01 / Section
A recorded sequence is not a complete procedure
A business outcome may take several steps across apps and sites. Shadow reconstructs evidence-linked attempts from captured batches so a reviewer can examine the sequence and compare it with other distinct attempts.
That record is useful, but it is still a record of what was captured. It may not explain why a person chose a step, what exception they were handling, or whether the same sequence is appropriate in a different case.
02 / Section
Separate evidence from interpretation
Keep three things distinct in a review: recorded evidence, interpretation, and planning assumptions. Evidence describes what appears in a captured attempt. Interpretation proposes what those steps may mean. Assumptions fill gaps about how a future workflow should behave.
A citation helps someone trace a claim back to its source. It proves provenance, not that the interpretation is correct or that the capture tells the whole story. Label uncertain explanations as questions to resolve, not facts to build on.
03 / Section
Compare attempts without flattening differences
Compare separate attempts to look for recurring workflow patterns, but keep meaningful differences visible. An apparent variation may reflect a different request, exception, or decision—not an inconsistency to remove.
App and site sequences can also point to places worth investigating. Treat those sequences as discovery signals, not proof that every visit belongs in one standard procedure.
04 / Section
Put a human decision before an automation plan
Before planning automation, ask the people responsible for the work to review the evidence and the proposed interpretation. Invite them to identify missing context, exceptions, and decisions that require judgment.
Then state the planning assumptions plainly: what cases the proposed workflow covers, what it leaves out, and what still needs validation. The result is a reviewable plan, not a claim that the recorded sequence is a complete or automatically executable process.
