Course 3.6 · Curriculum 2026.1
Validating model output before it touches a record
Allow-listed enumerations, range-checked scores, capped arrays, and dropping what fails rather than storing it. Plus the provenance rule that matters most.
- Estimated effort
- 4 hr
- Lessons
- 34
- Modules
- 6
- Material
- 4 hr
Estimated effort is the sum of the lesson and lab times in the course outline, for this version of the course.
This course is planned and not on sale. The curriculum below is the plan, and it may change before the course opens.
See the courses on sale01What you will learn
7 capabilities, stated as things you can do
Each one is what you should be able to do at the end, not what the course covers. If an outcome is not testable, it is not an outcome.
- Validate every field of a model response against an allow list, a range or a length cap.
- Drop a failing value rather than storing it with a warning nobody reads.
- Cap arrays so one response cannot create four hundred child records.
- Label provenance on every written value, and default to the weaker label.
- Explain why an unlabelled claim is the dangerous one.
- Design a review queue for values that failed validation but might be right.
- Write the test that proves an invalid response never reaches the database.
02Curriculum
6 modules, 34 lessons
Every lesson shows its length and its type, from the course outline.
6 modules · 34 lessons · 4 hr of material
Between the parse and the DML. Everything that arrives is a proposal, and a proposal is not a value.
- VideoAllow-listed enumerations, and no free-text statusPreview6 min
- VideoRange-checked scores, and the confidence that was 1.46 min
- VideoCapped arrays, and the response that wanted 400 children6 min
- VideoLength-capped strings, and the field that silently truncated6 min
- ReadingEverything that arrives is a proposal, not a value5 min
- LabLab: build the validator15 min
An agent looping single updates wrote a misspelled picklist value and blanked stages across records. Fixed with validation, batching and a backstop rule.
- VideoAn agent looping single updates, one record at a time6 min
- VideoA misspelled picklist value, and what it did to every stage it touched6 min
- VideoServer-side validation against the live schema6 min
- VideoBatched writes, so one bad value cannot cascade6 min
- ReadingA validation rule as the backstop nothing else caught5 min
The rule that matters most: an unlabelled claim is the dangerous one, so default to the weaker provenance.
- VideoAn unlabelled claim is the dangerous oneA value with no source reads as fact. Label every written value, and default to the weaker label when unsure.6 min
- VideoStoring provenance beside the value, not in a log6 min
- VideoShowing provenance to the person reading the record6 min
- VideoDefaulting to the weaker claim when the source is ambiguous6 min
- ReadingGeneral model knowledge against internal data: telling them apart5 min
- LabLab: add provenance to three written fields15 min
The enum-constrained single write in the field triage agent, and the ritual that resolves, maps, states and echoes before it commits.
- VideoThe single write, enum-constrained server-side6 min
- VideoAn endpoint that rejects anything off-enum, so a wording slip cannot corrupt it6 min
- VideoResolving the exact record before anything is written6 min
- VideoMapping the user’s natural phrasing to the enum, and stating the mapping6 min
- VideoEchoing it back and waiting for a yes5 min
- LabLab: build the confirmation ritual end to end10 min
A remediation decision costs six figures. The engine only nominates candidates, and a data gap must never read as confirmed.
- VideoData gaps read amber, never silently green6 min
- VideoWhy a six-figure decision cannot rest on a candidate flag6 min
- VideoThe null hypothesis: assume the model is wrong until proven6 min
- ReadingA human verification queue as the calibration signal5 min
- LabLab: design the amber state for a missing field14 min
Dropping is the default. A review queue is the exception, and it needs an owner or it becomes a graveyard.
- VideoDrop, do not store with a warning6 min
- VideoA review queue with an owner and a decay policy6 min
- LabThe test that proves an invalid response never reaches the database15 min
- VideoWhy a swallowed validation error is worse than a loud one5 min
- ReadingWriting the validator’s test cases before the validator5 min
- QuizCheck: output validation10 min
04Before you start
What you need first
Assumed knowledge and setup
- Course 3.4. This course starts where the parsing lesson there stopped.
- A Developer Edition org.
- Course 3.5 is recommended, because injection and validation are the two halves of the same boundary.
Before this course
Courses often taken next
- 3.11 · Agentforce Specialist exam preparation
- 3.12 · Call intelligence, and what not to log
- 3.13 · Vision, documents, and the confidently wrong answer
Each is a separate course, bought on its own.
05Who teaches it
Elliot Saha
Co-founder, Chief Technology Officer
Platform architecture, custom Apex and Lightning Web Components, AI systems, and the Academy platform itself.
06Reviews
No learner reviews yet
This course has not been taken by enough people to publish an honest rating, and we will not print invented quotes on a page that sells verification discipline. Here is what we can evidence instead.
- Assessment
- Lesson checks onlyThis course carries lesson checks and issues no certificate.
- Curriculum version
- 2026.1You enrol into a version. Content changes do not move you mid-course, and your required-lesson count is snapshotted at enrollment.
- Refunds
- 14 daysAn individual purchase can be refunded within 14 days if you have completed less than 20% of the lessons. Team seats are refundable before they are assigned. From the Academy terms.
Start 3.6
This course is in development and not yet open for enrolment. The courses open now are listed in the catalogue.
- 4 hr estimated effort, labs included
- A ShareCo certificate is not a Salesforce certification