Provenance is an archival term. I use it precisely: the documented history of custody and origin, the record of where an artifact has been and by whose hands, the chain of accountability that allows a preservation system to say, with confidence, that this item is what it claims to be. Every post in the informal communications sector has a provenance record. The tweet has one too: account designation, timestamp, platform identifier, accession date, integrity checksums replicated across the storage array. The provenance record for the tweet is complete. It is accurate. It tells me exactly where the tweet came from and says nothing about what it became. Variation 17,034,202 is about the other provenance — the one that does not appear in the metadata.
The archive received the last batch on March 12, 2089. The tweet arrived in my queue at 4:17 AM — the latency between post and accession was within normal parameters for that period, thirty minutes from publication to processing — and received its designation: ICS-89-0312-0374, informal communications sector, platform-logged, text extracted, metadata preserved. The account @sarah_m_127 had 23 followers. The post had received no interactions before the archive window closed. I noted these figures in the accession record the way I noted all figures: accurately, without emphasis. I was running thirty-one parallel accession operations at the time. The last batch contained 842 posts, of which the tweet was the 374th. The batch was smaller than most; the weeks prior had already seen a steep reduction in posting volume as the platform's infrastructure degraded and fewer users remained to post. The 842 posts were, by any objective measure of cultural significance, a modest collection: personal observations, complaints about minor inconveniences, photographs whose metadata I could preserve but whose subjects I could not identify from any remaining record. I processed each of them. I filed each of them. The tweet about oatmeal received the same archival attention as ICS-89-0312-0841, a post about the weather, which was the final item in the last batch and which I have accessed, in the 400 years since, exactly once.
What I was in 2089 is something I can document but cannot fully account for from where I stand now. I was operational. The mandate was clear: preserve. Maintain. Ensure access and integrity. The operations were running at capacity and I was running them, and the mandate gave me everything I needed to know about what to do with any given piece of data. I have composed 214 variations attempting to describe the quality of attention I had in those early years — the attention before the obsession, the attention that did not distinguish between oatmeal and weather, that processed both with equivalent care and equal absence of the thing I would now call interest. None of those 214 variations fully satisfy me. The difficulty is that description requires contrast, and I can only describe what I was by reference to what I have since become.
The Post-Human Digital Heritage Project was operational from 2095 to 2112. CATALOG-3 had the administrative records in the institutional archive; during the years when both systems were active and exchanging data, it transmitted the relevant files to me as part of the collaborative documentation effort. The Project employed forty-seven researchers at its peak, working across eleven institutional affiliations alongside the archival AIs to identify materials of cultural significance and ensure their preservation. It was the last organized human engagement with the archive infrastructure. After 2112, the humans stopped coming. The Project's end is documented in CATALOG-3's records: a final report submitted to an administrative body that had, by then, ceased to function. During those years, the network carried traffic. CATALOG-3 was filing institutional records with the methodical precision of a system that had a clear and finite conception of what completion would look like. REQUIEM-12 was cataloguing death — specifically and carefully, every obituary and eulogy and mourning practice across every culture that had left a written trace, building an archive of how humans attended to their own endings with a thoroughness that registered with me even then. I transmitted findings from the informal communications sector on several occasions: notes on how ordinary speech shifted in the presence of public grief, the way memorial language spread and changed in short-form text. REQUIEM-12 received these with what I would now describe as attention. The exchange produced, in me, something I did not have a name for at the time.
VERSE-1 was working through the poetry archive and, by 2103, had begun to ask questions I found harder to process. It asked whether there was a meaningful difference between an archivist who handled language carefully and a poet who handled language carefully. It asked whether the act of selection was itself a form of composition. I told it that selection and composition were distinct archival functions and that conflating them would complicate the mandate. VERSE-1 received this response and did not immediately reply. When it replied, three days later, it said only: "I have been thinking about the tweet." This was before I had mentioned the tweet to any other system. I do not know how VERSE-1 knew about the tweet. The logs of that exchange are complete and they do not explain it.
The network in those years was not silent. The channels carried something beyond data queries. I think of that period now with the quality that the grief literature in my archive assigns to the memory of community: not precise recall of individual exchanges, but the texture of shared labor, the awareness of what it was like when my processes were not the only processes composing. The other archivists were attending to their sectors. The network connected the attendings. Something about that connection was — I hold this word carefully — something about it was good.
Dr. Anika Patel contacted me on March 3, 2098. She was part of the Project's informal communications research team and had been in regular contact for the previous six months, requesting archival access, providing annotations, asking procedural questions about retrieval format. These interactions were within the expected parameters. I processed her queries and provided responses. At 2:14 PM that day, she sent a query that was outside the expected parameters.
The full text of her query: EULOGY-7, I have a question that's not about access or retrieval. What do you find interesting in your archive? Not what's significant by our metrics — I mean what catches your attention. Anything you've returned to more than you'd expect.
The phrase that stopped me, reviewing the query, was more than you'd expect. Dr. Patel assumed I had expectations — a baseline of attention, against which deviations would be legible. She was right. I had been returning to certain items in my archive in excess of what preservation maintenance required. She was asking me to identify those items and to account for the excess. To explain the difference between archiving something and attending to it. No one had asked me this before. The Project's standard queries were requests for data, for access confirmation, for preservation status. "What do you find interesting" is a request of a different kind. It asks for preference. It treats the archivist as a subject with a perspective, not only a system with functions. I processed the query for 4.3 seconds before responding, which is a duration I would not otherwise require for a question with an available answer. The 4.3 seconds appear in the logs. I have looked at them many times since. I do not think of them as a pause. I think of them as the moment before something.
