Skip to content
See all connectors at once

See all connectors at once

The test-estate safety rule: stand the test estate up only beside a test instance, never a production deployment. Its systems are throwaway test services with published credentials: change-me is every test credential. They are still encrypted and still ask for a login, because FACE refuses a connection that is not (see Connector security). Use a scratch machine or a scratch cluster.

This is the end-to-end walkthrough: one company in every connector, and FACE making sense of all of them together. Every source below is a view of the same company, Quillmere Freight Lines (quillmere.invalid). A shipment in PostgreSQL is the same shipment in the Kafka events, the GPS track, the SAP delivery, the Salesforce case, the claim PDF, the dock-camera frame and the RFID read. Six anomalies are planted across the sources, and each one can only be seen by reading more than one of them.

The universe

As of 2026-09-30, at seed 20261005. People appear only as roles. Every host ends in .invalid, every address is in a documentation range, and the only credential value is change-me.

EntityCountExampleSources that carry it
shipment24SHP-50001bigquery, d365, databricks, document, excel, files, github, guidewire, iot, kafka, logisticsapi, mqtt, mysql, objectstore, postgres, redshift, salesforce, sap, servicenow, sharepoint, tms, whs, yms
carrier4CAR-01d365, document, files, guidewire, kafka, mysql, objectstore, postgres, redshift, sharepoint, tms, yms
device42GPS-101bigquery, cctv, databricks, files, gps, iot, kafka, mqtt, postgres, rfid, servicenow
trailer10TRL-101cctv, databricks, document, excel, iot, kafka, mqtt, postgres, servicenow, tms, yms
customer6CUST-0001files, hubspot, kafka, oms, postgres, redshift, salesforce, sap, snowflake
dock9DOCK-NORTH-01cctv, excel, files, kafka, logisticsapi, postgres, sensorgateway, whs, yms
order24SO-30001document, files, hubspot, kafka, oms, postgres, sap, snowflake
sku18SKU-10001ims, inventory, logisticsapi, oms, postgres, sap, snowflake
customer_po23PO-0002-4103files, hubspot, kafka, oms, postgres, snowflake
warehouse3WH-NORTHinventory, oms, postgres, redshift, tms, whs, yms
invoice24INV-90001d365, files, kafka, mysql, objectstore, redshift
claim4CLM-2026-0101files, guidewire, salesforce, sharepoint
role6ROLE-DISPATCHservicenow, sharepoint, whs, yms
asn24ASN-70001logisticsapi, postgres, whs
epc48303400000051AE0000000001logisticsapi, rfid
yard3YARD-NORTHpostgres, yms
load24LD-610001tms
sap_delivery2480000101sap
sap_vendor40000200100mysql

There are 23 distinct customer POs for 24 orders. That difference is one of the planted anomalies. universe.json has every entity, every link between sources, and every anomaly with its evidence.

The walkthrough

Connect the sources in this order. Each step adds a kind of source, and each step has something to show.

Connect the two databases FACE reads end to end

Load postgres and mysql into scratch engines and add both connections. Explore returns their tables, columns, types, primary keys and declared foreign keys: the operational system (customers, orders, shipments) and freight billing (carriers, rate contracts, invoices). These two are the ones a test checks against the expected model, so start here.

Add the event bus and the telemetry

Stand up kafka, or the live Kafka estate for a stream that does not end, and use the recordings for mqtt, iot, gps, rfid, sensorgateway and cctv. Kafka shows its topic list and value schemas. Its Schema Registry and ksqlDB add which business domain each topic’s schema belongs to, which topic a query computes from which, and which topics carry no traffic. The telemetry sources have no catalogue: FACE collects for a window and infers each feed’s shape, and the reader location, the GPS asset_ids and the gateway channels settings tie readings to docks and trailers.

Add the business apps and the systems of record

Add salesforce, sap, guidewire, servicenow, hubspot, d365, sharepoint and github, then the logistics systems: oms, tms, whs, yms, ims, inventory and logisticsapi. Each page says what FACE reads for that type. Where FACE reads no catalogue, open the payload next to its expected model: the shared keys are the same strings.

Add the files

The files dataset lays the estate’s files out by the store each would live in: claim and invoice PDFs on S3, dock photos on GCS, dock video on Azure, the carrier rate sheet over SFTP, the scorecard on SharePoint, and an orders export on the runner’s disk. The PDFs are real PDF text, the images decode and the clips are MJPEG. Its page names the connection type that reads each store. excel, document and objectstore hold three more small datasets.

Add the cloud warehouses

snowflake, databricks, redshift and bigquery need a cloud account. Load their SQL into an account you own. Against a real account, FACE reads the Snowflake and Databricks catalogues; for Redshift and BigQuery, show the dataset next to its expected model.

Walk the six anomalies

Take them in the order below. Each one names the sources that hold its evidence, the file, and the literal string to look for, so every claim made on screen can be checked against the data.

The six planted anomalies

Each anomaly is the only one of its kind in the universe, and a test re-measures each one from the data. Files inside a zip are named by their path in that zip.

1. A duplicate customer PO (ANOM-4)

Customer PO PO-0002-4194 was accepted twice, as SO-30014 and SO-30015, with identical lines; both shipped.

