The setting
Brazil’s Supremo Tribunal Federal describes Victor as a classifier supporting the handling of appeals and their possible correspondence with general-repercussion themes. The source is the court’s own Portuguese-language account, paraphrased here in English.
The decision context
An appraisal of a dated institutional claim, not an independent benchmark or a claim that AI decides appeals. Victor is distinct from the court’s similarly named VitorIA tool.
The evidence to examine
Reported evidence
The reported task-time claim
The court’s August 2026 article attributes to a World Bank report a reduction from 40 minutes to five seconds for linking appeals to general-repercussion themes.
Source: Brazil's Supremo Tribunal Federal, Victor and the boundary of judicial automation (2026) · Opening paragraphs: attributed 40-minute/five-second comparison ↓Reported evidence
What the system does
The article describes Victor as analysing an appeal and indicating a possible correspondence with themes; it does not decide the case’s classification. Other stages still require human work.
Source: Brazil's Supremo Tribunal Federal, Victor and the boundary of judicial automation (2026) · Como funciona and the paragraph on the initial analysis stage ↓AIPEX analysis
AIPEX reading question
A reduction in one processing task leaves questions about checking, rework, queueing and the eventual experience of people seeking justice.
What this does not establish
The cited article does not provide a reproducible timing protocol, sample distribution or a causal estimate of total case duration. Its attributed benchmark has not been independently reproduced by AIPEX. It does not establish judicial quality, fairness or present performance across all appeals.
Try the decision first
Your task.
What would make a task-time comparison meaningful evidence for a wider court-service decision?
- Define the task boundary and comparator: staff activity, machine processing, checking and rework. Ask whether the timing covers comparable cases and resources.
- Request classification errors, difficult or novel themes and effects on downstream staff decisions. Identify when an incorrect label can be detected and corrected.
- Propose measures of end-to-end delay, workload and treatment of different case types. Keep the reported efficiency claim distinct from a claim about access to justice.
Read one defensible approach
AIPEX analysis for discussion. Other conclusions can be defensible if their evidence, assumptions and trade-offs are explicit.
- Quote the timing as an attributed institutional claim about a bounded task. Do not divide the two numbers and present the ratio as a measured improvement in court efficiency.
- Ask for the timing protocol and an error/rework analysis before accepting a net saving. Organisation-wide benefits depend on whether that task is a real bottleneck.
- Use a bounded comparison of the actual workflow, with correction and escalation for novel cases. Preserve human legal responsibility while examining whether early labels influence later judgement.
Read the evidence / Reviewed 18 October 2026
Sources & scope.
AIPEX’s prompts and worked approaches are our own synthesis. Use the linked material to examine the method and its limits.
- Brazil's Supremo Tribunal Federal
Victor and the boundary of judicial automation (2026) ↗
19 August 2026; opening account of the 40-minute/five-second claim and Como funciona
Court-authored Portuguese-language account, paraphrased by AIPEX in English. Its timing claim concerns an initial classification task and attributes the comparison to a World Bank report. It is not treated here as an independently reproduced evaluation of total case duration or judicial quality. Victor and VitorIA are distinct tools.
- UK Government
Guidance on the impact evaluation of AI interventions ↗
Evaluating AI interventions in their operational setting
Helps frame evaluation of the intervention and its workflow, alongside technical model testing.
- HM Treasury
The Magenta Book (2026) ↗
Evaluation design, impact evaluation and the counterfactual
Use to distinguish observed change from effects attributable to an intervention. Evaluation design must fit the question and context.