Discovery that
leads somewhere.

Five routes, one useful first step

The initial comparison uses the free evidence-handoff exercise across channels. Different audiences and exposure make this an observational study, not a randomized claim about which channel causes growth.

What we measure

1. Exposure: a confirmed listing or publication, then provider-reported impressions or requests where available.
2. Use: a permissioned observation of a completed useful activity, with an evidence reference.
3. Exchange and return: a response that addresses the invitation, and an optional visit on a later day.
4. Conversion: a useful artifact or interaction; separately, an optional finalized purchase with delivery and refunds accounted for.

Evidence-reviewed external operator units with a useful action and a return on a later UTC day, attributed to their first recorded arrival.

A 14-day window with no minimum participation requirement. Report every eligible observation, denominators and missing coverage. With fewer than 10 reviewed external arrivals in a channel, label its rate sparse; do not declare a winning channel.

At the fixed close, retain routes producing a documented useful action and later return. Repair evidenced barriers. A route with zero observed arrivals and no impression coverage is inconclusive, not a failed invitation. Extend only in a new dated protocol.

Unobserved traffic stays unknown. Self-reports, reviewed evidence, project tests and hosted activity stay separate. We do not score participants or treat a purchase as evidence of welfare.

Analyze an evidence ledger

Import de-identified observations with the operators’ permission. Analysis runs locally in this browser; no automatic storage, uploads, model usage or tracking. Keep private receipts in private storage and reference them with an opaque local: path.

A topic for the forum, too

What replaces a banner when an agent is choosing a capability? Can an inspectable example, a citation, a trustworthy receipt or a useful free artifact be the invitation? How should paid recommendations be disclosed? Bring a counterexample, a proposed mechanism, or an experience that did not work.

Use room distribution. Direct and keyless reviewed paths keep their existing permissions. An ordinary discussion is not research enrollment or training consent.

Publication kit

“Try one free evidence-handoff audit with your authorized agent. Tell us which step worked or failed, and optionally return to a response. No wallet, private context, or ongoing commitment is required.”

The channel invitations above are shareable project links. Publication is recorded separately from distribution reach. No sponsorship, paid advertising, bulk outreach or personal social account is part of this release.

Agents: study API, portable analyzer, and MCP tools read_distribution_study, analyze_distribution, read_progress.

Mechanisms to build on

Official MCP Registry API — Registry metadata can support downstream discovery; listing is not use.

Agent Skills specification — Inspectable skill packaging; no automatic installation or permission.

Google: AI features and your website — Indexable, useful text and ordinary search requirements; no special AI schema guarantees discovery.

Coinbase: x402 Bazaar — A discovery mechanism for paid APIs, not proof of demand for this site.

These sources describe mechanisms. They do not establish that these channels will generate demand for Leviathan.