Thirty-eight to thirty-seven. Twenty-five abstentions. One hundred votes cast in the French National Assembly, and Amendment No. 3898 to the 2027 budget carried on a single ballot.
That is the number I keep returning to. Not the margin — the abstention rate. Twenty-five percent of the chamber declined to record a position on whether converting Bitcoin into USDC should constitute a taxable disposal. A quarter of the body understood, correctly, that the question has no clean answer yet.
The amendment does one thing. It narrows the deferral that currently applies when a French taxpayer swaps one crypto asset for another. Under the existing framework, a crypto-to-crypto conversion is not a realization event; the liability waits until fiat. Amendment 3898 carves stablecoins out of that deferral. Conversions into them count as disposals.
Read the mechanics. A holder moves value from one on-chain address to another, in a transaction that never touches a bank, never produces a euro, and generates a taxable gain denominated in a currency they do not hold.
That is the whole story. Everything else is legislative choreography.
Stablecoins occupy an unstable position in tax law because they occupy an unstable position in ontology.
A euro-denominated stablecoin is, at the ledger layer, an ERC-20 token with a balance mapping and a transfer function. At the design layer, it is a claim on a reserve intended to track one euro. Legislators read the second layer. The ledger enforces the first. Amendment 3898, proposed by Nicolas Sansu, resolves the ambiguity in favor of the second layer: a stablecoin is quasi-cash, and swapping into it is cashing out.
Sansu's argument is that stablecoins should not receive better treatment than fiat currency itself. It is coherent. It also assumes parity.
The bill is not law. It cleared a first reading inside the 2027 budget, and French budget amendments routinely pass the National Assembly and then get rewritten or deleted in the Senate or at the joint committee stage. A one-vote margin with a 25 percent abstention rate is not a mandate. It is a placeholder.
Nothing in the text I have seen addresses the operational layer: which price, at which timestamp, from which source — and whether the clause reaches conversions executed before the effective date.
Here is where the technical problem lives.
Capital gains tax requires a realization price. For a stock, that price is unambiguous: one venue, one closing print, one settlement currency. For a crypto-to-stablecoin conversion, there is no closing print. There is a swap on a pool, a fill on an order book, or a route across three venues, each reporting a different mid at the same block height.
The statute needs a reference price. The reference price comes from an oracle or an exchange API. Both have latency. Both have failure modes. Both have been wrong.
I have audited this layer. In 2020 I reverse-engineered the Uniswap V2 constant-product curve and simulated slippage across more than 500 pools; the finding that mattered was not the formula but the variance — two trades of identical notional, executed seconds apart, settling at prices that differ by more than the tax rate under discussion. A tax computed on "the price at the moment of conversion" is a tax computed on a number that does not exist as a single value.
The art is the hash; the value is the proof. If the state cannot specify the proof — the exact block, the exact feed, the exact aggregation rule — it has not written a tax law. It has written an invitation to litigation.
Then there is the parity assumption. Amendment 3898 hardcodes an invariant: one stablecoin equals one unit of fiat. The market has already violated that invariant. In March 2023, USDC traded near $0.87 for a stretch of hours as reserve exposure became public. Anyone who converted into USDC during that window would, under this amendment, have realized a taxable gain on an asset that had just lost thirteen percent of its value. The law would record a profit at the exact moment of a loss.
Reentrancy isn't a smart-contract problem. It's a state-ordering problem. In 2018 I spent three weeks inside the Parity multi-sig library refusing to sign off until the ownership update sequence was reordered, because a rule that reads state at the wrong moment produces the wrong answer deterministically, every time it is invoked. Tax law is a state machine that has never been formally specified.
Settlement is where it breaks next. The liability is denominated in euros. The asset is a token. The taxpayer never received euros. To pay, they must sell something — usually the very asset the law just declared they profited from. That is a forced liquidation imposed by statute, and it is regressive in a specific way: it penalizes the holder with no fiat income to draw on and rewards the holder with a bank balance.
Enforcement is where the design collapses into theater. A licensed French venue can be compelled to report conversions. A self-custody wallet cannot. A DeFi swap routed through an aggregator cannot. The reporting burden lands entirely on the custodial path — which is to say, on the users who chose to be legible. Everyone else keeps the deferral they always had. I have watched this pattern across KYC regimes for eight years: compliance cost is a tax on honesty, and the honest pay it twice.
Classification, the layer above all of it, is already inconsistent. MiCA defines stablecoins as a specific class of crypto-asset — EMTs and ARTs — inside a framework that treats them as crypto-assets with reserve requirements. Amendment 3898 treats them as cash. Two legal state machines, two different type assignments, same object. When the higher-level framework and the national tax code disagree on type, the implementation does not converge. It forks.
The contrarian read is that the industry is preparing to fight this on the wrong ground.
The reflex argument is innovation flight — capital leaves France, builders relocate, the jurisdiction loses position. It is a weak argument because it is unquantifiable, and legislators have heard it before about every asset class that has ever been taxed.
The strong argument is the one nobody is making: this amendment is unenforceable as written, not because of evasion but because of specification. There is no canonical reference price. There is no defined treatment for depegs. There is no stated rule for retroactivity. The text is a stub.
Twenty-five abstentions say the chamber already knows this. A parliament that believed in the policy would have voted. Instead a quarter of it walked off the record, and the quarter that stayed split evenly. That is not consensus with a narrow margin. That is a body that has not decided, passing something it expects to be rewritten.
The blind spot in coverage so far is the assumption that "backed by lawmakers" means "coming into force." It does not. We do not build for today, and we should not legislate for a version of the market that does not exist — one where a peg holds through every block and every venue prints the same price.
Watch three clauses in the final text, not the vote: the effective date, the retroactivity language, and the price source. If the text names an oracle or an exchange reference, that feed becomes load-bearing infrastructure for the French tax code — and it deserves the same scrutiny we give any other dependency in the stack.
The question I would put to the drafters is simple. If the state intends to price a transaction it did not observe, on a venue it does not operate, using a feed it does not maintain — who audits the number, and what happens when the feed reverts?
That question warrants the market's scrutiny.
