Surplus Reservoirs
Where the apparatus stores what it has extracted.
A photograph uploaded to a free social-media service in 2014 was, in 2024, used to train a generative-image model that was then sold as a commercial product. The photographer was not asked. The photographer was not compensated. The photographer was, technically, never the customer; the photographer was the input.
A Surplus Reservoir is the data and capital aggregation system in which the apparatus stores what it has extracted. The Reservoir's defining feature is its scope: every input the apparatus has captured — every utterance, every gesture, every transaction, every biometric reading, every preference — is preserved indefinitely and remains available for downstream use. The Reservoir is the apparatus's archive of the occupant.
What the Reservoir physically is, today, is a cloud data warehouse — petabyte-scale storage, geographically redundant, with differential-privacy nominally invoked at the API layer and operationally compromised wherever a brand partner has paid for direct access. The headline access controls are theater. What the warehouse contains comes in three layers. The first is data the occupant produced consciously (posts, reviews, search queries). The second is data the occupant produced unconsciously (browsing patterns, dwell times, location traces, the gait their phone's accelerometer reads while they walk). The third — the one the contemporary regulators are still pretending isn't there — is data the occupant never produced at all, inferred from the first two: sexual orientation, political affiliation, mental-health trajectory, household composition, susceptibility to specific marketing rhetorics. If you want to know what the Reservoir actually is, look at the inference layer. The Reservoir is, primarily, a model-training environment with storage attached.
The Marxian frame, transposed: where industrial-era capital aggregated produced goods (the warehouse), platform-era capital aggregates the data produced as a byproduct of consumer activity (the data warehouse). The warehouse extracted surplus labor; the data warehouse extracts surplus attention, affect, and behavior. The transposition's most significant feature is that the data warehouse continues to extract value from the input after the input has ceased — a property the industrial warehouse never had. Shoshana Zuboff's account of behavioral surplus provides the contemporary political-economic name for what the Reservoir captures. The argument that the Reservoir is the principal site at which contemporary value is generated has, on my reading, been validated by every subsequent year of platform-economic data.
Inside the OLS, the Reservoir takes the form of an onboard service ledger that records, against the occupant's identifier, every biometric reading the Translator captures, every productivity output the Feedback subsystem registers, every content engagement, every pharmacological dosing event, every affective-state classification. The retention period is tier-graded — seven years post-cessation for Essential, twenty-five for Pro, ninety-nine fiscal years for Eternal — and the ledger continues to be queried, by the manufacturer's brand partners, for the duration of that period. Those brand partners purchase access on a per-occupant, per-data-class basis. The occupant is not told which partners are buying their record. The occupant is not told that the record continues to be sold after their death.
The Reservoir is the most patient subsystem in the apparatus. Looked at on the timescale of a single interaction, what it captures is modest. Looked at on the timescale of an occupant's lifetime, it accumulates. Looked at on the timescale of the apparatus as an institution — the timescale on which decisions about which platforms to acquire and which models to train get made — the Reservoir is where the apparatus's actual capital sits. Everything the occupant experiences as the apparatus is, in this longer view, the surface through which the Reservoir gets fed.