Sipcode — Impact (the savings verifier)

Call the sipcode:verify_sipcode_impact MCP tool to produce a structured before/after comparison.

Critical: respect the integrity contract. The tool's response is structured with these key fields:

• status — one of: "measured", "insufficient-post-data", "no-install-marker", "no-baseline", "no-post-sessions", "window-asymmetry-<N>d-vs-<M>d"
• delta — present (numbers) ONLY when status === "measured". Null in every other case by design. When delta is null, DO NOT compute a savings number from before/after totals. The windows aren't comparable.
• allTime — present only when no marker exists. Contains the user's total session count across all time, so the user sees their data exists even when no A/B split is possible.
• warningReason — explains why delta is null (e.g., "window-asymmetry-39d-vs-2d", "insufficient-post-data-2d-vs-min-3d").
• headline — the canonical one-line summary; trust this and surface it.

When status === "measured": lead with the output-ratio change (the only normalization-resistant metric). Mention token + dollar savings as secondary. Include the window-length caveat.

When status !== "measured": present the headline and the next-step recommendation from the notes field. Do not invent a savings number. The tool's delta: null is structurally enforced — do not work around it.

If the user pushes back and asks you to compute a number anyway, refuse and explain the structural reason: the comparison windows aren't fair, so any delta would be misleading. Suggest they pass since: "YYYY-MM-DD" to set a manual pivot, or run sipcode rules --install to start measuring forward.