Weft Research · Methodology v0.1
Agent Commerce Index
A map of what agents buy—not just where tokens move.
Payment data can show that value moved. The Agent Commerce Index connects that evidence to providers, capabilities, endpoints, prices, and observed demand so the market becomes legible.
Current release
The method before the leaderboard.
This first release defines the observable universe, entity model, classification rules, and metrics. It intentionally publishes no fabricated market totals. Quantitative snapshots will appear only when each number can carry its source, window, coverage, and revision.
The entity graph
Blockchain analytics sees a transfer. A commerce index must test links among service identity, advertised capability, payment evidence, and measured outcome—and preserve provenance and confidence on every edge.
A fact directly observed in a registry, challenge, route, or settlement record.
ClassificationA Weft category attached with a taxonomy version and confidence.
OutcomeSeparately measured evidence that a result was returned and useful; payment alone never establishes this.
How the index is built
Every published snapshot follows the same five-stage pipeline.
- 01
Collect
Ingest declared service metadata, runtime payment offers, public settlement records, and consented Weft routing events with source and observation timestamps.
- 02
Resolve
Connect provider names, domains, wallets, services, capabilities, endpoints, and protocol offers. Preserve conflicting candidates and confidence instead of forcing a match.
- 03
Classify
Assign capabilities through a versioned taxonomy. Separate provider-declared labels, deterministic rules, and Weft-inferred categories.
- 04
Measure
Aggregate only metrics supported by the visible coverage: active endpoints, routed transactions, price distributions, repeat demand, protocol mix, and concentration.
- 05
Publish
Release a dated snapshot with methodology, coverage window, revision history, known blind spots, and reproducible definitions for every metric.
Coverage and evidence
Each input answers a different question. None is treated as a complete view of agent commerce.
Metrics worth publishing
Definitions come before numbers. Every metric is scoped to observable coverage.
- Active providers
- Resolved providers with at least one observed payable endpoint in the window.
- Capability price
- Median and percentile settled or offered price, grouped by taxonomy version.
- Routed demand
- Calls selected through Weft, separated from broader public settlement activity.
- Protocol mix
- Observed offers and settled calls by protocol and payment method.
- Repeat demand
- Repeated buyer-provider activity where buyer identity is legitimately observable.
- Concentration
- Share of observable calls or volume captured by leading providers in a capability.
Known blind spots
What this index will not claim.
- It is not the entire agent economy.
- One wallet is not assumed to equal one person, company, or agent.
- Onchain volume is not treated as proof of useful economic activity.
- An offered price, settled payment, and measured outcome remain separate evidence.
- Failed calls, retries, refunds, and suspected self-trading must be separated.
- Inferred capability labels never overwrite source facts.
Primary inputs