Busiest airports
Freight handled in the latest month held, with change against the same month a year earlier.
loading…
What this is. Cargo tonnage handled by Indian and
international airports, assembled from published statistics. Every
number below was read from a source document and can be traced back
to it, nothing here is estimated or generated.
Airline cargo share
Total freight carried per airline across all periods held. Industry-wide totals are excluded so carriers are not double-counted.
loading…
One airport, in full. Monthly history, where the
statistical model expects it to go next, and anything flagged as
unusual. The shaded band is the range the forecast considers likely.
Select an airport
Monthly freight tonnage with a three-month projection.
loading…
More flights, or fuller flights?
Freight is flights multiplied by the tonnes each one carries, so a year’s change splits exactly between the two. They need different responses: more flights needs slots and stands, fuller flights needs handling and warehousing.
…
Reading this airport
Plain-language summary of what the series shows.
…
Flagged months
Months where observed traffic departed from the seasonal pattern.
…
What drives the number. Tonnage says how much moved.
These say how well, and where from. Cargo load factor is
freight tonne-kilometres over the capacity offered: it separates
growth that came from flying more from growth that came from filling
what was already flying.
What drove the change
loading…
Cargo intensity by airport
How much freight each departure actually lifted, and how much rode alongside each passenger. Volume and intensity are different things: the busiest airport is not always the most cargo-dense.
loading…
Market concentration
How evenly freight is spread across airports, month by month.
…
Freighter or belly hold?
Whether a carrier’s freight moves with its passenger flying.
…
Carrier efficiency
Load factor, tonnes lifted per departure, and mail’s share of what was carried.
loading…
How these are produced. Each series is fitted with
SARIMA, a statistical model for seasonal time series, and checked
by rolling-origin backtesting: the model is repeatedly asked
to predict a month it was not shown, and the average error of those
attempts is reported as MAPE. A 6% MAPE means past predictions were
typically within 6% of what actually happened. Lower is better; a
forecast without an error figure is a guess.
Projections
Predicted freight with an 80% interval: the range the model expects the true value to fall in four times out of five.
loading…
What counts as an alert. Seasonality is removed from
each series, then months whose remainder is far from typical are
flagged. Expected is what the seasonal pattern implied;
observed is what was published. A large gap is not
necessarily an error, a new terminal opening looks the same as a
data problem, and both are worth a human look.
Flagged movements
loading…
Written explanations
Narrated from the flagged figures. Every number is checked against a stored row before publishing; anything that does not resolve is rejected rather than shown.
loading…
Ask the data
Questions are mapped onto registered metrics and answered by SQL. The model never calculates a figure: it reads rows it was handed.
The monthly brief. A standing report generated from
the same figures shown elsewhere, written for someone who wants the
month summarised rather than explored.
Open standalone ↗
loading…
Why this exists. The data is assembled by agents that
choose their own steps rather than following a fixed script. Each run
records the goal, every tool it called, the reasoning behind the
choice and what came back: so a figure can be audited back through
the decisions that produced it. A tool called more often than the runs
that used it is one an agent had to retry.
Decision replay
Recorded trace, replayed at reading pace.
Tool usage
Retries are the loop recovering.
…
Who decided
Per-step policy provenance.
…
Recorded runs
loading…
Where every figure comes from. Each fact carries a
foreign key to the document it was parsed from, enforced by the
database rather than by convention: a fact without a source cannot
be stored at all.
Publishers
Documents retrieved and facts extracted from each.
…
Passage index
Source text indexed so a citation can point at a paragraph, not a whole PDF.
…
Search the source documents
Dense vector similarity and keyword search, fused. Returns the passage, never a computed number.
System health
Last pipeline run, stage by stage, and the counters exposed for monitoring.
…