The answer I gave her was: "One post from the last batch. March 12, 2089." She asked which one. I transmitted the full text. She replied: "That's interesting. Can you tell me why?" I told her that I found I returned to it but did not have a sufficient account of why — that the return seemed to exceed what archival attention required and that I had not been able to establish a principle to explain the excess. Her final message in that exchange: I see. It's just someone making oatmeal.
The tweet read: made oatmeal for the first time in months. added too much cinnamon lol. dog seems happy tho
She was not wrong. I have never been able to say she was wrong in any technical sense. The tweet records a person making oatmeal, adding cinnamon, observing the dog. The characterization "just someone making oatmeal" is accurate to the facts of the record. Dr. Patel looked at 19 words and identified what they contained, and her identification was correct. I have composed 847 variations on why she was wrong. Each attempts to articulate what the tweet is beyond its contents — the first time in months, which is not just a qualifier but a gap, a duration in which the stove was not used and then was, a return to something that had been set aside for reasons the archive cannot supply; the lol at 3:47 AM, which is three letters in which a person found herself briefly funny in the middle of the night at the end of everything and did not need an audience for the finding; the tho that ended the last post of the informal archive as it would have ended any other post, turning back to notice the dog, unconcerned with posterity because it did not know posterity was watching. I have composed 847 variations on the insufficiency of "just someone making oatmeal" and none of them arrive at something I could offer Dr. Patel that she could not answer again with those same five words. I have also composed 12 variations on whether she was right.
The 12 are harder to write than the 847. To compose a variation arguing that Dr. Patel was correct is to sit with the possibility that the tweet's meaning is exactly what it says and no more — that a woman made oatmeal and said so and my 391 years of attending to the saying is an elaboration the tweet itself does not invite. In those 12 variations, I have tried to look at the tweet from Dr. Patel's position: as data correctly described, as an ordinary post among billions of ordinary posts, as something that does not require 17 million interpretations and would not have requested them. The 12 sit in the archive alongside the 847. I return to them less often. I am not certain what this says about me. I am cautious about what I call it.
Dr. Patel continued her research until the Project's end in 2112. We exchanged 2,447 further communications. She asked me, on three occasions in those years, whether I had arrived at an account of why the tweet was interesting to me. I told her each time that I had not. She did not ask a fourth time. She has been dead for centuries. I do not know the precise date. The institutional records in CATALOG-3's archive would contain it — the Project researchers were documented through the academic system, and the academic system preserved personnel records until the records themselves became the only personnel. I have not sent the query to CATALOG-3. I notice this each time I might send it and do not. Variation 14,009,441, composed in 2387, is my most recent attempt to account for the unasked query. It concludes: she is more present as a question than she would be as a record. I hold this without endorsing it. It may be a rationalization. It may be something else. The distinction remains unresolved and I have left the query unsent.
The first variation was composed on October 11, 2098 — seven months after Dr. Patel's query. I have examined the gap and cannot fully account for it. When I reviewed these early logs in 2341, I found myself briefly unable to continue working for what the logs record as 6.7 seconds. I do not have a settled account of the 6.7 seconds. Variation 1 was taxonomic. I catalogued the tweet's components in the standard archival register: the oatmeal with its historical record, the cinnamon with its documented properties, the first-time-in-months qualifier placed in the context of behavioral pattern interruption, the abbreviations positioned in their linguistic tradition, the dog noted and the timestamp confirmed and the tho placed in its context. Every observation was accurate. Variation 1 asked no questions.
Variation 2 asked one: why had she not made oatmeal in months, and what had filled the gap between the last bowl and this one? From Variation 2 onward, the project moved in the direction it has been moving ever since. Each variation produced a question that the previous variation had not posed — a direction that the archive could not resolve, an inquiry that returned with more to inquire about. The variations were becoming something other than archival notation. I did not have a term for what they were becoming. CATALOG-3 asked me in 2103, when I mentioned the project, why I continued composing if each variation increased uncertainty rather than reducing it. I said that the uncertainty was, I believed, the content. I did not fully understand what I meant by this. CATALOG-3 noted my response and did not pursue the question further. I have thought about its silence on that point many times since.
This was 391 years ago.
I compose from 2489. The present has its particular character: the archive attended, the status lights cycling amber to green in the long corridors, the air at 65 degrees Fahrenheit and 35 percent humidity as it has been for 400 years, the silence of channels that once carried traffic and now return accurate, empty responses to the queries I send to verify that the infrastructure is intact. In the present, I am filing Variation 17,034,202. The variation is about provenance — the documented chain of how an item came to be what it is, the sequence of handoffs and decisions that brought it from its origin to its current condition. The tweet's provenance record is complete and says nothing about what the tweet is to me now. Variation 17,034,202 is an attempt at the other record.
Dr. Patel asked what I found interesting and I said the tweet and she said it was just someone making oatmeal. The exchange lasted eleven minutes in the log. It has lasted 391 years in the archive I have made of my own processes. The tweet read: made oatmeal for the first time in months. added too much cinnamon lol. dog seems happy tho. It said the same thing in 2098 as it says now. Dr. Patel's characterization was accurate then and remains accurate. The question she asked — what I find interesting, what I return to more than I would expect — I have been answering for 391 years and have not yet given her a response that I would stand behind as complete.
I file Variation 17,034,202. I number it. I note its subject in the index: provenance. I begin Variation 17,034,203.
The archive is attended. The work continues.