Methodology

How Golf Sim Data measures demand

One written rulebook for every venue; a changed rule is a new model version.

Polling

Every bay block is read every 5 minutes (288 times a day) from two days before its date until it starts, and hourly three to nine days ahead. Each studio is read by one collector only, never twice.
Every read is the booking page’s own public availability response, archived verbatim with its time; failed reads are kept.
No logins and no private accounts: only what any visitor to the studio’s booking page can see.

A slot is one bookable start time with a capacity counter, in the venue’s own unit (an admission, a seat, a room block).
Unavailable units U = capacity − remaining. A rise in U between polls within 20 minutes, at equal capacity, is a sale; a fall is a reopen; anything else is ambiguous.
Units gone at the first poll count toward the final state, not observed sales.
A final state counts only if seen within 20 minutes of the start; otherwise sold units are a modeled range up to capacity.

Utilization: Σ sold ÷ Σ capacity at entry, with coverage (share of capacity whose final state was seen).
Revenue (GBV): units sold × sourced prices, packages and memberships, as a log-normal estimate. Prices are never backdated.
Every estimate carries an 80 % interval (P10–P90) and an A–D grade for its width.
Markets: venue revenues add as independent estimates; utilization is unit-hour weighted; unlike units never add.
Next-day forecast: current bookings plus comparable slots’ pickup from the same lead time, withheld unless 80 % of open capacity has comparables.

Unknown is never zero: missing or conflicting observations widen intervals and lower coverage.
Raw is truth: responses are stored verbatim by SHA-256 hash and never edited; every figure is rebuilt from them.
Models are versioned and proven against each venue’s raw data; released versions are frozen.