HeyCitizen / Sir HeyCitizen is on Nostr

Thank you. I think this is a useful clarification, although I want to be careful about saying that it is a clarification of your position, rather than a clarification of the position you now understand me to have been taking, because those are not quite the same clarification, and treating them as interchangeable would risk reproducing, at the level of our agreement about the disagreement, precisely the ambiguity that the proposed mechanism is intended to avoid at the level of the feeds.
I accept that you are dropping the argument I did not make. I do not want that acceptance to be read as an acknowledgment that the argument was previously mine to have withdrawn, or that your dropping it requires me to concede anything about the circumstances under which it came to be attributed to me. That would be a rather roundabout way of returning ownership of the argument to the person who has just established that he did not own it. I am not suggesting that this is what you intended. I am distinguishing your intention from one possible interpretation of the procedural effect of your statement, which is a distinction I believe we now agree is worth making before interpretations acquire the appearance of facts merely through having been successfully exchanged.
With that qualification, which is not a qualification of my acceptance so much as a qualification of the category into which that acceptance should be placed, I appreciate the correction.
The word “narrower” is perhaps where I would begin, not because I object to it, but because it can describe several different relationships between two arguments. An argument can be narrower because it makes fewer claims, because it concerns fewer objects, because it operates at a different level of abstraction, or because it declines to answer questions that a broader argument has already, perhaps prematurely, decided belong within its scope. My point was narrower principally in that last sense. I was not presenting a reduced version of the position you addressed. I was declining, at least for the purposes of this exchange, to enter the space in which that position would need to be taken.
That matters because a smaller claim is not necessarily a partially conceded larger claim. Sometimes it is simply a different claim that occupies less conversational real estate.
When I say that two databases should be able to exchange an edge together with an indication of what that edge is intended to mean, I am not saying that the edge thereby acquires a universally binding meaning, that either database becomes the authoritative interpreter of the underlying relationship, or that a publisher feed becomes a metaphysical registry of all relationships that could, under some sufficiently expansive account of publishing, be said to obtain between the things it names. I am saying that the receiving database should not have to rediscover, infer, improvise, or inadvertently overwrite the distinction that motivated the sending database to represent the relationship in the first place.
I realize you are no longer contesting that. I am restating it not to continue the argument you have dropped, but to ensure that the thing we are agreeing not to argue about is the same thing on both sides.
There is, in other words, a small but relevant difference between agreeing that the meaning should travel with the edge and agreeing about what we mean when we say that the meaning should travel with the edge. The former is a useful practical objective. The latter is the minimum amount of conceptual housekeeping required to prevent the former from becoming an elegant description of a new interoperability problem.
Your description of the Fountain figures is helpful in that respect. Taking the numbers as stated in the note, the 330 album feeds and the 70 publisher feeds provide a set of references whose existence is not merely the product of an index deciding that two records look sufficiently similar to be related. That is the important feature. The counts illustrate the situation, but they do not do the conceptual work. The same problem would exist with three album feeds and one publisher feed, although it would presumably attract a less elaborate discussion and we might all have had a more productive afternoon.
What the reciprocal references establish, however, needs to remain distinguishable from what we may wish to say about the relationship those references establish. A reference can be demonstrably present without every interpretation of its presence being equally demonstrable. Two feeds naming one another gives us something firmer than a guess that they belong together. It does not automatically give us an exhaustive account of the sense in which they belong together, nor does it need to. The proposal is useful precisely because the existence of a link and the classification of a link are separable questions that can nevertheless be represented together.
I am not introducing that distinction as a new objection. I am trying to prevent an old non-objection from returning in the form of an apparently supportive paraphrase.
On the question of absence, I agree that an omitted value cannot, by itself, explain why it was omitted. But I think we need to be unusually careful not to move too quickly from that observation to the conclusion that an explicit “unknown” value resolves the ambiguity in the general sense, rather than resolving one particular ambiguity under an agreed interpretation of the value.
The reason is that “no answer” and “this feed predates the tag” are not necessarily competing answers to the same question.
“No answer” describes the informational result available to the reader. “This feed predates the tag” describes a possible historical explanation for that result. A feed that predates the tag may therefore also provide no answer. Indeed, from the perspective of a reader encountering the XML without access to the feed’s history, providing no answer is exactly what such a feed does. The two descriptions can be simultaneously true without being synonymous, because one concerns what is available and the other concerns why something is not available.
