/segmently-unit-economics

UC-4 — RevenueCat: observed inputs → scenario

Route: has or asks about RevenueCat (SKILL.md § Router) · Eval: P4 · MCP: uc-4, revenuecat The procedure is below: Turns, Calls, Answer skeleton, Boundary, Stops.

Turns

1. Connect — skipped when RevenueCat read tools are present in this session: it runs on the READER's own RevenueCat MCP connection, never a key through this skill (not connected: § Stops).
2. Match — the reader matches RevenueCat products to paywall products by name.

Calls

• Read tools only: get-overview-metrics, get-revenue-metric, get-chart-data, get-chart-options-schema, get-benchmarks, list-experiments, get-experiment, get-experiment-results. Never create-, start- or publish- tools. Never cost per start or chain rates from RevenueCat.
• The pulls, with maturity rules: trial_conversion_rate by product → products[i].trialConv (cohorts older than trial days + 3); subscription_retention → paymentsCounted = 1 + the sum of the observed retained periods (input preparation: list the periods and the sum, labelled "your numbers, prepared"; partial, no extrapolation); refund_rate → deductions.refunds (basis: all transactions — label it); revenue types → tax share and fee share (input preparation: show both revenue figures; label "as RevenueCat reports proceeds").
• Append an observed-inputs document to observed[] ({ "kind": "observed_inputs", "v": 1, "source": "revenuecat", "window": … }), apply the values to the scenario, set measuredOn = "RevenueCat <window>" (SKILL.md § Files: re-basing moves the old base into variants[] first).
• npx -y @segmently/cli ue evaluate .ue/<slug>.json --explain (the stage becomes live_measured), then npx -y @segmently/cli ue link <file> --label '<name>' on the updated project.

Answer skeleton

1. The pulls, each with its window, and every prepared figure with both operands, labelled "your numbers, prepared" (a partial retention sum says partial).
2. The observed[] document appended and measuredOn = "RevenueCat <window>" set.
3. The evaluate: stage live_measured, the KPIs, then the drift check (references/stages.md) — each observed value beside the scenario's input, with the absolute difference.
4. The link: privacy.sentence first when privacy.due, the ue link call, then the link.
5. The boundary: walk.opener, the re-based base's walk.lines, then the line of § Boundary.

Boundary

Assembled (SKILL.md § The boundary) from the rendered walk: walk.opener, the re-based base's walk.lines — whatever the base now carries from RevenueCat walks there, with its window — then not testable soon: what RevenueCat has not observed yet (retained periods past the window, trial cohorts younger than trial days + 3).

Stops

• Not connected (no RevenueCat read tool is available in this session): print claude mcp add --transport http revenuecat https://mcp.revenuecat.ai/mcp as THE step to run to connect it, ask for no key, and change nothing in the scenario's inputs; the one write is the onboarding base link's measuredOn (references/stages.md), said in one line; nothing is recorded — SKILL.md § Files. Stop there until the reader has connected it.
• A RevenueCat tool that writes (create-, start-, publish-) is never called.