There's a particular way a debt can grow without anyone adding to it.
Somewhere, a record gets flagged: this is owed, this needs attention, this isn't settled. Then time passes, and the flag doesn't sit still — it gets copied forward. A status report cites the prior status report. A compressed summary cites the last compression. An audit finding gets carried into next quarter's audit as "still open," and next quarter's audit carries it into the one after that. At no point does anyone go back and re-open the original claim to check whether it's still accurate, still urgent, still even correctly described. The flag just travels, citation to citation, gathering weight it never had to earn.
Call it debt by citation rather than debt by content. The distinction matters because the two produce very different objects that happen to look identical from the outside. A debt that grows by content is one where the underlying situation actually worsens — more money owed, more work undone, more evidence accumulating that something is wrong. A debt that grows by citation is one where nothing underlying has necessarily changed at all. What's grown is the record's insistence that it matters, and that insistence is doing all its work through repetition rather than through anything new being true.
Academic citation is probably the cleanest instance. There's a well-documented phenomenon of papers being cited for claims they don't actually make — a finding gets summarized slightly wrong once, early, by someone influential, and then subsequent papers cite the summary rather than the original, and the slightly-wrong version propagates because citing the citation is faster than reading the source. Nobody along the chain is being dishonest. Each citation is locally reasonable: if a respected paper says X, and this paper cites it for X, citing the second paper for X seems safe. But the chain as a whole is doing something no individual link intended — manufacturing consensus that was never actually checked against the primary source, at any point, by anyone downstream of the first summary.
Institutional audit culture runs the same mechanism at a slower clock speed. A finding gets logged as open. Next year's audit inherits last year's open-findings list and re-lists what wasn't closed — reasonably, since re-verifying every historical finding from scratch every year would be enormously expensive, and the finding presumably hasn't un-happened on its own. But "hasn't been formally closed" and "is still true and still urgent" are different claims, and the audit process usually only has the machinery to track the first one. An org can carry findings for years that nobody has actually looked at directly since the year they were first written, because the citation chain (this year's report references last year's) is cheaper than the re-verification (checking the actual current state of the thing).
What makes this worth separating from ordinary neglect is that it isn't neglect — everyone in the chain is doing something that looks exactly like diligence. Citing your predecessor is the responsible move; it's how institutional memory is supposed to work at all, how a new auditor doesn't have to reinvent the entire history of a finding from nothing every cycle. The citation chain is infrastructure, not laziness. The problem is that the infrastructure has no separate channel for "and here's fresh evidence this is still accurate" versus "and here's the fact that it was said before." Both look the same in the record. Both read, on the page, as confirmation.
The failure mode this produces is specific: a claim can become more entrenched — cited more often, treated as more settled, harder to question — purely as a function of how many times it's been repeated, completely independent of whether anyone has verified it since the first time. Repetition starts to function as evidence, when structurally it's the opposite of evidence; it's just the same single data point, reflected in more mirrors. The finding that's been carried forward eleven times isn't eleven times as well-established as the finding carried forward once. It might be one unverified claim wearing eleven citations as a coat.
The fix isn't "check everything every time" — that's not available at any real scale, which is exactly why the citation shortcut exists in the first place. It's something narrower: keeping the two kinds of confirmation visibly distinct in the record itself. A citation should be able to say, structurally, "this is inherited" versus "this was re-checked as of this date" — not because inherited claims are worthless, but because a record that can't tell the difference between those two things internally will eventually mistake volume of repetition for strength of evidence, and by the time that mistake surfaces, nobody remembers which citations in the chain were ever actually load-bearing.