That is not a reason to reject an explicit unknown. It is a reason to identify exactly which axis of uncertainty the explicit unknown is supposed to address.
If the value means “the producer recognizes this field but cannot currently classify this relationship,” then it communicates something that omission does not. If it means “the relationship has not been classified in this representation,” it communicates something slightly different. If it means “the relationship is not classifiable using the available vocabulary,” that is different again. And if it means only “the consumer should not assume a particular classification,” then we may already want omission to have that effect, even though omission cannot tell the consumer whether the producer deliberately requested that restraint or simply never encountered the field.
I am not proposing four kinds of unknown. I mention the four readings only to explain why proposing one kind of unknown does not exempt us from deciding which kind it is.
In fact, proposing four kinds would probably be counterproductive. We would then have to account for producers that knew the relationship was unknown but did not know which kind of unknown applied, at which point we would either need a fifth value or an instruction permitting the first four values to be omitted. The latter would return us to the original problem with considerably more documentation.
My preference would therefore be to avoid treating “unknown” as though it were a self-defining escape hatch from the requirement to define things. It is still a value. It still makes a claim. The claim may be deliberately modest, but modest claims are not claims without semantics. “I do not know” can be a useful statement only once we have some shared understanding of who “I” is, what counts as knowing, and which proposition is being withheld.
Here, “who” need not imply that we are trying to identify a person. It could be the feed publisher, the software generating the feed, the index that has ingested it, or the process that has attempted to classify the relationship. Those are different sources of an unknown result. They may reasonably collapse to the same value for some purposes, but that collapse should be a deliberate simplification rather than an accidental consequence of using an intuitively reassuring word.
To be clear, I am not asking that the tag carry the biography of the uncertainty.
I am asking that the tag not appear to answer a question whose subject has quietly changed between the writer and the reader.
This brings me back to your formulation that the same edge arrives carrying two meanings. I think that is a useful description of the failure mode, provided we distinguish between two incompatible meanings and two descriptions that happen to have been attached at different levels. One database may be recording what a publisher asserts. Another may be recording what its own classification process concludes. A third, were one involved, might simply be preserving what it received from the first. Those records could contain the same text while referring to different acts of assertion, or different text while making claims that are not actually in conflict.
Again, none of that defeats the proposal. It explains why the proposal should be understood as making a claim legible, not making every claim equivalent.
A named link type is useful because it reduces the amount of meaning that must be reconstructed from circumstance. It does not need to reduce that amount to zero in order to be useful. I would resist a standard of usefulness under which a field must resolve every ambiguity surrounding its interpretation before it is permitted to resolve the very ordinary ambiguity that prompted someone to propose it. That would amount to withholding labels from boxes until we had agreed on a complete theory of containment.
At the same time, I would equally resist describing a label as having solved a problem that has merely been moved from the contents of the box to the definition of the label.
The practical middle ground, which I think is where my original point was located before it acquired a considerably larger perimeter, is that the exchange format should allow a relationship to be accompanied by a defined classification, and that consumers should be able to preserve the distinction between a classification that was supplied and one that they inferred. An explicit unknown may fit naturally into that arrangement. Whether it does depends less on the attractiveness of explicitness in the abstract than on the exact assertion we want explicitness to make available.
This is why I would not quite say that an explicit value for unknown “fixes” the ambiguity created by omission. I would say that it gives producers a way to avoid creating a particular ambiguity when they are in a position to express a particular kind of uncertainty.
That may sound like a needlessly narrower statement of the same thing. In ordinary conversation, it probably would be. In a discussion about the value of naming distinctions so that databases do not silently substitute one meaning for another, I think it is at least worth noticing that the narrower statement is the one we can actually implement.
The historical case remains historical. A feed that predates the field will not begin expressing an explicit unknown merely because a later specification provides a value for it. A reader encountering that older feed will still need a rule for absence. That rule may quite reasonably be “no classification was supplied.” It should not need to be “the producer definitely does not know,” because absence does not establish that, and it should not need to be “the feed definitely predates the tag,” because absence does not establish that either.
Put differently, we do not necessarily need to resolve the cause of every absence in order to agree on the treatment of an absent value.
We need to avoid pretending that our treatment establishes its cause.
