Counting Without Reconstruction: How Reporting Infrastructures Shape Accountability in AI-Mediated Decision Systems — 4S 2026 Presentation

Presentation record for the talk Counting Without Reconstruction: How Reporting Infrastructures Shape Accountability in AI-Mediated Decision Systems, given online at the Society for Social Studies of Science (4S) Annual Meeting 2026, Toronto, in the open panel "Counting matters: The ontologies of (ac)counting with and beyond data", session 2 (Session 388), 9 October 2026. Chair: Anushree Gupta. Discussant: Janaki Srinivasan. Argument. A reporting schema decides, before any decision is made, which features of that decision will survive into the public record. Two public-record regimes are read at the level of one decision. In 46 high-stakes UK algorithmic transparency records (ATRS), the record describes the class of decision and never the instance. In the US CFPB consumer complaint database, a fixed response category turns a company's refusal to give a public account into something counted. The archive grows; the capacity to establish how a particular decision was produced does not. Files. 4S_Deck_v2_0.pdf (the 9 slides as shown); 4S_2026_Spoken_Text_as_prepared.pdf (the spoken text as prepared, from the slides' speaker notes, v3.2, 27 September 2026 — not a transcript); README_4S_2026_Presentation.pdf (this note, with the source of every figure). Sources of figures. UK: 10.5281/zenodo.20181288 (strict single-rater scoring, N=46 of 131 records captured 13 May 2026). US: 10.5281/zenodo.21873689 (working corpus 2,040,778 complaints with both narrative and company response; 1,876,833 = 91.98% select one of 11 fixed response categories). Not claimed. The UK audit is single-rater and preliminary, not validated. What a record does not preserve is not evidence the practice did not happen. Complaint figures are descriptive; no causal claim. Other panellists' work, discussant comments and audience discussion are not reproduced.