CrawlClick

System, end to end

A commercial test with every boundary visible.

CrawlClick keeps serving truth, money truth and outcome evidence separate. That separation prevents a successful insertion from being reported as a successful answer-engine outcome.

Product evidence map

What we can prove—and what the pilot must learn

No illustrative traffic claims

  1. 01 / built

    Verified request

    Identify the crawler from vendor-backed network evidence, not a forgeable user-agent alone.

  2. 02 / built

    Disclosed insertion

    Bind the reviewed claim, destination, price and publisher to a traceable event.

  3. 03 / built

    Money movement

    Prepay, settle balanced books, and make publisher payouts through Stripe Connect.

  4. 04 / validating

    Answer surfacing

    Measure whether an engine carries the sourced, disclosed claim into an answer. No outcome is guaranteed.

Request path

What happens when someone requests a publisher page?

01

Classify

Separate ordinary readers, directional referral signals, training agents, index crawlers and live user agents.

02

Verify

Require network evidence for billable crawler traffic. A user-agent or human navigation header alone cannot become a charge.

03

Fetch

Resolve the verified public origin once, connect over TLS and preserve safe origin semantics.

04

Decide

For an eligible verified crawler, match reviewed and funded campaigns against page context and safety policy.

05

Disclose

Insert one visibly sponsored passage and durably record it. Human and ineligible requests receive the unchanged origin response.

Data plane

Why does serving avoid Postgres?

Crawler traffic can be enormous and bursty. Eligibility, targeting and pacing therefore read immutable object-storage snapshots. The relational control plane is never a dependency of a request hot path.

Control plane

Where does commercial state live?

Postgres owns identities, organisations, sessions, verified sites, Stripe objects, the operation outbox and balanced ledger. Tokens are stored only as hashes and every tenant query is organisation-scoped.

How does failure behave?

Origin problem

Unsafe, private or malformed origins are rejected during setup instead of discovered on a live crawl.

CrawlClick problem

The Cloudflare edge fails open to the publisher's real page, preserving availability.

Money problem

Stripe events and operations are idempotent and retryable; ambiguous state cannot silently credit or pay twice.

Invite-only pilot

Help answer the commercial question.

We are selecting a small set of US publishers and advertisers for a test-mode-first pilot. Every request is reviewed by a person.