Cadence and sampling
How often a prompt runs, how many times per cycle, and what that buys you.
Each tracked prompt runs three times per engine per cycle, from clean sessions, on a weekly cadence by default. Those two numbers set both the statistical character of the output and most of the cost of serving it.
Why three, not one
Because one run is not a measurement. Identical prompts to the same engine on the same day return substantially different answers — day-to-day overlap in *cited sources* for the same prompt sits around 34–42%. A single-shot tracker is reporting one sample from a distribution and calling it a score.
Why not thirty
Because per-prompt stability is expensive and mostly unnecessary. Stabilising a single prompt's estimate to a standard error under 0.10 takes roughly seven runs a day; nobody's budget survives that across fifty prompts and six engines. What the contract actually delivers is the aggregate: fifty prompts at three samples is about 150 observations per engine per week, which supports an honest share-of-voice band of roughly ±7–8 points.
So the answer is not a bigger N. It is reporting the aggregate as a band, adding breadth through paraphrases and languages, and comparing fortnight over fortnight rather than day over day.
Daily cadence
Available as a premium lever on higher tiers, and worth it for a launch, a crisis, or a live campaign. It costs roughly seven times as much to serve, which is why it is priced as a lever rather than folded into the base — a bundle that hides a 7× cost driver is a bundle that gets withdrawn later.
Most cells are not volatile at all: roughly three-quarters of brand × prompt × engine combinations are near-deterministic — the brand is always mentioned, or never. Volatility concentrates in a contested middle band, which is where extra sampling eventually gets spent.
See it against your own client list
A working demo runs your prompts, in your market, on live engines — not a sandbox with seeded data. Bring one client brand and three competitors.