SDR Reverse-Engineering
Automated generation and upkeep of an Adobe reference catalogue across several report suites. Algorithmic matching between Report Suites and Launch JSON rules, orphan detection, and regression analysis.
- Status
- prototype
- Year
- 2025
- Stack
- Python, LLM (Claude), Adobe Analytics, Adobe Launch
Why this project
Keeping an Adobe Analytics implementation’s documentation up to date quickly becomes unmanageable once several report suites run in parallel: every Launch rule change, every variable added or removed, has to be reflected in the reference catalogue. Rather than documenting by hand after the fact, the idea is to generate and maintain this catalogue automatically, straight from the source code.
Use cases
- Automated generation of the variable reference catalogue across several report suites
- Algorithmic matching between data actually collected per Report Suite and the Launch rules’ JSON code
- Detection of orphan variables (declared but never populated, or the reverse)
- Regression analysis between two versions of the catalogue, to isolate what changed and why
How it works
A Python script extracts the JSON code of Adobe Launch rules and matches it against the data observed on the Report Suite side, variable by variable. An LLM is brought in to formulate readable hypotheses for ambiguous cases (partial matches, inconsistent naming), while strict matching stays driven by explicit algorithmic rules rather than being fully delegated to the model.
What I took away
Keeping a reference catalogue current is a data-matching problem before it’s a text-generation one: the LLM helps interpret ambiguous cases, but orphan detection and regression analysis stay more reliable when built on explicit, verifiable matching logic.