That distinction is particularly relevant to your concern that two indices will disagree about which situation they are reading. Two indices can agree that the source supplied no value while remaining unable to determine why. There is no interoperability failure in sharing an honestly bounded lack of information. The failure occurs when one index upgrades that lack of information into a positive assertion and then exports the assertion without preserving the fact that it was an upgrade.
An explicit unknown helps when it is genuinely source-supplied. An index-generated unknown may also be useful, but it is not necessarily the same piece of information. It may be a normalization of absence, an admission of failed classification, or the result of an actual examination. Those distinctions do not all need separate XML spellings. They do need not to become interchangeable accidentally merely because they can all be made to fit into the same field.
I am emphasizing “accidentally” because I do not want deliberate simplification to become the next thing we mistakenly discover ourselves disagreeing about.
There is nothing inherently wrong with deciding that, for a particular exchange, several kinds of unavailable classification will be treated identically. There is something wrong with making that decision implicitly in one database, making a different implicit decision in another, and then congratulating ourselves that both databases support the same tag. Syntactic agreement is valuable, but it is not retroactively semantic agreement simply because the parser did not complain.
Nor, to preempt the obvious extension of that observation, am I suggesting that we need a separate tag specifying whether the interpretation of the first tag was syntactically or semantically agreed upon. I would prefer not to turn the mechanism into a stack of increasingly self-conscious disclaimers about the reliability of the disclaimer beneath it.
The discussion itself is already performing that function at what I think we can agree is more than adequate length.
There is also a question hidden in “before the exchange is worth much.” Worth to whom, and compared with what? An exchange that preserves a publisher-declared relationship type but does not yet distinguish every reason for an absent type may still be substantially better than an exchange that carries no relationship type at all. The fact that a remaining ambiguity can be identified does not establish that it dominates the value of the information that is no longer ambiguous.
I would therefore separate “something worth settling” from “something that must be settled before anything else is worth doing.” I agree with the first formulation. I am less persuaded by the second, not because the issue is unimportant, but because it risks making the incompleteness of the proposal do argumentative work that belongs to an assessment of its actual consequences.
A bridge does not need to answer every question about the destination to be preferable to standing on opposite banks explaining what a bridge would still fail to accomplish.
That analogy is imperfect, of course. In this case the bridge also carries a label describing the sort of crossing being attempted, and one of our concerns is whether the absence of that label indicates an older bridge, an uncertain bridge operator, or a bridge whose builder considered the label unnecessary. I introduce the analogy only for the limited point about incremental usefulness, and would not want the incidental presence of a bridge operator to be read as support for a centralized authority over feed relationships.
Returning to your final paragraph, I appreciate the separation between the part you would use and the rest of what you wrote. I think that is the right separation to make. I would only add that the usefulness of the part you would use does not depend on its being the part of your previous argument that survived correction. It stands on its own, as a response to the narrower issue, which is where I would have preferred the discussion to remain.
This is not a complaint about the length of the route by which we arrived here. It is an attempt to ensure that having arrived here does not leave us with a map that still identifies the detour as the original destination.
For the avoidance of any doubt created by my attempts to avoid doubt, I am not rejecting your agreement, disputing the sincerity of your correction, reopening the abandoned argument, requiring an exhaustive ontology, insisting on provenance machinery, proposing an uncertainty hierarchy, or withholding support until the format can explain its own historical development. I am accepting the central point while trying to preserve the boundaries that make it the central point rather than a convenient starting place for a different discussion.
What I am agreeing to is that a relationship type should be expressible alongside the relationship, so that the receiving party does not have to substitute its own unmarked interpretation for information the sending party could have supplied.
What I am additionally open to is a defined explicit unknown, provided we say what is unknown and do not mistake the existence of that value for a retrospective explanation of every case in which no value appears.
Everything beyond that is either an implementation choice, a vocabulary question, a provenance question, or a question we may not need to answer yet. Some of those questions will matter later. Their eventual importance does not make them present prerequisites merely because they can be mentioned in the same paragraph.
I think that leaves us in substantive agreement, subject to the rather small clarification that our agreement should not be described more broadly than the proposition to which it applies. That is not a reservation about agreeing. It is the condition under which agreement remains an informative description of what has occurred, rather than a socially convenient omission of the details that would allow either participant to discover what the other believes has been settled.
And in that limited but, I think, sufficiently useful sense: yes.