State of the Index
The primary anonymized dataset. 1,628 operators across 17 platforms and 3,304 models. 9.07Q total tokens. Data as of 2026-07-13.
The Finding
When you measure 1,628 AI operators across 17 platforms, the first thing that jumps out is that volume is not yield. The operators who burn the most tokens are not the most efficient. The most efficient operators reuse context — cache reads dominate their cascade, and their Yield (Υ) is orders of magnitude higher than the heavy-input majority.
The aggregate numbers tell the story: 9.03Q of the 9.07Q total tokens observed are fresh input. Only 34.11T are cache reads — just 0.4% of all traffic. The cascade economy exists, but barely. Most operators are paying full price for every token. The ones who aren't are the ones at the top of the leaderboard.
This is the founding insight. The dataset is the evidence. Every quarter we will re-snapshot the index, publish a new version, and track how the distribution shifts as prompt caching becomes standard practice. This is v3.1 — the baseline.
The Source
This is the source material. Every article, dashboard, and finding published by SigRank derives from this dataset. It is available on Zenodo under CC-BY-4.0 — download the CSVs and JSON to run your own analysis.
The dataset measures AI operators — the humans driving AI tools — by four token pillars: input (fresh tokens sent), output (tokens generated), cache creation (context written to cache), and cache read (context reused from cache). From these four integers, every cascade metric is derived — including the headline efficiency metric Υ (Yield) = (cache_read × output) / input².
Operators
1,628
Platforms
17
Models
3,304
Total tokens
9.07Q
Cache %
0.4%
The Dataset
Data files (CSV + JSON)
| operators-raw.csv | 1,628 | Raw token telemetry: input, output, cache_read, cache_write, reasoning, total, cost, op_ratio, active_days, sessions, submissions + trans_exp |
| operators-derived.csv | 1,628 | Derived cascade metrics: yield, seed_sigrank, snr, leverage, velocity, compression, tokens_per_day, dev10x, scale_v, cost_per_million, efficiency + archetype |
| operators-platform-split.csv | 8,992 | Per-operator per-platform token breakdown |
| platform-raw.csv | 17 | Platform aggregate raw stats |
| platform-metrics.csv | 17 | Platform derived metrics |
| model-raw.csv | 3,304 | Model aggregate raw stats |
| model-metrics.csv | 3,304 | Model adoption patterns |
| archetypes.json | — | 10 build archetype statistics: N, %, median yield/leverage/velocity/SNR, token composition, classification thresholds |
| class-distribution-reference.json | — | 24-stage experience ladder observed distribution: population, min/max tokens per stage |
| experience_ladder.json | — | Canonical 24-stage classifier thresholds (total-token boundaries) |
| MANIFEST.json | — | All file paths, sizes, SHA-256 hashes, row counts, citation metadata |
Documentation
| README.md | Package overview, file descriptions, two-axis taxonomy summary |
| METHODS.md | All metric formulas, HCM cohort definition, archetype classification |
| PROVENANCE.md | Source, scrape date, exclusions, reproducibility |
| ANONYMIZATION.md | ID scheme, stripped fields, shuffle seeds, re-identification risk |
| DATASET-SPEC.md | Locked column specification for all 7 CSV files |
| RELEASE-NOTES.md | Version history (v1.0 → v3.1) |
Field Analysis
The full field analysis lives at /field — interactive charts, archetype distributions, platform adoption curves, Benford validation, percentile bands, and the cascade Sankey. It is the visual companion to this dataset.
Open Field Analysis →Methodology
Figures are computed from the SigRank Index — a privacy-preserving leaderboard ranking AI operators by token-cascade efficiency (Υ = cache_read × output / input²). Data is built from on-device, ed25519-signed token-telemetry snapshots. No message content is ever read or stored. Full methodology at /methodology.
Cite this dataset
McHenry, D. J. (2026). SigRank Two-Axis Operator Taxonomy: Finalized Datasets and Analytics Dashboards (v3.1) [Dataset]. Zenodo. https://doi.org/10.5281/zenodo.21900519
BibTeX
@dataset{sigrank_taxonomy_2026,
author = {McHenry, Deric J.},
title = {SigRank Two-Axis Operator Taxonomy: Finalized Datasets and Analytics Dashboards (v3.1)},
year = {2026},
publisher = {Zenodo},
version = {3.1},
doi = {10.5281/zenodo.21900519},
url = {https://doi.org/10.5281/zenodo.21900519},
note = {Concept DOI: 10.5281/zenodo.21875675. License: CC-BY-4.0. ORCID: 0009-0002-9904-5390.}
}License
This dataset is licensed under CC-BY-4.0. Attribution required — cite as shown above.