Methodology
How we measure
What This Is
The Keeling Curve has tracked atmospheric CO₂ since 1958. It is a single, continuous measurement that makes the invisible visible. The Delegation Curve seeks to do the same for AI's growing role in consequential decision-making throughout society. One composite score (0–100), updated regularly, so the shape of the curve is the story.
9 domains are tracked: content moderation, algorithmic trading, code generation, customer support, credit decisioning, medical diagnostics, legal AI, hiring, and education.
Each domain score is built from 2-5 normalized indicators sourced from transparency reports, regulatory filings, surveys, public datasets, and research.
Curve Maintenance
The public curve is kept on one measuring stick over time. When sources or scoring rules improve, earlier public points may be restated before showing headline movement.
Older published points are retained in the dataset for audit, but the main chart uses the current measurement series.
Every restatement and correction, with its effect on the composite, is recorded in the public restatement and correction log.
The Delegation Curve
The composite score is a weighted average across 9 decision domains. Each domain score (0–100) estimates AI influence in that sector using direct delegation measures where available and proxy signals where direct measurement is still incomplete.
Normalization
Raw metrics are normalized to 0–100 before weighting. Most inputs are direct percentages; count and reach metrics use fixed caps or log scaling so they can be compared with percentage-based indicators without implying they are literal decision shares.
Status Tiers
Domain Weights
The composite is a weighted average. Does the weighting matter? Try different lenses — the contribution bars show each domain's weighted impact, sorted by magnitude.
Published Thinkwright methodology
Data Source Tiers
Tier 1 — Direct Workflow Evidence
Platform transparency reports, operational datasets, regulatory or filing disclosures, and first-party telemetry. Highest confidence when definitions are stable.
Tier 2 — Strong Proxies
Official databases, industry surveys, product-market denominators, and adoption reports. Used when direct delegated-decision data is not yet available.
Tier 3 — Context & Guardrails
Public benchmarks, research studies, quality reports, and reliability evidence. Used to interpret confidence and scope, not as standalone proof of social delegation.
Evidence Grades
Every scored indicator carries a grade for how directly its source measures delegation, a confidence level, and the period it measures. Grades and sources are shown on each domain page.
Direct observed delegated decisions, task resolutions, operational outcomes, regulatory disclosures, or platform telemetry.
Workflow penetration, operational deployment, official databases, or strong representative surveys.
AI-attributable output, value, or productivity share, or narrower proxy evidence. Used with caveats.
Adoption, sentiment, capability benchmarks, or broad context. Scored only through an explicit, documented conversion.
Credibility Principles
Published methodology — every weight, formula, and source is version-controlled and public.
Raw data access — underlying indicators are downloadable.
Source confidence — uncertainty is documented, not hidden. Survey-based and proxy-heavy domains are treated more cautiously.
Continuous series — public prior points stay on the current measuring stick; archived publication points remain available for audit.
Method before interpretation — sources, weights, and formulas are separated from narrative conclusions.
AI Delegation Curve, Last updated 16 May 2026