Skip to content

Turn any idea, plan, or codebase into a beautiful interactive diagram. An agent skill for Claude Code, Codex, and more.

3.4 KB
Raw Blame

Artifact → Install v2 measurement contract

Artifact → Install v2 is a bounded change to the existing journey. Proof Lab cards link to start.html?type=<type>&source=gallery, and the Start page selects the matching typed recipe and can copy the selected agent's install command plus the bounded prompt in one action. Generated artifacts no longer emit the former footer link; source=artifact remains accepted only for compatibility with already-shared artifact URLs.

The page records only coarse interaction steps:

  • start_view
  • starter_copy
  • global_install_copy
  • project_install_copy
  • prompt_copy
  • proof_open

Each event contains schemaVersion, step, the allow-listed source, diagram type, agent, and language. It never contains a repository path, diagram title, prompt text, node, relationship, source JSON, or user identifier. Unknown source values are normalized to direct.

Privacy boundary

Events are capped at the latest 24 entries in sessionStorage under archify.start.events.v1. The page also dispatches the same detail as the archify:start-funnel browser event and exposes a read-only snapshot through window.ArchifyStartMetrics.snapshot().

No network request is made by Archify. The page contains no fetch, sendBeacon, or XHR path. Closing the browser session discards the local receipt. A hosting operator must not attach an external event sink without a separate privacy decision, public disclosure, retention policy, and test update.

Observable metrics

Within a controlled browser test or an explicitly instrumented host, compute:

  • Legacy artifact arrival share: source=artifact start_view / all start_view
  • Gallery arrival share: source=gallery start_view / all start_view
  • Combined starter copy rate: starter_copy / start_view
  • Any install copy rate: (starter_copy + global_install_copy + project_install_copy) / start_view
  • Prompt-only copy rate: prompt_copy / start_view
  • Proof reassurance rate: proof_open / start_view

Count at most one occurrence of each step per browser session when reporting conversion, even though the local diagnostic receipt retains repeated clicks. Segment only by the allow-listed source, type, agent, and language.

What this does not prove

First-diagram success is not observable from this static page. A successful copy is an intent signal, not proof that installation completed, the agent generated a candidate, validation passed, or the user retained Archify. Do not label these ratios “activation,” “success,” or “retention.”

The Ordinary-Model Floor benchmark measures candidate quality independently. A future end-to-end activation metric would require an explicit, privacy-reviewed receipt boundary at the installed CLI or a voluntary user submission; it must not be inferred from Start-page clicks.

Evaluation window

Before publishing a conversion claim:

  1. Pin the repository commit and the exact v2 UI.
  2. Validate the event contract in a real browser for artifact, gallery, direct, and rejected unknown-source URLs.
  3. If an approved aggregate sink is later added, run a predeclared observation window and publish both session counts and rates; do not compare against a historical period that lacked equivalent instrumentation.
  4. Keep deterministic page tests, package tests, and visual review as release gates regardless of copy rate.

Until an approved aggregate sink exists, the event receipt is for local QA and experiment readiness only.