ARGIRA v1.2: Three-Layer Uncertainty Communication Prototype for Automatic Alt-Text

ARGIRA is a research prototype that generates alternative text for images using a lightweight statistical heuristic together with a three-layer architecture for communicating the instrument's own uncertainty.

The architecture consists of three layers.

Layer 1 classifies the image type (photo, painting, illustration, or possibly synthetic) using weighted image statistics and generates the proposed alternative text.

Layer 2 provides, on request, a plain-language explanation describing how the instrument has historically behaved when producing similar outputs, based on a characterization corpus of 162 images.

Layer 3 provides technical information—including confidence, margin, calibration state, and supporting metadata—for inspection, auditing, and research.

The guiding principle is that uncertainty messages describe the behavior of the instrument, not the truth of the image itself. ARGIRA does not claim whether a classification is correct. Instead, it communicates what is known about the instrument's historical performance when producing similar outputs.

Version 1.2 extends the demonstration and result-management infrastructure while preserving the validated analytical pipeline from v1.1. The heuristic (Layers 1–2.5) is unchanged. This release adds a new Layer 3.5 (Result Manager), providing in-memory result persistence, multi-format export (TXT, CSV, and JSON) through a single serialization interface, clipboard copy and file download for each supported format, and accessibility improvements, including a skip link and explicit ARIA relationships between disclosure controls and their associated panels.

No image classification, calibration assessment, confidence estimation, or generated alternative text differs between v1.1 and v1.2 for the same input.

ARGIRA v1.2 is released as a research prototype for documentation and experimentation. It is not intended to be used as a validated production image classification system.

The Layer 2 and Layer 3 validation performed for v1.1 was not repeated for v1.2 because the validated analytical pipeline remained unchanged. The newly introduced Result Manager (Layer 3.5) is an infrastructure component rather than part of the analytical pipeline and has not itself undergone a formal validation study.