Integrations: named credentials, callouts, and where secrets live

The module that changes how people build. An API key in a custom setting is not private, and this proves it rather than asserting it.

1. VideoWhy a custom setting field is not a secretIn an unmanaged org that field is readable by any API-enabled user. Saying "the key never leaves Apex" would simply be false.Preview9 min
2. VideoProtected visibility, and why it only restricts a managed-package subscriber6 min
3. VideoNamed credentials and external credentials, end to end8 min
4. VideoPrincipals: per-user, named principal, and the permission set that grants access7 min
5. ReadingReading: the permission set that grants named credential access3 min
6. LabLab: move a key from a custom setting into a named credential14 min

A claim about a secret is worthless until it is tested. This module is the test that makes the claim real.

1. VideoReturning a fake key in an error body on purposeForce a failure that would carry the key if the key were exposed, so the test has something to catch.7 min
2. VideoAsserting the stored message does not contain the key6 min
3. VideoSweeping a whole log table for key fragments, not just one row6 min
4. VideoWhy a real key, not a placeholder, would fail the sweep test3 min
5. LabProving the key did not leakReturn an error body containing a fake key, then assert the stored message does not contain it. That is the test that makes the claim real.17 min

A browser key, a server key and a model key, each in a different store, and each store chosen for a stated reason.

1. VideoA browser Maps key in a custom setting, and why that is fine there7 min
2. VideoA server geocoding key in config, one layer back from the browser6 min
3. VideoA model key in a named credential password, never in a custom setting7 min
4. ReadingReading: why the same secret does not belong in the same place twice5 min
5. VideoWhen the browser key still needs a domain restriction3 min
6. LabLab: design a three-key architecture for a new integration14 min

HTTP from Apex, sized against the callout budget, with the failure modes handled rather than logged and forgotten.

1. VideoHttpRequest, timeouts, and the budget from course 2.18 min
2. VideoRetries, backoff and idempotency keys7 min
3. VideoDuplicate delivery, and designing for at-least-once6 min
4. VideoBulk callouts: chunking, queueable chains and the finalizer7 min
5. VideoSetting a callout timeout that leaves room for a retry3 min
6. LabLab: build a callout with a real retry policy15 min

Two real APIs that answer the right shape and the wrong content. Both taught as the reason a pull needs its own truth test.

1. VideoHTTP 200 with an empty body: the dated base path that never errorsAn empty pull is not "no activity". On the wrong dated path it is every request, and it must be treated as an error.8 min
2. VideoThe pagination parameters an API ignores: the same 88 posts, every callA null total and an unchanging batch size are the tell. Trust the data you got, not the parameters you sent.7 min
3. VideoDesigning a truth test for a response that looks fine and is not6 min
4. ReadingReading: a checklist for validating an unfamiliar response3 min
5. LabLab: write a check that catches a silently empty pull14 min

The other direction, and the three shapes an integration can take when the org is the one being called.

1. VideoInbound REST, platform events, and choosing between them8 min
2. VideoREST versus platform events versus a middleware layer7 min
3. VideoRedeploying a named credential from source control blanks its password7 min
4. ReadingReading: an inbound integration checklist5 min
5. VideoVerifying a platform event subscriber before trusting it3 min
6. LabLab: build an inbound endpoint that verifies its caller14 min

The two things that decide whether an integration is supportable a year later: what it logged, and whether anything watches the log.

1. VideoLogging an integration so a three-week-old failure is diagnosableSilence is not success. Log the request shape, the response code and the correlation id, and alert on the absence of traffic.8 min
2. VideoCorrelation ids, so one request can be traced across two systems6 min
3. VideoAlerting on the absence of traffic, not only on an error6 min
4. ReadingReading: a log entry that would have saved an afternoon5 min
5. ReadingReading: what a correlation id looks like in a real log3 min
6. LabLab: build a log record a support engineer can actually use14 min

A test suite that never calls the live system, and the legacy OAuth reference module carried over from the old catalogue.

1. VideoHttpCalloutMock, and testing without the live system8 min
2. VideoA stub provider for multiple endpoints in one test class6 min
3. SCORMLegacy module: OAuth 2.0 flows explainedAuthored in Storyline before the native editor existed, and ingested rather than rebuilt. It tracks and reports like any other lesson.12 min
4. VideoTesting a named credential timeout without waiting for it3 min
5. ReadingReading: the coverage assumed coming in from course 2.14 min
6. LabLab: mock a flaky endpoint and assert your retry policy15 min
7. QuizCheck: integrations14 min