ARGIRA v1.4.5: Three-Layer Uncertainty Communication Prototype for Automatic Alt-Text — Public Uncertainty-Notice Behavior Fixed

ARGIRA v1.4.5 is a public release of the ARGIRA prototype for automatic alt-text generation and three-layer uncertainty communication. This version establishes a deliberate, consistent, and transparent public behavior for communicating calibrated uncertainty: when a result is identified as known_unreliable, the prototype provides a brief automatic notice explaining that uncertainty to the user.

It is a new version of ARGIRA v1.4.4 (DOI: 10.5281/zenodo.21924218).

A deliberate public uncertainty-communication behavior

ARGIRA v1.4.5 provides a single, stable uncertainty-notice behavior in both English and Spanish HTML artifacts. KNOWN_UNRELIABLE_MODE is fixed to brief_auto_notice, so the automatic notice is shown consistently whenever a result's calibration state is known_unreliable.

This decision allows the public prototype to present its current uncertainty-communication approach directly and consistently, without turning the public demo into an experiment on its users. The current version therefore stands on its own as a transparent research prototype, while leaving comparative evaluation of alternative notice modes for a properly designed future study.

Experimental infrastructure preserved, not deleted

An internal, unpublished working checkpoint built on top of v1.4.4 had connected KNOWN_UNRELIABLE_MODE to SESSION_CONDITION, an experimental session-level condition assigned 50/50 between 'request_only' and 'brief_auto_notice' on each page load. That A/B behavior was never part of a published Zenodo release.

SESSION_CONDITION and assignSessionCondition() remain present in the code but are disconnected from public behavior. They are kept, unused, as infrastructure for a possible future experimental phase — one that would require its own methodology, data-collection plan, informed consent, and privacy design before being reconnected. No user feedback mechanism, experimental data collection, or mode selector was added in this release, and no A/B testing is exposed to the public demo.

Visible explanation added

Both HTML artifacts now include a short, visible section explaining the current uncertainty-communication mode: that the public version uses the automatic notice, that the request_only variant and any A/B comparison are not part of the current public experience, and that a future phase may study both variants experimentally under an appropriate protocol.

Experimental scope and continuity

ARGIRA v1.4.5 does not introduce a new image-classification algorithm, calibration model, or change to experimental thresholds. The underlying heuristic classification (classifyImageVisual), the calibration table and lookup logic, the runner-up/margin computation and its 0.18 threshold, level1PlusNotice(), the notice_mode_used field and its CSV/JSON export, and the res.alt.text consistency fix (ensuring the on-screen text, aria-label, copy/apply-to-alt actions, and all exports agree) are all preserved unchanged from the prior working checkpoint.

The release remains an experimental research artifact and is not presented as a finished product or as validated accessibility technology.

The purpose of this release is to establish a clear and honest public behavior for uncertainty communication, while preserving the infrastructure for a properly designed future study. The prototype communicates openly what uncertainty mode the user is experiencing, rather than silently varying that experience between sessions.

Accessibility verification scope

Manual developer verification with Android TalkBack carried forward from previous releases remains part of the accessibility evidence for the prototype. Apple VoiceOver was not evaluated because an available Apple/VoiceOver testing environment was not available. This remains an untested scope, not a failed validation. No external user study or participant evaluation was conducted for this release.

Package contents

prototype/argira-demo-v5.html — English prototype artifact.

prototype/argira-demo-es-v5.html — Spanish-localized prototype artifact.

README.md — release overview and usage information.

CHANGELOG.md — version history.

CITATION.cff — machine-readable citation metadata.

LICENSE.txt — CC BY-NC-SA 4.0 license text.

argira-results-4.csv to argira-results-7.csv — CSV exports captured during the prior working checkpoint, with the experimental A/B condition still active (historical evidence, not current public behavior).

argira-results-9.csv, argira-results-10.csv — CSV exports captured under v1.4.5's fixed public behavior, confirming notice_mode_used is consistently brief_auto_notice.

The release provides two independently executable prototype artifacts, in English and Spanish, with structural and functional parity between the variants.