Dispatch · delivery-promise error

One global speed leaves NORTHGATE 2.2× as late as RIVERSIDE, and the pooled average hides it.

generated 2026-10-09 18:21 UTC from the warehouse · every figure queried at build time

How late each store actually runs

Actual delivery time minus the time that was promised, averaged per trip. Longer is worse. The dashed line is the pooled average, the single number a report would quote.

NORTHGATE RIVERSIDE pooled +7.79 min

pooled +7.79

NORTHGATE +10.16 min late

RIVERSIDE +4.61 min late

5.55 minutes apart. The pooled average sits between the two and describes neither: it reports a formula that is roughly +7.79 minutes optimistic, and hides that it is 2.2 times as wrong at one store as at the other.

Synthetic data. The events come from a seeded generator and defects are injected on purpose. The gap measures the store speeds written into that generator, not a real fleet. What this shows is that the pipeline finds a per-store difference that a single pooled figure would report as one problem.

Breach rate by day

NORTHGATE breached more often than RIVERSIDE on 14 of 14 days, so the gap holds from day to day rather than coming from one bad afternoon. Each column carries both figures on hover or keyboard focus; the end of each line is labelled regardless.

100%

50%

NORTHGATE 82.5% RIVERSIDE 66.0%

Sep 01NORTHGATE 81.3%
RIVERSIDE 61.9%

Sep 02NORTHGATE 83.8%
RIVERSIDE 65.8%

Sep 03NORTHGATE 83.2%
RIVERSIDE 69.4%

Sep 04NORTHGATE 81.4%
RIVERSIDE 69.1%

Sep 05NORTHGATE 81.1%
RIVERSIDE 66.8%

Sep 06NORTHGATE 82.2%
RIVERSIDE 66.3%

Sep 07NORTHGATE 81.8%
RIVERSIDE 63.2%

Sep 08NORTHGATE 80.3%
RIVERSIDE 67.9%

Sep 09NORTHGATE 83.3%
RIVERSIDE 68.7%

Sep 10NORTHGATE 85.7%
RIVERSIDE 65.7%

Sep 11NORTHGATE 81.4%
RIVERSIDE 67.8%

Sep 12NORTHGATE 81.5%
RIVERSIDE 65.4%

Sep 13NORTHGATE 85.1%
RIVERSIDE 66.7%

Sep 14NORTHGATE 82.5%
RIVERSIDE 66.0%

Sep 01Sep 14

Per store

Every rate carries the trips it was computed from. Click or press Enter on a column heading to sort.

Store	Trips	Mean signed	MAE	p50	p90	Breach
NORTHGATE	11,083	+10.16	11.20	+7.08	+25.22	82.5%
RIVERSIDE	8,229	+4.61	7.11	+2.85	+15.79	66.4%

Pipeline run

What the DAG moved on the run that produced this page.

125,264

events landed

5,854

collapsed in silver

20,000

trips

quarantined

Kafka is at-least-once, so redeliveries are expected. Collapsing them in silver is deduplication working, not data lost.

Trip completeness

99.5% of trips carry a trustworthy duration. The rest are missing an event, and are excluded rather than guessed at.

State	Trips	Share
complete	19,902	99.5%
broken	98	0.5%

Quality gates 9/9

Each gate runs before the load, not after it. A failure stops the DAG rather than annotating a warehouse that already has the bad rows in it.

Stage	Check	Verdict	Detail
bronze_to_silver	freshness	PASS	newest event is 1 day, 0:33:18.738184 old (limit 3 days, 0:00:00)
bronze_to_silver	reconciliation	PASS	125264 in = 125264 accepted + 0 quarantined
silver_to_gold	fk:driver_id	PASS	all driver_id values resolve in dim_driver
silver_to_gold	fk:store_id	PASS	all store_id values resolve in dim_store
silver_to_gold	measurable_share	PASS	19312/20000 trips measurable (96.6%)
silver_to_gold	not_null:trip_id	PASS	trip_id has no nulls across 20000 rows
silver_to_gold	row_conservation	PASS	fct_trip_x_dim_driver preserved all 20000 left rows
silver_to_gold	unique_key	PASS	trip_id unique across 20000 rows
silver_to_gold	volume_band	PASS	only 0 baseline day(s) - band not evaluated