Printing and the fixity of the text in central banking

The objection that should win

Here is the strongest case against everything that follows. A rate decision is not a belief that can be quietly revised next quarter. It is a number — 25 basis points, 50, unchanged — set at a meeting, minuted, and acted upon by every mortgage desk and bond trader within the hour. If the model behind that decision is permitted to keep revising its picture of the economy after the fact, you have not built a better central bank. You have built one that cannot explain, in a court of parliamentary accountability, why it did what it did on the day it did it. Central banking needs a text that holds still. That is the entire discipline of a minute, a forecast round, an inflation report: a fixed statement, dated and defensible, standing against a world that will not stop moving under it. A revisable belief is the opposite of what a rate-setting committee is for.

That objection is not a strawman. It is the working theology of every central bank communications department. It deserves to be taken seriously before it is narrowed.

What the frozen edition gets right, and where it breaks

An economist walks into the meeting with a forecast built on a vintage of data: last month's flash CPI print, the labour force survey from six weeks back, credit aggregates reported with their usual lag, market-implied expectations pulled from the futures curve that morning. That forecast is, in the fullest sense, a printed edition. It is struck off once, in the form of the minutes and the accompanying projections, and it is distributed identically to markets, media and the public. Everyone reads the same page.

The trouble is that the underlying data continue to move after the page is set. Flash CPI is revised. The labour force survey is reweighted. Credit aggregates are restated as more banks report. By the time the next meeting happens, a meaningful share of what justified the last decision has quietly changed value, and nobody re-sets the type. The minutes stand, verbatim, describing a state of the economy that no longer exists in the data warehouse. This is not hypothetical: real-time data revisions to GDP and employment routinely exceed the size of the policy move they informed. The Bank of England and the Federal Reserve both publish, apologetically, real-time datasets specifically because their own past decisions cannot be understood using current data — you must go back to the vintage as it stood on the day.

That is typographical fixity, transplanted into a central bank. One state of the world's numbers, authoritative in tone, ageing from the morning of release. And the fix that gets reached for — footnotes in the minutes acknowledging data uncertainty, fan charts widening the error bars, occasional "special factors" paragraphs — is the errata slip. A leaf of corrections bound into a text whose body cannot be re-set.

Why the live scene does not save you either

The alternative sounds like an obvious improvement: watch everything continuously, decide from the freshest read available, never work from a printed snapshot at all. This is the Large World Model move — a reading taken from the scene as it stands, right now, with nothing stale in it.

But a rate-setting body that decided purely from the freshest available read of price indices, labour flows, credit aggregates and market expectations would be deciding from noise. High-frequency price data is volatile precisely because it has not yet been reconciled against anything. Weekly card-spend proxies for consumption move on payday timing. Market-implied inflation expectations swing on auction technicals that have nothing to do with the macro outlook. A committee that reacted to the freshest signal every time would produce policy as jumpy as the data feed, and would have no account of why last week's freshest signal is now being ignored. The live scene is true about what is present and mute about everything else — including the fact that what is present might be an artefact of measurement rather than a fact about the economy.

So neither posture is adequate on its own. The frozen minute asserts with false permanence. The live feed asserts with false confidence. Both fail the same way: neither carries a revision date, and neither tells you what would have to happen for the number to change.

The narrower claim

What actually survives contact with central banking is not "revise everything constantly" — the objection is right that this destroys accountability — nor "freeze and defend" — the ledger of real-time revisions shows that this destroys accuracy. What survives is dated belief, held apart from dated decision.

The decision can and should stay fixed. A rate rise announced in March is a fact of history; nothing revises it. What must not stay fixed, and currently does, is the economist's belief about the state of price indices, labour flows, credit aggregates and market expectations that the decision was reasoned from. That belief should be a snapshot: timestamped, sourced to a named data vintage, marked explicitly with what would supersede it — the next release date, the known revision schedule, the standing caveat that this labour survey undercounts self-employment by a documented margin. Fixity moves from the minute itself to the citation inside it. "Inflation stood at 4.1 per cent (ONS, vintage 14 March, subject to routine revision within ±0.3pp historically)" is a different kind of sentence from "inflation stood at 4.1 per cent." The decision is unmoved. The evidence behind it is dated rather than eternal.

The two objections that actually bite

The historians' correction matters here directly. Adrian Johns's point that fixity was manufactured, not intrinsic to the press, maps exactly onto central bank authority. A rate decision commands obedience not because the arithmetic is unimpeachable but because the institution — mandate, governor's signature, the ritual of the press conference — has spent decades building the expectation that this is the authoritative number. Strip the institutional scaffolding and a central bank forecast is exactly as contestable as any other model output built on revisable, often-wrong preliminary statistics. That does not weaken the case for dated belief; it strengthens it, because institutional authority is precisely what gets damaged when the underlying data are later shown to have been wrong and nobody can locate, in the record, what vintage was actually used.

The second objection is the sharper one for this domain: continuous revision destroys what fixity bought, namely a citable, auditable decision that cannot be relitigated after the fact. That must be conceded almost entirely. A rate-setting committee cannot be allowed to say, six months later, "we now believe, in light of revised data, that we would have acted differently" and treat that as costless. Markets priced against the original decision. Contracts were written against it. The answer is not to let the decision move — it is to make sure the belief that produced it was already dated when it was made, so that a later revision is legible as "the labour data used on the day has since been revised by the ONS from 4.2 to 3.9 per cent unemployment," which is a fact about data provenance, not an admission that the committee's reasoning was arbitrary. Version control did not abolish the idea of a release; it abolished the idea of a release with no record of what it was built from.

Where this leaves the ladder

None of this asks a central bank to abandon fixed decisions, and none of it promises a model that is never wrong. It asks for something narrower and available now, without waiting for the next generation of forecasting tools: that every number entering a policy decision carry its vintage, its source, and its known revision history, so the decision built on it can be defended on the day and understood correctly years later. That is the loose-leaf codex, not the printed edition and not the open page. It is also, on the intake axis, the last plausible move. Faster feeds, more streams, better reconciliation between price indices and labour flows — all of that is coverage and latency, improving the codex. None of it is a fourth way of taking in the world.