A ratio without its denominator is a rhetorical device.

The same eligibility loss has been published as 236 million tonnes and 32 per cent, as around 240 million and roughly a third, and as 268 million and 30.6 per cent. Every one of those is correct. They divide by different things and none of them says which. In a market where the headline numbers are chosen rather than derived, a ratio without its denominator is not a measurement. It is a rhetorical device, and it will be treated as one by the first auditor who asks.

Three true numbers, three different questions

The figures above are not contradictory. They answer different questions.

Unretired credits affected, against the total market. Credits in circulation. Total available supply. Change the denominator and the percentage moves by several points without a single fact changing underneath it.

None of this is dishonest, and the analysts producing these numbers are careful people. The problem is downstream: once a percentage is separated from its denominator and enters a slide, a board paper or a press release, the information required to interpret it is gone and cannot be recovered from the number itself.

Why this market is unusually exposed to it

Carbon has three properties that make denominator drift almost inevitable.

The population is genuinely ambiguous. Issued, retired, cancelled, held, in circulation, eligible, on hold. A credit can be several of these at once, and the boundaries between them are set by registry conventions that differ between programmes.

The asset is heterogeneous. A tonne is a tonne only in the accounting. A tonne of durable removal and a tonne of legacy avoidance are different instruments with different failure modes and different prices, and averaging across them produces a number that describes nothing.

The market is argued about rather than measured. When a market is contested, participants reach for the framing that helps. That is human and it is not going to stop. The only defence is a convention that makes the framing visible.

The convention: a ratio is a type, not a number

We treat every published ratio as a structured object rather than a value. It carries four things and it cannot be rendered without them.

The numerator, with what it counts. The denominator, with what it counts. The as at date, because populations move. The source, resolvable to the document the figures came from.

In practice this means a document generator refuses to print a percentage that does not resolve to a complete ratio object. Not a style guideline. A hard failure, enforced in code, because style guidelines are the first thing to go under deadline pressure and this is precisely the error that survives review: the number looks right, it is internally consistent, and it is only wrong in a way that shows up when someone else tries to reproduce it.

How it reads in practice

Consider our own anchor figure. It is tempting to write "California's buffer pool is nearly a quarter consumed", which is memorable and nearly useless. The version that survives scrutiny is longer and it is the only one worth publishing.

8.97 million tonnes of reserve credits retired against 36.23 million tonnes ever contributed, which is 24.8 per cent, as at the retrieval date, from the regulator's own issuance and reversal tables. Roughly thirteen years into a programme whose permanence claim runs for one hundred.

Now the reader can check it, disagree with the choice of denominator, or reuse it in their own analysis with confidence about what it means. That is what makes a figure citable rather than merely quotable.

The second discipline: abstention

The same instinct produces a rule that surprises people. Where attribution is not published, the honest output is not an estimate. It is a labelled abstention with a reason.

If a registry records that units were relinquished but does not publish which holder bore the loss, the record says so. If a delivery shortfall is visible in aggregate but cannot be attributed to a named contract, the record says so. The alternative, filling gaps with a modelled guess, produces a dataset that looks complete and cannot be defended anywhere it matters.

A model that always has a number is not confident. It is unfalsifiable. Every underwriter and every auditor knows the difference, and the willingness to say "the evidence does not support a view here" is what makes the rest of the numbers worth reading.

Why a small convention carries weight

None of this is sophisticated. Stating denominators and declining to guess are the sort of thing everyone agrees with in principle and abandons under pressure. That is exactly why doing it consistently is a differentiator rather than a hygiene factor.

A record that becomes the market's shared reference has to survive being checked by people who would rather it were wrong. It survives on two things: every figure resolving to a document, and the discipline never to publish a number the evidence cannot carry.

If that feels pedantic, wait until the first auditor asks where the number came from.