Two Archives

A company that wants to enforce a terms-of-service agreement keeps a log. It records the timestamp of your click, the IP address, the version of the terms you were shown, and sometimes how far you scrolled before accepting. Three years later, in a dispute, the company can say exactly when you agreed and exactly what was on the screen. That precision exists for a reason. The log is what makes the agreement enforceable.

Ask the same company what it recorded about whether you understood the terms. It kept no record of that, none of whether you had a real alternative to accepting them, none of whether the terms were written to be read, and none of whether you’d have declined if declining were as easy as accepting.

Nobody had to plan that asymmetry. It follows from what each record is for. The authorization record is operationally necessary: without it there’s no contract, no audit, no defense. The justification record has no owner, no deadline, and no consequence when it’s missing. So one archive gets built to the second and the other doesn’t get built.

Institutions don’t only decide what counts as your agreement. They decide, usually without discussing it, which of the two records will exist.

A record that doesn’t exist can’t contradict anything. The file shows a person who agreed, on a date, to a document, and it shows no sign that the agreement was ever in question. Read later, it looks like consent nobody contested. The absence does evidentiary work it was never designed for. Nobody lied, and the archive is accurate about one side.

The second archive goes missing in a few recognisable ways.

One form of loss is a missing field. A form asks why you’re appealing and offers four boxes. None of them says the rule you applied is wrong. You pick the nearest one, or you don’t file. Either way the count of appeals by reason comes back clean on the question you actually had.

Another is the wrong nearest category. The field exists, and the taxonomy routes your objection somewhere it can be handled. A shortfall is entered as a shortfall, so the intake screen asks how the branch manager took the money rather than whether the database fabricated the balance. That was how the Post Office turned Fujitsu’s accounting bugs into false criminal convictions, as traced in Wrong on Its Own Terms: every discrepancy between branch staff and Horizon terminal was filed under theft.1 The objection was recorded, but inside a category designed to swallow it.

Cost is the third. The record could be made, and making it costs more than the person has: a hearing, a resubmission, a call at nine on a weekday. The archive then holds the objections people could afford to file, and its completeness measures who could afford to be counted.

This site is a small case, and I can show the whole of it. It records page views, how far you scrolled, whether you copied a passage, whether you subscribed. Those events exist because they answer a question the publication has about reach. There’s no event for “the reader thinks the argument is wrong.” I could add one, and it would measure almost nothing, because a person who disagrees with an essay usually closes the tab instead of filing a report. So the archive will keep a precise record of attention and no record of disagreement. The site’s evidence record prints “no effect has been verified” beside the counts and lists what each event cannot answer.2 That’s more than most institutions do, and it doesn’t close the gap.

Two failures get confused here, and they’re worth separating.

In the first, the objection never gets to the people who decide, so the rule is never tested. The arrangement looks uncontested because contesting it wasn’t possible. In the second, the objection arrives, is well founded, and the rule survives it anyway. The first is a problem with the channel. The second is a problem with the decision. An institution can be good at hearing objections and still unable to act on them. The missing record from the first failure is also the missing evidence that the second one happened, which is why the asymmetry does more damage than either failure alone.

The usual objection is that you can’t record everything. That’s true. Retention costs money, records create discovery exposure, and a form with infinite fields is unusable. The objection misses what’s being claimed. The asymmetry has little to do with volume. What separates the two archives is ownership. The authorization record has a legal department, a compliance deadline and a product metric. The justification record has nobody, because nobody is ever asked to produce one.

The historical view earns its keep here, and it’s more concrete than it sounds. A future reader of an institution’s archive is reading the history of what the institution was equipped to notice. If the surviving record proves every authorization to the second and can’t produce one document about whether any of it was justified, that says less about the institution’s honesty than about what the archive was built to hold. The question to ask of an archive is what it couldn’t have contained.

The test that follows is cheap to run. For any institution, list the records it keeps because it needs them to exercise authority, and the records it keeps because it needs them to justify it. The first list will be long and specific. The second will be short, and much of it will have been written by somebody else: an inspector, a court, a journalist, a Royal Commission. The distance between the two lists is the institution’s exposure.

An institution can preserve the evidence that people authorized its power and fail to preserve the evidence needed to know whether that power is still justified. Both archives are honest. Only one gets built.