Discount approval
Greater than 20% became greater than 30%.
The changed boundary allows discounts between those limits to bypass the intended Finance approval requirement.
← Back to selected work
Case study · Salesforce Change Impact Lab
When a valid change breaks the business rule.
A release gate that checks what Salesforce metadata means, alongside whether it is structurally valid.
The problem
A change can deploy successfully and pass static analysis while weakening a business control. A discount approval boundary can move; a flow can change from AND to OR; a permission set can grant authority it should withhold.
The lab asks a specific question: does the implementation still enforce the requirement that was approved?
The design decision
Freeze the business intent independently of the metadata. Select the relevant controls from the pull request, extract the implemented semantics, and compare them with an independent deterministic oracle.
The core reviewer path runs without Salesforce credentials. Conventional Salesforce checks remain useful supporting evidence.
Three controls, three deliberate regressions
Case escalation
High priority AND Strategic entitlement became OR.
A truth-table comparison exposes two outcomes that no longer match the frozen escalation rule.
Least-privilege access
Finance approval became editable.
A permission change allows sales reps to modify the approval field instead of only reading it.
The release evidence
Paired live pull requests demonstrated both sides: harmless governed changes received GO and merged; controlled regressions received NO-GO and remained closed and unmerged.
Recorded multi-control release candidate · PR #38
- Demo CI
- GREEN
- Salesforce Code Analyzer
- GREEN
- Business-intent gate
- NO-GO
One protected control passed. Another had two semantic mismatches. The aggregate release decision stayed NO-GO.
A passing supporting check cannot mask a semantic failure. Missing required evidence produces REVIEW.
What this establishes
The experiment demonstrates intent-aware release reasoning for three narrowly governed Salesforce controls. It does not establish universal metadata coverage or replace org-backed runtime testing.
The wider lesson is practical: quality evidence should explain whether a change preserves the business requirement, and make the reason for a release decision visible.
Let's talk
What are you trying to make more dependable?
Tell me about your system, your team, and the problem you need to solve. We can start with a conversation about where I can help.