski and Weinstein, and then shows that switching to it may increase total electricity consumption.

The mechanism is Jevons, and the authors say so. A proof-of-useful-work miner is paid the same subsidy, but part of that payment now also buys a computation that someone might have wanted anyway. The subsidy is therefore a discount on the useful work: it lowers its effective price. If demand for that work is elastic, meaning a discount of x percent raises the quantity demanded by more than x percent, total electricity consumption rises rather than falls. If demand is inelastic, consumption falls. And even then, the authors note, the consumption attributable to the scheme never disappears entirely, because the subsidised work would not have been produced at all without the subsidy.

In a simple model with electricity-only costs, no overhead, and an unchanged subsidy, they prove the result; the paper says the analysis extends to less restrictive settings with adjusted thresholds. It is the cleanest piece of economics published about Bitcoin this year, and it says something unfashionable in both directions: the “waste” critique is not answered by redirecting the subsidy into useful computation. It is answered, if at all, by the subsidy ending.

Meanwhile, the payout layers that already exist failed twice

Everything above is about a margin that is mispriced. The two stories below are about what happens when you delegate the payout entirely, which is what almost everyone actually does.

On 8 November, LNbits told its users to shut their servers down immediately. There were active attacks. LNbits is the open-source multi-user account layer that a great many Lightning operators run on top of their nodes, founded in 2019 by Ben Arc and widely described as the WordPress of Lightning. On 10 November it shipped version 1.7.0 with the words “SECURITY UPDATE”, and the release guidance is unusually blunt: versions before 1.7.0 are affected, and “funds running on prior versions of LNbits are at risk of being stolen.” Updating alone is not enough, because a patch cannot remove access an attacker already has. Operators were told to stand up a fresh server, rotate every user wallet key as Super Admin, replace funding-source credentials (API keys, macaroons), rotate the other secrets in settings, and re-check the connections for extensions including NWC Provider, LNDHub, Bolt Cards, TPOS, Bleskomat and FOSSA.

One operator published numbers. Kevin Ravensberg, who runs the Telegram tipping service BitTipBot, said attackers created new accounts on his instance and withdrew 2,285,545 satoshis between roughly 09:22 and 14:00 UTC on 8 November. That is 0.0229 BTC, about $1,680 at this morning’s price. The project has not published an aggregate loss figure. A dollar amount like that is trivial next to a single block reward. The shape of the failure is not: the attack surface was the account layer, not the Lightning Network. A multi-user hot wallet with its own login, its own extensions and its own credential store is a bank with a smaller lobby, and it fails like one: by the front door, with valid credentials.

Four weeks before a transfer ban even takes effect, the custodial version of the same lesson arrived from the other side. FinCEN’s order against MEXC Global was published in the Federal Register on 10 November, which means US financial institutions are not bound by it until 10 December. Binance did not wait. It announced that it will refuse transfers to or from MEXC and its affiliates, and that funds arriving from them will be held pending a compliance review that can end with the user removed from the platform. The notice lists thirteen names; one of them, MEXC (Comoros) Ltd, appears nowhere in FinCEN’s order, which Unchained verified by comparing the two documents. The order invokes Section 9714 of the Combating Russian Money Laundering Act and has no end date. FinCEN puts the volume linked to Russian illicit finance, sanctions evasion and cybercrime at $16.7 billion between May 2021 and May 2026, with about $1.5 billion in Russian sanctions evasion and about $1 billion in business with more than forty Iranian exchangers. Treasury Secretary Scott Bessent’s line was: “Financial institutions cannot have it both ways: Either you support the Iranian regime, or you have access to the U.S. financial system.”

You can think the underlying enforcement is justified and still notice what it demonstrates. A balance on an exchange is not a claim on bitcoin. It is a claim on a company’s continued permission to move it, priced at whatever the compliance department decides on the day. The people who found that out this week did not lose a key. They lost an authorisation.

