Get Overview

Where the property's remaining measured visibility gap is concentrated.

has_measured_gap is False at 100% measured visibility (total deficit 0); the lists are then empty and the UI renders a no-gap state — no gap shares are computed (no division by a zero total).

Recommendation Stability: how consistently repeated searches return the same result.

stability_rate is the majority-agreement rate in [0.5, 1.0] (rendered as a percentage), the unweighted mean over Stability-eligible cells of each cell's unweighted mean daily max(p, 1 - p) — the same equal-cell hierarchy as ai_visibility_rate, never a 0–100 score. None when no cell had ≥2 usable repeats on any measurement (insufficient repeated evidence — distinct from 100%, which needs eligible evidence with no variation). The *_cells / *_cell_measurements counts are factual support only (they never weight the score): eligible requires ≥2 usable repeats, measured requires ≥1. eligible_cell_measurements vs measured_cell_measurements exposes when a Stability value rests on few repeated measurements even though many cells were Visibility-measured.

Period-level confidence: technical coverage + per-cell temporal support.

Separate from model-output variability (which lives in the Stability Diagnosis): high variability is not low measurement quality.

The URL-provider source environment observed around the property this period.

concentration_topk_share is the top hosts' share of captured source appearances (Σ distinct-result-per-host, a result×host quantity) — never a measure of "what AI trusts". cross_provider_host_count counts hosts recurring across ≥2 assistants over the full eligible set (not just the shown top-k). hosts is a single ranked evidence list (top-k, presentation only).

How the property's observed source environment relates to the domains LLMs cite for its destination (Source Tracker).

Association only: Source Tracker observes which domains are cited for the destination — never that they are "trusted", and absence is not proof that gaining presence would improve AI visibility. Matching is exact normalized host (subdomains stay distinct); Gemini is excluded (reconstructed hosts aren't join-safe). snapshot_week is always present when available so an older weekly benchmark is never shown as contemporaneous with the whole period.

How much of the captured source activity pointed at the hotel's own domain.

Grain is the source occurrence (one GeoResultSource row), NOT the result and NOT exclusively the literal citation: these rows span answer_citation, search_result, grounding_source and unknown evidence types, so one response referencing the hotel's site three times contributes three. This is deliberately a different grain from the Sources tab's n/M, which are distinct-result counts.

Scoped to the Sources run population for the selected period (see _resolve_sources_run_ids) so the numbers reconcile with the Sources tab rather than with the surrounding Overview metrics.

Two distinct empty states, never collapsed: owned_domain is None means the property has no website_url configured; a real owned_domain with zero counts means the hotel IS configured and the selected period simply produced no source-captured runs.