Debugger - Synthreo Pylon
Inspect a Pylon DAG run in the test console - review Steps, Problems, Output and Request, check node payloads, and diagnose variable and retrieval issues.
What you’ll accomplish
Section titled “What you’ll accomplish”Inspect a Pylon DAG test run to see the data nodes received and produced, examine retrieval evidence, and investigate unexpected results.
What you’ll need
Section titled “What you’ll need”- A DAG you can test.
- Representative input and an expected outcome.
- The variable reference for the nodes you are inspecting.
1. Run a representative test
Section titled “1. Run a representative test”Open the DAG, supply the test input and select Run test. Use Show test console if the console is hidden.
You’ll see the run trace and its Steps, Problems, Output and Request tabs.
2. Inspect the node trace
Section titled “2. Inspect the node trace”Open Steps and inspect the entries recorded for the run. Compare each relevant node’s Input and Output with the data you expected.
You’ll see the payloads available for those trace entries. Compare the trace with the canvas rather than assuming every canvas node must appear.
3. Inspect retrieval and problems
Section titled “3. Inspect retrieval and problems”For an LLM node using retrieval, look for RAG selection items. Check whether the retrieved material contains the information needed to answer the test question. Review Problems, then compare the final Output with the Request.
You’ll have evidence separating input, retrieval and response problems instead of diagnosing from the final answer alone.
If it didn’t work
Section titled “If it didn’t work”A node is absent from Steps
Section titled “A node is absent from Steps”Check its connections, the upstream output and the conditions leading to its branch. An unmet branch can be absent from the trace. Absence alone does not establish a particular engine defect or guarantee that a skipped-node reason will be shown.
A variable produces an unexpected result
Section titled “A variable produces an unexpected result”Compare the reference against the upstream Result property name character by character and inspect the actual output shape. The name after Out. is an output column, not a node ID.
Shipped examples use bare references in node configuration and Out.-prefixed references in terminal output templates. Runtime case sensitivity and unresolved-reference rendering are not established by the verified source; do not diagnose a specific cause solely from blank or literal-looking output.
Tool Eval receives no tool-call data
Section titled “Tool Eval receives no tool-call data”Check the upstream LLM node’s Omit full response setting. It defaults on; switch it off when Tool Eval needs the full response. Inspect which function the model selected and compare that with the Tool Eval configuration.
Retrieved material is missing or irrelevant
Section titled “Retrieved material is missing or irrelevant”Inspect RAG selection items, confirm the source material contains the answer, and review Knowledge & Retrieval on the LLM node. Test changes to Minimum relevance, Results to use and data structure against the same representative questions.
An email or output template behaves unexpectedly
Section titled “An email or output template behaves unexpectedly”Check the consuming node’s template engine and the upstream data. Email supports Mustache template. Output Data offers Synthreo String Compiler; its Mustache choice is disabled. Do not transfer triple-brace examples between these engines or assume a particular unresolved-reference symptom.
Historical interface
Section titled “Historical interface”The image below shows the retired Test Results → Debugger interface with two tables. It is retained to help identify older screenshots; use the current test-console procedure above.
