01 / Necessary informational conditions
From notice to
contestable grounds.
Notice, explanation, and an appeal procedure can leave an affected person unable to investigate the determination they are asked to accept.
The first NIC paper, published as a preprint in October 2026, identifies effective incontestability as a distinct failure mode: a determination that remains formally open to challenge yet stands practically beyond it, because the affected party cannot investigate the grounds for an objection or obtain a reasoned response. Open access to a model can show how it computes a result while leaving the person unable to check the data used in their case or the standard the institution applied.
The framework connects a material objection to the epistemic task it sets, the informational operations that could perform that task, and a test of conditional necessity: an operation is necessary only if no feasible, adequate route can dispense with it. Six functions organize the inquiry: determination and consequence; sociotechnical roles and evidence control; data, source, and quality; governing-standard–operation fit; evidential warrant, uncertainty, and threshold; and contextual applicability.
A five-part delivery architecture (baseline orientation, contemporaneous preservation, inquiry-activated access, protected mediation, and usable support) lets affected parties discover what warrants investigation before they are expected to substantiate an objection. Meaningful contestability also requires institutional uptake: a forum that answers with reasons and can secure correction or warranted reaffirmation. K.W. v. Armstrong, the Idaho Medicaid budget-tool litigation, demonstrates how informational and procedural obstacles interact to produce effective incontestability.
Read the preprint (DOI 10.5281/zenodo.23387270)02 / Institutions and private power
Why the obligation
exists at all.
The NIC paper specifies what meaningful contestability requires. The companion paper, Private Power without Epistemic Preemption: Relational Institutional Answerability in Algorithmic Decision-Making, argues why private institutions owe it.
Its thesis is that institutions exercising consequential private power over people through computational systems must not convert epistemic advantage into epistemic preemption. That exercise of power can itself generate a duty to keep determinations open to informed challenge and consequential revision. Keeping the two questions, what is owed and what it requires, in separate papers makes each argument more exact.
The preprint is expected in January 2027.
Read the companion paper’s public outline03 / Making the register empirically vulnerable
Claims built
to be tested.
The research program tests three claims: that the six-function register covers material objections, that its distinctions improve investigation, and that particular operations are indispensable for specified tasks under specified conditions.
Planned work codes a corpus of successful and failed contestation episodes against preregistered rival partitions, including a merger of three functions into a single evidence-adequacy category. Omission and substitution experiments use prespecified adequacy floors and stopping rules. Field studies measure institutional uptake separately from informational access. A linked deference study asks whether reliance on system outputs stays contributory or becomes preemptive, and distinguishes justified agreement with automated advice from displaced judgment.
Development is planned to begin with consumer-credit cases, followed by public benefits and health care, with portability tested in held-out settings. These studies are proposed; data collection and results are ahead.
Read the proposed study plan04 / Approach to inquiry
Make the claim precise.
Specify how it could fail.
My approach moves between conceptual analysis, institutional case studies, and empirical design. Necessity claims are bounded to stated conditions; tasks and adequacy floors are justified independently of the operations being tested; and an adequate route that omits an operation defeats its claimed necessity, however the categories are drawn.
The method reflects long practice in knowledge engineering: eliciting tacit expert judgment, reading regulatory and underwriting standards closely, implementing rule-based systems, and building the trace-and-explanation facilities that let users, auditors, and regulators see which rules produced a determination. I have built the passage from governing standard to implemented rule myself, and I know where it can diverge.
Planned work includes preregistration, documented materials, reproducible analyses, and full reporting of null results, burdens, and unequal effects.
05 / Beyond the paper
Knowledge people
can work with.
The framework is meant for use by affected people and their representatives, by institutions designing review procedures, and by researchers comparing failures and remedies across domains.
- A contestability toolkit. Connected affected-party and reviewer interfaces, a case generator, a record linking grounds, evidence, and disposition, scoring tools, and an adaptation guide, with domain modules for credit, public benefits, and health care.
- Tested portability. Independent users adapting a frozen version of the toolkit to held-out settings, so that transfer claims rest on evidence.
- Public understanding. Clear explanations and worked examples that preserve the distinctions people need to evaluate a determination.
- Research education. Teaching modules in experimental design, qualitative coding, and open science, for students with practical experience who want a route into inquiry.
- Open materials. Materials and negative findings shared where privacy permits.
The preprint, public outlines, and worked example are available now. The toolkit, studies, and teaching modules are planned contributions as the research proceeds.
Visit the research program06 / Publications and working papers
Published and
in preparation.
- Thornhill, Kerry. 2026. “From Notice to Contestable Grounds: Necessary Informational Conditions for Algorithmic Contestability.” Preprint, Zenodo, October 15, 2026. https://doi.org/10.5281/zenodo.23387270
- Thornhill, Kerry. In preparation. “Private Power without Epistemic Preemption: Relational Institutional Answerability in Algorithmic Decision-Making.” Preprint expected January 2027.
- Thornhill, Kerry. 2026. “From Operative Determination to Contestable Judgment: One Argument Map, Three Uses.” Working document, v. 3, September 2026. University of North Texas.