How APEX works

APEX turns a stream of raw freight prices into trustworthy market percentiles. This page explains the pipeline conceptually: where the data comes from, how it is normalized, and how an estimate is produced. It deliberately does not cover where the service runs — see APEX › Reference for operational and API details.

Observations

The atomic input is a price observation: a single recorded freight price for a specific origin, destination, distance, and date, from a specific source. APEX ingests observations from European freight data — both spot-market postings and executed-freight prices (actual booked rates). Executed prices are the stronger signal and are weighted more heavily than spot postings.

Each observation is normalized and quality-tagged on the way in: geocoded to consistent locations, distance-enriched where missing, and marked with a quality score reflecting whether its price was directly observed or inferred from a broader median.

Corridors

Observations are aggregated into corridors — directed origin→destination lanes. "Directed" matters: PL → DE and DE → PL are separate corridors with separate prices, because freight markets are rarely symmetric (a lane can be well-paid one way and cheap the other).

A corridor accumulates the observations that fall on its lane. Percentiles (p25 / median / p75) are computed across those observations, filtered to exclude zero-price and outlier rows.

Source weighting and trust

Not all sources are equal. APEX weights observations by source trust so that stronger signals dominate the percentiles — executed-freight sources count for more than spot postings. Each observation also carries a quality score: highest when the price was directly observed, lower when it was filled from a corridor or country median during enrichment. Low-quality and outlier rows are excluded from the percentiles that back an estimate.

Serving an estimate

When you request an estimate, APEX computes the percentiles live from the observation data for the matched lane, resolving through the cascade (exact → reverse → postal prefix → country pair) until it finds a level with enough data. The estimated total price is the median per-km rate times the requested distance.

The estimate's confidence reflects two things: how much data backs the matched lane (volume) and how recent that data is (freshness), scaled down by how far the cascade had to broaden. A lane with many recent exact observations scores HIGH; a thin or stale lane scores lower; a lane with no usable data returns INSUFFICIENT_DATA rather than a guess. The precise formula and thresholds are documented in APEX › Reference.

APEX also computes a separate, richer quality assessment during periodic corridor aggregation (based on sample size, source diversity, and price variance). That aggregate score is used for corridor-level analytics — it is not the same number as an estimate's confidence. The two are documented side by side in the reference so they are never confused.

The data model, in brief

• Price observations — the raw, normalized inputs.
• Corridors — directed lanes, with pre-aggregated statistics for analytics.
• Locations and postal ranges — the geography observations are matched against.
• API keys and usage — access control and metering for the API.

Field-level detail lives in APEX › Reference. How these numbers feed a carrier's bid is covered in Using APEX from TOVR.