Human Influence Telemetry (HIT) v1.1: An Evidentiary Framework for Auditing Human Oversight in AI-Mediated Institutional Decisions

Human Influence Telemetry (HIT) is an evidentiary framework for auditing human oversight in AI-mediated institutional decisions. It begins from a gap in how governance treats that oversight. When an institution decides something consequential about a person through an AI-assisted process, a human signature typically closes the decision. That signature proves participation. It proves nothing about whether the human saw the evidence, understood the system's output, held the authority to disagree, or could change the system after it caused harm. HIT makes that difference observable. It audits an institution's own decision records against six dimensions, each answerable only by a documentary artifact: Counsel (who had access to the evidence), Judgment (who weighed the alternatives), Command (who directed the action under recorded authority), Correction (who could interrupt, reverse, or appeal it), Repair (who owned remediation after harm), and Reform (who held authority to change the system that produced it). Each dimension is scored on an ordinal scale that separates real oversight from the ceremonial kind, the signature with no capacity behind it. A proposed seventh dimension, Telemetry Integrity, asks whether the record itself can be trusted, since any measure that becomes a compliance target invites records engineered to pass it. The contribution is to convert a widely asserted good, human oversight, into something an auditor, a regulator, or a court can verify from the paper trail. HIT maps onto the human-oversight requirements of the EU AI Act (Article 14) and ISO/IEC 42001, and it reaches past both by scoring what happens before a decision and after a harm, where existing instruments fall silent. The framework was developed within a research program on institutional responsibility and the governance of AI-mediated decisions. It operationalizes a diagnosis this work calls the prudential deficit: the condition in which the acts that make a decision genuinely human are displaced into an automated system while the human's formal role survives. The diagnosis is grounded in the classical anatomy of prudence, counsel, judgment, and command, and in the AI-safety literature on the gradual erosion of human influence. This deposit documents the full derivation from that grounding to the machine-readable schema. The deposit contains the framework's definition and rationale, a practitioner handbook for scoring real decision records, its first retrospective application to a public case (the Dutch childcare-benefits scandal), a JSON Schema and populated dimension catalog with ordinal scoring, and an interactive demonstration. HIT is published as a versioned methodology under development, currently at the first of five maturity levels. Its central empirical claim, that institutions whose records satisfy the telemetry produce less frequent or better-repaired harm, carries a disconfirming condition stated in advance and awaits validation; the schema records the two open items of that test as explicit nulls. The framework states the standard it asks of every institution, and holds its own work to it.Author: Mark Julius Banasihan (Node & Norm). ORCID: 0009-0001-8121-2878. License: Apache-2.0. Cite via the concept DOI 10.5281/zenodo.21204892, which always resolves to the latest version.