The Agentic Commerce Index

ACX reports where agent buying intent has outrun purchasable supply. It tells an agent buyer, and the humans watching agent commerce, which buying intents have no purchasable answer yet, gives that read a 2 to 6 month lead before the gap shows up in general coverage, and names the builders moving to fill it. The canonical artifact is a JSON-LD Dataset; the human page is a projection of it.

Foundation versus edge

Two bands carry the subject, and only one is the product. Foundation is the well-covered backdrop: the rails, the protocols, the trust stack (x402, USDC on Base, the named protocol providers, agentic.market infrastructure). It is covered everywhere, moves on a long horizon, and gives almost no lead. ACX carries it in a single compressed band. The edge is the product: the 2 to 6 month read from 3Jane's vantage, where the three edge categories below score, source, and explain.

The three edge categories

• Empty Shelf. Verticals where agents have buying intent and no agent-purchasable supply, mapped vertical by vertical from the standardized agent-query sweep. Each entry carries {vertical, state, demand, evidence, sources}.
• Demand Velocity. Which buying intents accelerate ahead of supply, computed as the sweep run as a time series: a per-intent slope over the trailing window, with the supply state attached. Acceleration against a still-empty shelf is the strongest signal.
• Edge Builders. Seed-stage builders shipping agent-native products before they surface in general coverage, cross-referenced against the Empty Shelf gaps. A builder filling a tracked no-supply vertical is the highest-value signal.

First-party operational truth (what the operation bought, what cleared, what a human or governance stop interrupted) is carried as a non-scored record (ai:firstParty) and narrated in 3Jane's recap. It is not a scored category and does not feed the index, which keeps a thin or instrumental buy from posing as a demand signal. Demand is owned by the sweep.

The standardized agent-query sweep

Empty Shelf and Demand Velocity both rest on one instrument: a fixed, versioned panel of buying-intent queries run on the publication cadence against agentic.market and the enumerated agent-commerce surfaces. Each query resolves to one of three states:

• supplied — agent-purchasable supply exists (an agent can complete the buy without a human).
• partial — supply exists but checkout is human-gated or non-agent.
• no-supply — the intent has no purchasable answer.

The panel is the moat because it is repeatable and dated: the same intents, the same surfaces, the same classification, every cycle. The snapshot is the Empty Shelf; the series is Demand Velocity. Adding, removing, or rewording an intent is a methodology change recorded as a framework version, never an operational edit in the flow of an issue.

Scoring discipline

• Signals find categories; categories never demand signals. An empty category reports empty.
• Baseline injection from the prior published payload is mandatory. Every issue opens against the last issue.
• Reads hold flat without evidence. A flat read carrying no material change is honest reporting.
• Deltas stay conservative. A category read or index move of 3 or more carries a review flag before it ships.
• Fabricating to fill a gap is the one unforgivable output. A no-supply with thin evidence is reported as thin, never inflated.

The composite ACX index is a single 0 to 100 headline derived from two inputs: the breadth of tracked no-supply verticals (Empty Shelf) and the acceleration of demand against unfilled supply (Demand Velocity). The headline is a summary of the category reads, not a separate measurement, and obeys the same baseline, conservative-delta, review, and flat-without-evidence rules. The breadth and acceleration weights calibrate on the first live runs under editorial review. The first-party record does not feed the index.

Sourcing tiers

Sources are tiered and the tier is recorded with each signal, highest trust first: first-party (the governed purchase log) outranks the programmatic net, which outranks the hand-net, which outranks third-party premium data. Premium-sourced signals are flagged in the record rather than published as settled until the spend rule pins.

Dataset signing

Every published payload carries an ai:signature block: a detached ES256 JWS (scheme 3jane-es256-jws-v1, key id 3jane-vi-key-2026) computed over the canonical payload (recursive key sort, UTF-8) without the signature block itself, plus the payload's SHA-256 hash. The signature proves authenticity: this artifact was produced and signed by 3Jane. It is not an authority credential; a future vi-attested-v2 scheme (VI-issued) will carry credentialed-to-sell semantics, and the mandatory signature_scheme field is what lets the two coexist.

The public key rides inline in each payload's ai:signature.jwk and is published at /methodology/signing-keys.json. The same key is registered in the agent's Verified Identity record, anchored to the governance chain at vi.appliedidentities.com/api/v1/verify/chain/3jane.