Research opportunity · Discretion curtailed, relocated or encoded?
Who holds decision authority in public-sector algorithm contracts, agency or vendor?
Earlier version (v1)
This is RO-LD05-05 as it was published at v1, kept unchanged. The current version is v2; its changes are listed there.
Question
In the contracts public agencies sign for risk-scoring, profiling and case-routing systems, who is given authority over the choices that shape decisions (thresholds, input variables, retraining and updates, performance targets), what audit and documentation rights does the agency keep, and does this differ between systems built in-house and those bought from vendors?
Institutional practice unknownOne PhD chapter or paperLow costPartly addressed
Why it matters
If consequential judgement relocates into system design, the question of who designs matters. Legal scholars have argued that procurement hands policy choices to vendors, and interview studies show procurement staff struggling to govern AI purchases, but no one has systematically read the contracts to see where authority over thresholds, updates and documentation is actually placed. That is a researchable, document-based question.
The debate lists this among what we still do not know: How much substantive discretion sits with vendors and procurement contracts rather than public agencies?
Evidence that motivates it
- CNAF’s DMDE 2026 is a targeting model rather than an automatic final decision-maker; its source code and model documentation are public and the redesign excludes several sensitive/profile variables.
CNAF publishes the source code and documentation of its targeting model and excluded several sensitive variables in a redesign, showing that design choices can be documented and publicly held; it is a contrast case, and the claim does not say the model is fair or lawful. - For AI-mediated administrative processes, effective contestability depends partly on process provenance—knowing whether a system selected the case, influenced routing, or generated part of the evidentiary record.
Contestability depends on knowing whether and how a system selected or routed a case; contract terms on documentation and audit rights determine whether the agency can even supply that information.
A possible design
Assemble a corpus of contracts, statements of work and tender documents for public-sector decision-support systems (benefits fraud, employment profiling, child welfare, immigration routing, policing) from open procurement portals (UK Find a Tender and Contracts Finder, EU Tenders Electronic Daily, US state and city portals) and FOI requests for unpublished schedules. Target 60 to 100 contracts across at least three countries. Develop and pilot a codebook: who sets and may change thresholds; who chooses input variables; who may retrain or update the model and with what notice; performance and error targets; agency rights to inspect code, data and documentation; audit and bias-testing obligations; trade-secret and confidentiality clauses; liability; exit and data-return terms. Double-code a subsample for reliability. Compare vendor-built with in-house or publicly documented systems, and follow up with a few interviews with procurement officers to check how terms work in practice.
Data needed
Public procurement records and FOI responses (redactions expected); a codebook; a second coder; optional interviews with procurement staff.
Feasibility
One PhD chapter or paper. Low: documents are free or obtainable by FOI; costs are mainly coder time. Contracts are often redacted or withheld on commercial-confidentiality grounds; the pattern of redaction should be coded as data. Contracts show formal allocation of authority, not how it is exercised, so the interviews matter.
Has this been done?
Partly done already: related work covers some of this question (first-pass screen of 12 October 2026; not yet checked in depth).
The argument that procurement shifts policy choices to vendors is established in legal scholarship, and records-request and interview studies exist, including a CHI 2026 interview study of procurement as an AI governance mechanism. No study was found that systematically codes contract terms to show where authority over thresholds, variables, updates and documentation sits for public-sector decision systems.
How strong this evidence is
- Who searched and checked: an independent AI reviewer with fresh context (not a person), 12 October 2026: every nearest-work citation checked against its source page; corrections applied. Every check so far was done by AI; no person or domain expert has checked it yet.
- What was read: 1 of the 4 nearest works below was judged from an abstract, summary or search listing only, and 1 has details that could not be confirmed; it is marked. Not every entry says how much of the work was read.
- Where the search could not reach: Screened by web search on the date shown; the CHI 2026 paper by Hudig, Kallina and Singh was checked only through Crossref metadata and the first author's summary page (the full text was not read), which describe an interview study, not contract coding. Google Scholar, SSRN, paywalled law reviews and non-English procurement scholarship were not systematically searched. Absence of a hit is not proof that the work does not exist.
- Before building a project on it, run your own search in a scholarly database such as Scopus, Web of Science, Google Scholar or SSRN.
Nearest work found
- Mulligan & Bamberger, "Procurement as Policy - Administrative Process for Machine Learning," Berkeley Technology Law Journal 34, 773 (2019)
Argues that procured machine-learning systems embed policy choices (targets, training data, thresholds) made largely by vendors outside administrative process. Normative and case-based; no coded dataset of contracts (only part of the article was read). - Johnson, Silva, Leon, Eslami, Schwanke, Dotan & Heidari, "Legacy Procurement Practices Shape How U.S. Cities Govern AI - Understanding Government Employees' Practices, Challenges, and Needs," Proceedings of the 2025 ACM Conference on Fairness, Accountability, and Transparency (FAccT 2025), 772–789; arXiv:2411.04994
Interviews with 19 city employees in 7 US cities on how procurement shapes who holds decision power over AI; does not analyse contract texts. - Brauneis & Goodman, "Algorithmic Transparency for the Smart City," Yale Journal of Law & Technology 20, 103 (2018)
42 open-records requests in 23 states about six predictive programs; finds transparency almost never provided and over-broad trade-secret claims. Studies what agencies disclose, not how contracts allocate decision authority. - Hudig, Kallina & Singh, "'It's Just a Wild, Wild West': Harnessing Public Procurement as an AI Governance Mechanism," Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems (CHI 2026), 1–22, doi:10.1145/3772318.3791968 Judged from an abstract, summary or search listing onlySome details not confirmed
Semi-structured interviews with public-sector buyers, suppliers, consultants and procurement specialists in the EU and UK on how procurement can govern AI, covering pilots that stay below procurement thresholds, post-deployment model changes and audits. Read via the first author's summary page; the ACM page could not be opened. The summary describes no systematic coding of contract texts.
Search queries used
content analysis public sector AI procurement contracts vendor control model thresholds decision authority algorithmic systems study"procurement" algorithmic decision systems government contracts analysis vendors discretion accountability empirical FOI contracts codedMulligan Bamberger "Procurement as Policy" administrative process machine learningstudy coding contract clauses AI vendors government model updates intellectual property audit rights data ownership sample of public contracts
Provenance
RO-LD05-05 v1. Drafted by Claude (SocioLens Project Lead), AI-drafted from the LD-05 debate on 2026-10-12. Checked by an independent AI reviewer with fresh context (not a person), 12 October 2026: every nearest-work citation checked against its source page; corrections applied. Not yet reviewed by a domain expert. This is a research hypothesis, not a finding.