We measure the distance between what an organisation declares about one AI workflow and what its own material establishes.

Independent, Brussels. Nothing is sold after the report.

The established models place a person inside a decision. In the loop when the system waits for their approval, on the loop when they watch and may intervene, in command when they hold final authority. All three are local, and each describes one control point at a time.

None of them answers the question a director actually asks, which is not who approves this output, but where the humans are across everything now running, and what the whole set costs. That question is about the map, not the loop. Human in the Map is the position it is asked from, and the one an organisation cannot occupy from inside itself.

Each level can conclude that there is no next step. Two of the free tool's four outcomes route away from the offer rather than toward it: a deployment that is already documented, and no decision ahead.

The visibility gap is not a commercial intuition. Okta and Apprize360 (March 2026, 292 executives and 492 knowledge workers across seven countries) find 90% of leaders confident in their organisation's visibility over AI tools, while more than half of workers admit using tools without approval. TrustedTech and Censuswide (2,001 employees, UK and US) find senior decision-makers using unapproved AI tools at 65%, against 31% for the staff they manage. Both studies were commissioned by vendors with an interest in establishing a governance deficit, run on modest samples, with no breakdown by company size. The Baseline measures that distance in one particular case, not in general.

• The unit is a distance, not an inventory. The scope is one workflow, named and bounded in writing before anything starts. What is measured inside it is the distance between what the director describes and what the elements establish. At the Baseline the single source is the director, and what they cannot see does not become visible because we question them better.
• Exports, not access. Nothing is installed, no agent runs on the client side, nothing is left behind.
• Declared first, then checked. Starting from the systems would show what they were configured to do, not what the organisation does.
• Every finding says what it rests on. Confirmed · declared without an independent trace · gap established · open question.
• A gap needs a second source of a different kind before it is presented as established.
• No score. No grade, no maturity level, no percentage of an invisible total.
• Dated. Each family of findings carries the point from which it stops being reliable.

Two findings sit at the Full Map and not at the Baseline, for the same reason: rework time, which rests on an estimate that has to come from the people doing the work, and structural incompatibilities, which require a view from above that a single source looking from one position does not have.

Full method: methodology · sprinklingact.com/methodology

Until July 2026 Sprinkling Act sold a regulatory position: where an AI system falls under the AI Act, and which obligations follow. Four facts moved it. The April screening of 50 European companies returned 96% with no published AI Act position, a figure that supports two opposite commercial readings equally well. The Digital Omnibus provisional trilogue agreement of 7 May 2026 moved standalone Annex III obligations from 2 August 2026 to 2 December 2027, and an offer whose argument is a date loses its argument when the date recedes. Classifying a system article by article presupposes knowing how the work runs, which is precisely what is not known. And the 0-100 band aggregated exactly what the analysis exists to pull apart.

The gate framework was not retired, it was moved downstream: it names obligations once a workflow has been reconstructed. The papers published in April and May 2026 are not being edited to match the current position, because a timestamped document guarantees that this text existed on that date, and editing it destroys that guarantee. Read them as of their date.

Full account: methodology/TRANSITION.md

Four conditions have to hold before an engagement is accepted, and six named grounds lead us to refuse one. Published, with the refund capped at the price of the mission and a public erratum if a finding is later found to be wrong: sprinklingact.com/acceptance-criteria.

Not an audit, not a certification, not legal advice, not a monitoring tool, not a reseller. Sprinkling Act is not a law firm, not a notified body, not a certification body, and not affiliated with the European Commission, the European Parliament, or any national supervisory authority.

Nothing is sold after the report: no tool, no licence, no automation, no reselling of anyone's platform. Measuring and solving are different trades, and a measurer who sells the remedy destroys the economic condition of the measurement. That is what makes a finding checkable, rather than a promise of neutrality.

• methodology · how the analysis is conducted, and the AI Act gate framework that sits downstream of it.
• zenodo-mirror · index of published research, with DOIs.
• timestamps · OpenTimestamps proofs for the published PDFs, anchored on Bitcoin.

Open access, CC BY 4.0, DOI-indexed on Zenodo, cross-filed on SSRN, cryptographically timestamped.

• EU AI Act Readiness: A Structured Screening of 50 European AI Companies (April 2026). DOI 10.5281/zenodo.19671328. 74% of the 50 companies trigger high-risk classification, 96% publish no AI Act position, and none of the 14 already holding a certification had mapped the AI Act layer on top of it.
• Annex A: The Deployer Multiplier (May 2026). DOI 10.5281/zenodo.20042174.
• The AI Act as a Third Structural Pole (May 2026). DOI 10.5281/zenodo.20343243.
• The Deployer Multiplier: Classification, Cascade, and the Third Structural Pole of the EU AI Act (Eliva Press, June 2026, 82 pages). ISBN 978-99993-4-609-2. Consolidates the three papers above; the open-access versions remain the reference.

Public registry of published artefacts: sprinklingact.com/index. Lexicon of human oversight in AI workflows, 27 entries, each carrying the standing of its source and a caveat stating what that source does not allow anyone to claim: sprinklingact.com/lexique. Five of the 27 are marked not yet formalised, and they are the ones this method rests on.

Registered in Brussels, BCE BE 1034.962.482. Self-funded. No venture capital, no advisory board, no commercial relationship with AI vendors, regulators, or certification bodies.

• sprinklingact.com
• Analyst: Lamar B. Shucrani · ORCID 0009-0002-5093-8550