Abstract

Governments worldwide increasingly deploy algorithmic systems to allocate scarce public goods. The dominant transparency paradigm focuses on expert audit rather than citizen comprehension. We propose a substantive standard, Meaningful Algorithmic Transparency (MAT): information delivered to affected users must be proactive (placed into awareness, not merely published), tailored (referring to the user's decision context), and actionable (delivered before the decision closes). We develop the argument through Chile's Sistema de Admisión Escolar (SAE), a national school admissions platform serving roughly 470,000 families per year. Two measurements document the residual comprehension gap. Linking the SAE Satisfaction Survey to administrative truth for four cycles (2020–2023, N≈140,000), we show that families systematically overstate their placement probability: among applicants the platform can pre-identify as high-risk, the median family believes the risk is around 20% when it is in fact around 80%. The bias is uneven: at the same objective risk, lower-SES applicants understate by more, with a conditional gradient that is statistically stable across all four cycles. In the SAE Parent Surveys (2023–25), self-reported familiarity with a key procedural rule nearly doubled across cycles while verifiable comprehension barely moved (2.3% to 4.9%). The comprehension gap is therefore persistent and unevenly distributed, falling more heavily on the families whose outcomes most depend on effective information delivery, even when the underlying mechanism gives those families priority. We propose six operational components and show how Chile's existing regulatory architecture (access-to-information law, algorithmic-transparency recommendations, data-protection safeguards, and sector-specific regulation) can be articulated to make the standard enforceable.

@article{ algorithmic-transparency-mat_2026,
  title = "Making Algorithmic Transparency Meaningful: Evidence from Chile's National School Admissions Platform",
  author = "Medina, Exequiel and Mesías, Leonardo Ortiz and Neilson, Christopher A.",
  journal = "ConsiliumBots Working Paper Series, Issue 05",
  year = "2026",
  note = "wp"
,
  abstract = "Governments worldwide increasingly deploy algorithmic systems to allocate scarce public goods. The dominant transparency paradigm focuses on expert audit rather than citizen comprehension. We propose a substantive standard, Meaningful Algorithmic Transparency (MAT): information delivered to affected users must be proactive (placed into awareness, not merely published), tailored (referring to the user's decision context), and actionable (delivered before the decision closes). We develop the argument through Chile's Sistema de Admisión Escolar (SAE), a national school admissions platform serving roughly 470,000 families per year.Two measurements document the residual comprehension gap. Linking the SAE Satisfaction Survey to administrative truth for four cycles (2020–2023, N≈140,000), we show that families systematically overstate their placement probability: among applicants the platform can pre-identify as high-risk, the median family believes the risk is around 20% when it is in fact around 80%. The bias is uneven: at the same objective risk, lower-SES applicants understate by more, with a conditional gradient that is statistically stable across all four cycles. In the SAE Parent Surveys (2023–25), self-reported familiarity with a key procedural rule nearly doubled across cycles while verifiable comprehension barely moved (2.3% to 4.9%).The comprehension gap is therefore persistent and unevenly distributed, falling more heavily on the families whose outcomes most depend on effective information delivery, even when the underlying mechanism gives those families priority. We propose six operational components and show how Chile's existing regulatory architecture (access-to-information law, algorithmic-transparency recommendations, data-protection safeguards, and sector-specific regulation) can be articulated to make the standard enforceable."
,
  url = "https://consiliumbots.github.io/working-papers-cb/algorithmic_transparency/working_paper/algorithmic_transparency.pdf"
,
  url_article = "https://consiliumbots.github.io/working-papers-cb/algorithmic_transparency/working_paper/algorithmic_transparency.pdf"
}

Overview

A decade after adoption, Chile’s SAE illustrates both the promise and the ceiling of algorithmic transparency. The system sits at the frontier of what publication-centered transparency can achieve: a strategy-proof deferred-acceptance mechanism, unrestricted preference lists, and full publication of the algorithm, the rules, the data, and a simulator. The paper asks whether that is enough for the families who must actually use it, and finds that it is not.

Key points

• Two paradigms. “Transparency 1.0” (formal transparency) publishes the algorithm so experts and auditors can scrutinize it. Meaningful Algorithmic Transparency (MAT, or “Transparency 2.0”) requires information that is proactive, tailored, and actionable, with compliance judged by verifiable user comprehension rather than by the existence of documents.
• Miscalibrated beliefs at population scale. Linking the SAE Satisfaction Survey (2020–2023, N ≈ 140,000) to objective risk recovered by re-running the published algorithm, high-risk families believe their non-placement risk is about 20% when it is about 80% — a median gap of 49 to 62 percentage points in every cycle.
• Declared versus verified understanding. Declared familiarity with the family-application rule nearly doubled between the 2023–24 and 2024–25 Parent Surveys (36.8% to 60.4% among the highest-education group) while verifiable comprehension moved only from 2.3% to 4.9%. Among parents who claim to understand the rule, only 6 to 8 percent actually do.
• The gap is regressive. At the same objective risk, lower-SES applicants understate risk by more, consistently across two independent SES proxies and stable across four cycles — so the informational layer runs against the equity logic built into the allocation mechanism itself.
• An enforceable standard. Six operational components (identifying moments of consequential choice, concrete acts of notification, tailored case-specific information, actionable explanation with time to act, periodic measurement of comprehension, and external audit) can be grounded in existing Chilean legal materials without new legislation.
• Beyond Chile. The same question faces Brazil’s SISU, Kenya’s secondary placement, India’s RTE reservations, and Colombia’s SISBEN: at what point does the state’s obligation shift from making an algorithm auditable to making it understandable to the people it affects?

• Coauthors: Exequiel Medina, Leonardo Ortiz Mesías
• Status: Working paper · soliciting feedback
  Venue: ConsiliumBots Working Paper Series, Issue 05
• Date: 2026-08-10
• PDF