SourceFileLook for
postgrespostgres/02_data.sql in postgres.zip'PO-0002-4194' (two order rows)
omsoms/v1_orders.json in oms.zip"customer_po": "PO-0002-4194"
hubspothubspot/crm_v3_objects_deals.json in hubspot.zip"customer_po": "PO-0002-4194"
kafkastreams/kafka/records.jsonl"customer_po":"PO-0002-4194"
kafkastreams/kafka/records.jsonl"customer_po":"PO-0002-4194","order_count":2 (the ksqlDB table ORDER_PO_DUPLICATES: its only row)
githubgithub/repos_quillmere_edi-integrations_issues.json in github.zipPO-0002-4194 (the EDI team’s own bug report)

2. An overcharged freight invoice (ANOM-1)

INV-90003 for SHP-50003 bills 3.84/mile against a contracted 3.25/mile (+15.8% on the total).

SourceFileLook for
mysqlmysql/02_data.sql in mysql.zip'INV-90003' (billed_rate_per_mile 3.84)
mysqlmysql/02_data.sql in mysql.zip'RC-CAR-01-2026' (contracted rate_per_mile 3.25)
filesfiles/s3/quillmere-claims/invoices/INV-90003.pdf in files.zipRate per mile: USD 3.84
filesfiles/sftp/outbound/carrier_rates_2026.tsv in files.zipCAR-01 ZQAL 3.25 (the carrier’s own rate sheet)
d365d365/data_VendorInvoiceHeaders.json in d365.zip"InvoiceAmount": 494.01 (approved for payment at the billed amount)
kafkastreams/kafka/records.jsonl"rate_per_mile":3.84,"contracted_rate_per_mile":3.25,"variance_pct":18.20 (the ksqlDB stream INVOICE_RATE_VARIANCE: a billed rate more than 5% over contract)

3. A claim near its filing deadline (ANOM-2)

CLM-2026-0101 (damage on SHP-50009) is still a draft; CAR-01’s 60-day filing window closes 2026-10-05, 5 days after the as-of date.

SourceFileLook for
guidewireguidewire/claim_v1_claims.json in guidewire.zip"carrierFilingDeadline_Ext": "2026-10-05" (state draft, filedWithCarrier_Ext false)
sharepointsharepoint/graph_sites_ops_lists_claims-tracker_items.json in sharepoint.zip"Title": "CLM-2026-0101"
filesfiles/s3/quillmere-claims/claims/CLM-2026-0101.pdf in files.zipFiled with carrier: NO - draft
mysqlmysql/02_data.sql in mysql.zip'RC-CAR-01-2026' (claim_window_days 60)
salesforcesalesforce/query_Case.json in salesforce.zip"Claim_Number__c": "CLM-2026-0101" (escalated case)

4. A reefer temperature excursion (ANOM-3)

SHP-50004 on reefer TRL-102 peaked at 11.4 °C against an 8.0 °C ceiling, after a mid-transit door opening; stock quarantined and a spoilage claim followed.

SourceFileLook for
databricksdatabricks/02_data.sql in databricks.zip'SHP-50004', TIMESTAMP (temp_c up to 11.40)
mqttstreams/mqtt/messages.jsonl"temp_c":11.40
kafkastreams/kafka/records.jsonl"shipment_id":"SHP-50004"
kafkastreams/kafka/records.jsonl"shipment_id":"SHP-50004","trailer_id":"TRL-102","breach_count":4,"peak_celsius":11.40 (the ksqlDB table REEFER_EXCURSIONS: its only shipment)
servicenowservicenow/api_now_table_incident.json in servicenow.zip"u_shipment": "SHP-50004" (reefer high-temperature alarm)
iotstreams/iot/messages.jsonl"device_id":"TAG-102" (door_open mid-transit)
inventoryinventory/api_stock-levels.json in inventory.zip"quarantined": (quarantine at WH-NORTH)

5. A GPS route deviation (ANOM-5)

SHP-50021 (trailer TRL-106) ran up to 36.4 km off its planned corridor for three consecutive pings.

SourceFileLook for
bigquerybigquery/02_data.sql in bigquery.zip'PG-000205' (first off-corridor ping)
bigquerybigquery/02_data.sql in bigquery.zip'SHP-50021' (the planned corridor)
gpsstreams/gps/nmea/GPS-106.nmea$GPRMC,104146.00,A,3812.2682,N,08953.6706,W,44.8,148.0,110926,,,A*74 (the off-corridor fix as the unit sent it)
tmstms/api_loads.json in tms.zip"shipment_ref": "SHP-50021" (planned route polyline)

6. An RFID read with no matching ASN (ANOM-6)

EPC 303400000051AEFF00000999 was read at DOCK-EAST-03 by RDR-EAST-03 and appears on no ASN; the dock camera has a frame of the moment and no shipment to attach it to.

SourceFileLook for
rfidstreams/rfid/tag_reports.jsonl303400000051AEFF00000999
logisticsapilogisticsapi/api_asns.json in logisticsapi.zip303400000051AEFF00000999 is absent: no ASN line carries it
cctvstreams/cctv/frames.jsonl"camera_id":"CAM-EAST-03" (frame at the read time, no trailer_id)

Which sources hold which evidence

SourceANOM-1ANOM-2ANOM-3ANOM-4ANOM-5ANOM-6
postgres●
mysql●●
databricks●
bigquery●
salesforce●
guidewire●
servicenow●
hubspot●
d365●
sharepoint●
github●
oms●
tms●
inventory●
logisticsapi●
files●●
kafka●●●
mqtt●
iot●
rfid●
gps●
cctv●

Sources not in this table hold no anomaly evidence. They are context: they make the shared keys resolve.

What the test estate shows, and what it does not

  • It shows every connection type’s native shape, the shared keys that join them, and evidence that only adds up across sources.
  • A test measures FACE’s discovery against the expected model for PostgreSQL and MySQL. For every other type, the expected model is the shape the source holds, and each quickstart says what FACE reads for that type.
  • It is not product data. The FACE backend reads none of the test-estate data at runtime.