The claim
The design-rationale literature (ADR-030) is largely an account of why capture systems went unused. Grudin (LIT-759, unread; LIT-757, read) identified the disparity: the work of capture falls on one person and the benefit goes to another. Buckingham Shum and Hammond (LIT-761) and Lee (LIT-760) ask when capture repays its cost.
A collaborator with no memory pays the rediscovery tax every session (LU-ADR-001). When that same collaborator does the filing, capturing and benefiting fall to the same party, and the capture is cheap because the context is loaded at the moment of filing. Luria puts it this way: “Filed with context loaded it costs a paragraph; re-derived cold it costs a session” (LU-DP-008).
The disparity does not disappear. It moves. The owner now pays for reading and correcting the record, rather than for writing it. That is why the failure mode becomes flooding, and why the second-occurrence throttle (CLAIM-176) matters.
What it does not say
- It does not say agent filing is accurate. Several workbench entries exist to correct what an assistant said about the record.
- It does not say the owner’s review costs nothing. A week of operation is no evidence that the owner will keep reviewing.
Note of 2026-10-11: flooding as unused anchors
Under the owner’s reframing (CLAIM-185), the cost of capture is the cost of making things persist, and flooding is persistence without use: a code that no later entry points at. That makes over-capture measurable as codes with no inbound references, beyond those that are new. Whether an anchor goes unused because it was filed too eagerly, or because it is not discoverable (QUESTION-038), the count alone cannot say.
Note of 2026-10-11 (later): a size limit on settling
May (LIT-930, unread here) is the classic result that a large system with random interactions is almost surely stable only below a threshold, s√(nC) < 1, in its size n, connectance C and interaction strength s. Past it, more components or more links destabilize. Read across, flooding has a second cost beyond unused anchors. A record that keeps adding entries and relations could pass a point past which it no longer settles. The record’s structure is not random, and the debate after May is about which structures escape the bound, so this is a question to test, not a result to apply. It is also a further reason for the second-occurrence throttle (CLAIM-176).
Added 2026-10-14 (NOTE-851): Ashby (LIT-787, S. 20/10) is an earlier, empirical version of May’s result. Of random linear systems sampled up to 4 × 4, the fraction stable fell roughly as 1/2ⁿ, and “large linear systems should be assumed to be unstable”. He also names the escape: weak joining and many constancies. May (LIT-930) is now Unavailable, so the bound is still cited unread.
Note of 2026-10-14: what the founding paper and Hopfield’s model say
- IBIS reports adoption, not abandonment (LIT-758, read in NOTE-811). The founding paper reports “eager acceptance by the users” and gives no account of what keeping an IBIS costs. Its operating cycle has staff edit issues “in contact with” the participant who raised them, so filer and arguer were already separate in the original design, as they are with an agent filer. That suggests the disparity Grudin describes may depend on who the filer is. It is one sentence of design, not evidence.
- A capacity limit cuts the other way (LIT-918, read in NOTE-819). In Hopfield’s model spurious states occur at low load too, and overload is total loss: it “makes all memory states irretrievable unless there is a provision for forgetting old memories”. Read across, a record past its load would lose access to its old entries rather than blur them. The record’s rule, retire by status and never delete, is a provision against forgetting. This is analogy only.
Note of 2026-10-14: what Grudin’s read article adds
Grudin’s 1994 article (LIT-757, read in NOTE-835; THEORY-311) states the disparity as the first of eight challenges, so the premise can rest on a read paper and not only on the unread 1988 one (LIT-759). It qualifies the claim twice.
- Removing the disparity is not enough. “Even a net benefit with equal costs and benefits for all users will not guarantee groupware success” (pp. 96–97): critical mass and freeloading, among others, can sink a system the disparity does not. The defeater above names only the disparity, so a record abandoned for another of Grudin’s reasons would not defeat this claim, and would not be explained by it either.
- The one issue-based system he cites failed for another reason. It was abandoned before use because “the explicit record of opposing positions that the application would immortalize was politically unacceptable” (p. 97). That is challenge 3, social disruption, not challenge 1. It bears on a record that retires by status and never deletes.
What Lee’s seven issues are (LIT-760) remains unknown here: the paper is Unavailable.
Note of 2026-10-30: what holds it back
Its premise, Grudin’s disparity account (THEORY-311), is Proposed: no study in the record measures, member by member, the extra work asked and the benefit received across groupware. The record’s own reading of LIT-757 finds the one issue-based system Grudin cites abandoned for a political reason, not the disparity, and LIT-758 reports IBIS adopted. The title says the disparity is removed, while the body says it moves to the owner’s reviewing, which is the mechanism the defeat condition names. Nothing beyond the record’s own five weeks says whether that sinks agent-filed records.