SigRank Two-Axis Operator Taxonomy: Finalized Datasets and Analytics Dashboards

When 1,627 AI operators are classified along two axes, structure emerges that is invisible in aggregate. Axis 1 — Build archetypes: 10 deterministic composition types based on leverage (cache_read/input), velocity (output/input), and construction (cache_write/cache_read). No clustering. When the HCM cohort (1,586 operators) is classified this way, yield spreads 9,387x, leverage spreads 133x, and fresh input ranges from 0.27% to 26.62%. Axis 2 — Experience ladder: 24 stages (8 tiers x 3 sub-stages) by descending first-match over fixed total-token thresholds. The canonical classifier thresholds are in experience_ladder.json. The observed distribution is produced by applying those thresholds to all 1,627 source records (class-distribution-reference.json). Stage populations follow Option C target proportions, not equal-population binning. The two axes are designed as separate dimensions (orthogonal by design, not empirically tested for independence). v3.1 adds the full anonymized operator-level CSV dataset (7 CSV files: operators-raw, operators-derived, operators-platform-split, platform-boards, model-boards) with SHA-256 anonymized IDs, 4 interactive HTML dashboards, 2 SVG charts, 10 animated GIFs, and full provenance/anonymization documentation. 1,628 operators across 17 platforms and 3,304 models. Contents: canonical classifier, observed distribution, archetype statistics, 7 anonymized CSVs, 4 interactive dashboards, 10 animated GIFs, METHODS.md (all metric formulas), PROVENANCE.md (source, exclusions, reproducibility), ANONYMIZATION.md (ID scheme, stripped fields, re-identification risk), DATASET-SPEC.md (locked column spec), generator scripts for both axes, release notes. CC-BY-4.0.