Use cases

One platform, configured for the course in front of it.

BlueForge is a single product, but the failure modes that matter change by course. The presets below are sample shares by audience. Each one says plainly why a faculty member would reach for it.

Archetype · e-commerce

A multi-table e-commerce flow: customers, orders, and returns, with a slowly-changing customer segment and a return rate tied to order history. The share streams for the length of the term.

WhyA data engineering course needs a pipeline that survives late data, broken joins, and a schema that moves under it. This share has all three on a schedule the instructor controls.
Inject3% late arrivals on returns, delay drawn from exp(λ=90s)
Inject0.5% orphaned customer_id on orders
Inject2% MAR missingness on segment
DriftWeek 6: region encoding changes. Week 8: new column promo_code

returns / late-arrival view8 of 2,310 rows

return_id	order_id	event_time	ingest_time	reason
rtn_4a01	ord_88120	09:14:02	09:14:05	sizing
rtn_4a02	ord_88121	09:15:40	11:02:18 +107m	damaged
rtn_4a03	ord_88140	09:18:11	09:18:13	changed_mind
rtn_4a04	ord_91002 orphan	09:20:55	09:20:57	sizing
rtn_4a05	ord_88133	09:22:30	09:22:31	null
rtn_4a06	ord_88150	09:24:09	09:24:12	damaged
rtn_4a07	ord_88151	09:25:44	10:48:02 +82m	sizing
rtn_4a08	ord_88160	09:27:18	09:27:20	changed_mind

Archetype · marketing and web analytics

A marketing analytics share: web sessions joined to campaign spend and conversions, with the kind of structure that makes a naive aggregate misleading.

WhyAn analytics course wants students to question an aggregate before they report it. This share has a composition effect built in, so the headline number and the segmented number disagree.
SetupChannel mix differs by week, by design
Inject4% MCAR missingness on attributed_revenue
InjectSlowly-changing campaign_tier on the campaign entity
DriftWeek 9: a new acquisition channel comes online

campaign_web / conversion by channelaggregate vs. segmented

channel	week	sessions	conversions	rate
paid_search	3	8,120	244	3.01%
organic	3	12,940	198	1.53%
email	3	3,210	141	4.39%
Blended, all channels
all	3	24,270	583	2.40% aggregate

Archetype · finance and fraud

A card-transaction share for fraud detection: heavily imbalanced, with a positive rate that drifts and a derived feature that quietly leaks the label.

WhyA machine learning course wants students to evaluate beyond accuracy and to catch leakage before they trust a model. This share is built to punish both mistakes.
SetupFraud at 0.4%, drifting to 1.3% by week 10
Trapmanual_review_flag is set after the label is known
InjectConcept drift: P(y | x) shifts after week 7
InjectHeavy-tailed amount destabilizes tree splits

card_fraud / class balance over time16-week run

week	txns	fraud	rate	notes
1	42,100	168	0.40%	baseline
4	43,880	281	0.64%	ramp
7	44,210	486	1.10%	concept drift
10	45,020	585	1.30%	concept drift

Archetype · operations telemetry

A general-purpose share for courses outside the three core archetypes: a facility telemetry stream with gaps, drift, and outliers. Useful for statistics, visualization, and operations classes.

WhyNot every course is data engineering, analytics, or machine learning. A configurable telemetry share covers regression diagnostics, charting decisions, and operational monitoring from one base scenario.
SetupSensor drift on a subset of devices after week 11
Inject2% dropped readings, gaps rather than nulls
Inject1% data-entry outliers mixed with a legitimate tail
SetupBimodal load that should not be summarized by a mean

facility_sensors / reading sample6 of 18,400 rows

device_id	ts	temp_c	load	status
dev_07	02:00	21.4	0.62	ok
dev_07	02:05	21.5	0.64	ok
dev_07	02:10	null	0.63	gap dropped
dev_12	02:10	88.1	0.71	ok outlier
dev_12	02:15	22.0	0.70	ok
dev_19	02:15	23.8	0.81	ok

Try it live

See the kind of rows a student receives.

This pulls a small sample from a real demo share, no account needed. It is the lowest-friction way to confirm the data is real and behaves the way the presets describe.

static sample · live endpoint pending

web_events / demo share, first 10 rows10 rows

event_id	event_time	ingest_time	user_id	action	region
evt_8af2c1	14:02:18	14:02:19	u_4812	view	UT
evt_8af2c2	14:02:19	14:02:20	u_2207	click	CA
evt_8af2c3	14:02:20	14:48:51	u_4812	view	UT
evt_8af2c4	14:02:21	14:02:21	u_9913	view	NV
evt_8af2c5	14:02:22	14:02:23	u_2207	add_cart	CA
evt_8af2c6	14:02:23	14:02:24	u_4471	view	null
evt_8af2c7	14:02:24	14:02:25	u_4812	click	UT
evt_8af2c8	14:02:25	14:02:25	u_1180	view	TX
evt_8af2c9	14:02:26	14:02:27	u_3320	view	AZ
evt_8af2ca	14:02:27	14:02:28	u_2207	checkout	CA

This demo calls a real student share endpoint when one is configured. The endpoint is not connected for this build, so the table above shows a fixed sample of the kind of rows a student would receive. The shape is the contract: one row per event, with an event time and a separate ingest time so late arrivals are visible in the data itself.