And the political route, in the same week, buys the asset while carefully preserving the custodian. A bill filed in Texas on the first day of prefiling, House Bill 815 by Representative Ron Reynolds, would let each city and county invest up to $10 million of public money “in Bitcoin or another cryptocurrency”. Its text runs to two sections. It is silent on custody, silent on audits specific to digital assets, and silent on how a loss would be handled. It deliberately does not import the safeguards written into last year’s state reserve law: the requirement that an asset hold at least a $500 billion average market capitalisation over twenty-four months, and the cold-storage and qualified-custodian provisions. The authority, if it becomes law, starts on 1 September 2027. It is a Bitcoin bill that also authorises everything else, with no rule for who holds it.

The payout rails are being rebuilt, by the people who already own them

One more primary document belongs here, because it is the cleanest statement of what a payout layer looks like when the issuer designs it. Speaking to the European Parliament’s economics committee on 10 November, ECB executive board member Piero Cipollone laid out the digital euro timeline: a pilot beginning in the second half of 2027, running a beta version without legal tender status, with thirty-six payment service providers selected in July from more than fifty applicants, and merchants to be announced early next year. He also described the infrastructure underneath it. Pontes, launched in September, connects distributed-ledger platforms to the Eurosystem’s TARGET Services, onboarded eighteen market participants on its first day, and on 2 November settled the issuance of the European Stability Mechanism’s first digitally native benchmark bond, one billion euros.

That is a genuine achievement, and it is worth saying so plainly: tokenised settlement in central bank money is not a gimmick, and the ESM bond is a real transaction. But notice the architecture. The ledger is being built by the institution that issues the unit, for institutions it has selected, with a legal-tender status deferred until a regulation passes. Every property that makes a payout final, from who may hold the unit to what it can be spent on and how large a balance may be, is a parameter set by a committee. Bitcoin’s version of that parameter list is a schedule and a script. Both are engineering. Only one of them can be changed by the people who use it.

The record is being repriced too

There is a second market in this week’s news, and it is not about money but about the record.

Microsoft Technology Licensing filed patent application US 2026/0341309 A1 on 29 April 2025; the USPTO published it on 29 October 2026, and it was first flagged by the patent tracker Patentlyze and reported by Futurism on 1 November. The title is bureaucratic (“Near Real-Time Artificial Intelligence (AI) Enhancement of Video or Augmented Reality (AR) Presentations of Live Events”), but the first independent claim describes an agent that selects spoken words to modify “based on a profile of a user”, regenerates the audio, re-renders the frames so the speaker’s mouth matches the substituted words, and does it within one to ten seconds. The examples of what a profile might include are age, language, region, religion, ethnicity, education, and voice or video samples. Two people watching the same broadcast could hear different sentences. There is no disclosure requirement in the filing, and the ground-truth recording is described as optional, server-side, and controlled by whoever runs the pipeline. The filing also describes a fact-checking mode that pulls replacement words from an internet search. That is a system that decides a statement was inaccurate and silently substitutes what it thinks should have been said. The application is not a granted patent; the examiner raised objections on 29 July 2026, and Microsoft did not respond to Futurism’s request for comment.

The same day, on the other side of the same problem, the European Central Bank’s chief economist gave a keynote in London about the science of monetary policy communication. Philip Lane described how the ECB now uses large language models to measure the tone of its own communications, publishes research showing that the information in policy statements improves forecasts of euro-area core inflation several quarters ahead even after controlling for staff projections, and estimates “communication rules” that describe how the bank’s tone responds systematically to inflation and activity. His framing is that communication steers expectations, and expectations shape “the full set of financial market prices across the maturity and risk spectrums”.

Read fairly, Lane is describing measurement and analysis, not manipulation, and he is candid about the limits: the results depend on how questions are formulated and what data the models were trained on. But put the two documents side by side and the structure is identical: the spoken record is treated as an output to be optimised per audience, not as an input that everyone agrees on. One does it with a language model and a viewer profile; the other with a language model and a communication rule. Neither needs a watermark.

