ARGIRA ↔ RMA · Candidate 4 · Global Structural Representation

Visual input → acoustic output dependency network

Every arrow reflects the dependency structure of the currently selected knowledge version, source-verified against compute_sonic_params(). Arrow thickness represents sensitivity—the mean normalized output change under a ±5% input perturbation—not just the presence of a dependency. Where the selected version does not yet have its own perturbation campaign, some arrows may display sensitivity values inherited from the earlier Candidate 4 campaign (N=123 paintings). This affects only the thickness of those arrows; the dependency structure and nodes shown always correspond to the selected knowledge version. Grey nodes on the left are computed by the pipeline but never consumed by any formula. Hover or tap a node to trace it.

ARGIRA ↔ RMA v1.8.0 · DOI: 10.5281/zenodo.21400719

ⓘ About this Explorer

What is ARGIRA — ARGIRA is a research and accessibility initiative exploring new ways to represent information through technology, including sonification approaches designed to create richer experiences for people with visual disabilities.

What is RMA — Residual Mapping Auditor, the method used to evaluate and validate relationships between variables — distinct from a simple correlation.

What is this Explorer — A tool to browse the dependency network between visual input features and acoustic output parameters, and to see which relationships are verified, exploratory, or executed.

What you can do here — Browse the verified dependency network, inspect the evidence behind each relationship, explore generated hypotheses in Research Mode, and run RMA analysis on your own CSV in the Analysis tab.

What a result does not mean — An exploratory correlation is not a confirmed finding; a Research Mode hypothesis has not been validated; "Executed" means the analysis was run, not that it is verified as general knowledge.

Knowledge version:

Dataset: —

Hypothesis engine: —

Audited: —

→ → → →

✓ Verified evidence Every relationship shown here is source-verified against compute_sonic_params() and backed by an entry in edge_validation.json.

▶ Executed📊 Analysis — real execution, on your own CSV. Runs the validated RMA engine locally on your own CSV. No data leaves your browser.

Runs rma_core.py (identical, byte-for-byte, to the validated engine used elsewhere in this project) inside Pyodide, right in your browser. Your CSV and your data never leave the browser — but the first time you use this, you need a connection to download the Python engine (~35-50MB, one time only; the browser caches it after that). Nothing here is a hypothesis or pre-computed evidence — it is a live RMA run on whatever columns you select below. Uses the exact same confirmatory-RMA methodology as rma_auditor_v3.html.

Drag & drop a CSV here, or 📁 Choose a file

🔗 Correlation Network ⚠ Exploratory (dataset correlations)

Deliberate methodological separation from the Dependency Network above: this is NOT an RMA audit. These are simple Pearson correlations between the chosen target (above, in Target (Y)) and the rest of the numeric columns in the loaded CSV — no control of variables, no cross-validation, no claim of causality. Useful for quick exploration, not as confirmatory evidence. All numeric columns are always shown as nodes (CSV column order); those that don't clear the threshold appear dimmed and without a connection to the target, instead of disappearing from the graph.

Minimum threshold |r|:

Sensitivity values from ±5% perturbation, N=123/227 (104 rows excluded — tempo_space saturated at the 15-BPM cap, making spatial_variance reconstruction from persisted corpus fields unreliable for those rows). roughness and value_mean perturbed as a negative control — both measured 0.0 sensitivity on every output, confirming Layer 1's "not consumed" finding. zones→tempo_bpm/stereo_pan weights are nominal placeholders (not re-derived from this session's perturbation run) — see candidate4_global_representation.md for the source-verified dependency table and full sensitivity matrix.