The formula
A GBI of 20 means each $100 buys 20 reliable benchmark points. Higher is better. The reciprocal, cost per point, is also shown on each watch page for readers who prefer a “lower is cheaper” view.
Confidence adjustment
A raw mean from a handful of watches should not beat a mature result by accident. Version 1 uses a deliberately transparent penalty:
reliable score = average score × (0.70 + 0.30 × confidence / 100)
Confidence is stored as a percentage from 0 to 100 and is supplied by the benchmark pipeline. The raw mean, sample count, confidence, and adjusted score remain visible so the ranking can be audited.
Eligibility
- At least five independent benchmark samples.
- Confidence of 70% or higher for a formal rank.
- An exact, currently sold SKU mapping.
- An in-stock USD price checked within the last 48 hours.
Lower-confidence results may appear as provisional, but they do not receive a formal rank. Discontinued models stay available as historical records.
The model directory includes every valid group from the latest membership-ready WristBench synchronization. Until a full refresh succeeds, the directory is explicitly marked incomplete. A missing or ambiguous retail mapping leaves price and GBI blank; it does not hide the benchmark result.
Price policy
The US table uses the public, in-stock Garmin US price for the mapped base SKU. Member-only discounts, trade-ins, bundles, refurbished units, tax, and shipping are excluded. Sale and list prices are stored separately in the price pipeline.
USD and CNY offers are never mixed. A future JD ranking will be a separate China market view because regional catalogues, warranty, tax, and availability are materially different.
Interpretation and limits
GBI answers a performance-value question for the WristBench Connect IQ workload. It does not tell you which display, case, sport features, maps, battery, fit, or design is best for you. A lower-ranked watch can still be the right purchase.
Prices move daily and the benchmark population grows over time, so rank changes are expected. Every published snapshot records its inputs and algorithm version.