Confidence & quality

APEX produces two different "how much should I trust this?" numbers. They have different inputs, are computed at different times, and answer different questions. Confusing them is the single easiest mistake to make with APEX, so this page puts them side by side.

Estimate confidence

Returned on every /estimate response as confidence_score. It measures how well-backed this specific lookup is, from two signals about the matched cascade level:

volume    = min(observation_count / 100, 1.0)          // saturates at 100 observations
freshness = max(1 − data_freshness_hours / 720, 0.1)   // penalises data older than 30 days
base      = round((volume × 0.6 + freshness × 0.4) × 100) / 100
confidence_score = base × cascade_multiplier            // ×1.0 exact … ×0.6 country_pair

It is bucketed into a quality_status: HIGH (≥ 0.5), MEDIUM (0.3–0.5), LOW (< 0.3), or INSUFFICIENT_DATA when no level had enough data. It is computed live, per request, so it always reflects current data.

Corridor aggregate quality

Computed during the periodic corridor aggregation pass (not per request) and surfaced in analytics as a corridor-level quality/confidence figure. It measures how good a corridor's aggregate is, from three different signals:

quality = min(1.0,
    min(count / 20, 1)          × 0.4      // sample size
  + min(sources / 3, 1)         × 0.3      // source diversity
  + (1 − min(σ/μ, 0.5) / 0.5)   × 0.2      // price variance (lower is better)
  + 0.1)                                   // baseline

Side by side

The estimate confidence is the one to act on when pricing a lane. The aggregate quality is a corridor-health signal for analytics — it never appears in an estimate response. (An earlier version of these docs mistakenly presented the aggregate formula as the estimate's confidence; that is corrected here and in Estimate.)