The machinery for this is getting cheap, which is the part that should worry anyone who thinks the capability will stay in responsible hands. Meta’s new Muse Ember is priced at $0.10 per million input tokens and $0.50 per million output tokens, against $8 and $40 for Claude Fable 5.2 released the same day, a factor of eighty, as Simon Willison noted. Rewriting a sentence in real time, with matching lips, is a rounding error in a broadcast budget. It will not be one company’s product for long.

Bitcoin’s answer to this is not a policy. It is a data structure. A chain where the record is the input, agreed once, appended, never revised per viewer, is the only kind of ledger that can serve as ground truth for anything else. That is a weaker claim than “Bitcoin fixes speech”, and it is the only claim the technology actually supports.

The steel-man

The case that I am wrong is decent, and it goes like this. Fee revenue is cyclical, not structural: through 2023 and 2024, blockspace demand from new protocol experiments pushed median fee rates into the tens of satoshis per byte for months at a time, and that demand did not disappear, it rotated. There is a fifteen-block backlog in the mempool, which means the queue exists and can be monetised. Lightning and second layers move small payments off-chain and leave the base layer for settlement, which is exactly the base layer’s job, and settlement is where the willingness to pay lives. The subsidy has twenty-nine halvings left before the last satoshi is issued (my arithmetic, from the 50 BTC genesis subsidy and today’s 3.125), which is more than a century of runway; the network has time to grow into the fee market, and the current 0.66 percent is a snapshot of a bear market at $73,734, not a verdict.

Most of that is right, and none of it changes the arithmetic. A fee market that appears for months at a time in the middle of a speculative cycle is not a security budget; it is a windfall. The testable version of the optimistic case is simple: a full difficulty period with blocks at or near capacity and a median fee rate above ten sat/vB, sustained, in a market that is not being driven by a single protocol’s novelty. That has not happened yet. If it does, this article’s premise weakens.

What I could not verify

The honest limits, in order of how much they matter.

One difficulty period is not a trend. I measured 2,016 blocks because that is the natural unit of Bitcoin’s own adjustment, not because it is statistically sufficient to describe a decade.

Two blocks sampled for their transaction mix are two blocks. The OP_RETURN share swung from 6.5 percent to 88.1 percent inside three blocks; that is not a property of the network, it is a property of who happened to be writing this morning. The aggregate over the period (0.66 percent fee share, 99.8 percent fullness, 1 sat/vB median) is robust because it covers 8.2 million transactions. The composition is not robust.

Mempool figures are local. Forty-four thousand transactions is what one node saw. The chain is the same for everyone; the queue is not.

All three papers are models, simulations and a discussion document. The DCPP post is explicitly “a framework for discussion” with no implementation. Splitting for Stability has no field test. The proof-of-useful-work result is a theorem about a model in which electricity is the only cost.

And LNbits has not published what was actually taken, across how many instances. One operator’s 2,285,545 satoshis is a data point, not a total.

What follows

Follow the payout. That is the whole method, and it sorts this week’s news better than any price chart. The mining proposals all ask who receives the coinbase and under what rule; the Lightning failure asks who holds the keys to the account that receives it; the exchange freeze asks who is allowed to move it at all; the patent and the central bank’s communication research ask who edits the record of what was said. Four different layers, one question.

At the base layer the answer is currently uncomfortable. Blocks are sold out, and the marginal byte pays one satoshi. The subsidy covers the difference, on a schedule that no one votes on and no one can renegotiate, and the schedule is the only part of Bitcoin that has never missed a deadline.

That is not a prediction of collapse. It is an observation that the network’s security is currently paid for by a transfer, and transfers end. When this one ends, the bill will be settled by whoever is willing to pay for a byte of block space at a price that covers the cost of defending the chain that holds it. Everything else (the pool payouts, the reward splits, the useful-work subsidies, the custodial shortcuts) is an attempt to answer that question without asking it.

Sources

Write a comment