Data honesty

Lokrix is a system of record for how brands appear in AI answers, so the data has to be trustworthy. Two commitments make that concrete: every result carries its provenance, and every probability is reported with its statistical uncertainty.

Provenance labels

Every data point is labelled with how it was produced, and the label is always visible in the UI:

• Live — captured from a real engine response during this scan. See Live vs predictive.
• Cached — reused from a recent valid live run rather than re-queried, to save credits.
• Mock (sample) — clearly-labelled fixture data used in dev and demos. Mock data is never presented as if it were live.

Statistical honesty

A single run is never reported as a probability. Lokrix always samples, shows the 95% confidence interval, and discloses the locale, persona and time context behind a measurement. Non-determinism is treated as a first-class concept, not smoothed away.

Every score is explainable

Scores come with feature-level attribution — why this result, and versus which competitors — and recommendations are simulated, not asserted. We only claim what we can show. For example, adding quotations, statistics and cited sources is a measured lever, while llms.txt has no measured effect and is never sold as one. See The levers.