agenthub · transform.radar_data
Radar Data
Reshape several scored LLM outputs into the series a radar chart needs.
Why the node earns its place
The adapter between assessment and picture. A scoring workflow typically runs one LLM per dimension; each returns a score and some supporting detail. This node pulls the score out of each and turns the set into axes.
It is tolerant about what it is fed on purpose: JSON strings are parsed, and assistant message objects are unwrapped, so the outputs of several LLM nodes can be wired straight in.
How it works at run time
- 1Normalise every input — parse JSON strings, unwrap message objects.
- 2For each source that carries
score_key, add an axis: the source name as the dimension, the score, andmax_scoreas the full mark. - 3Collect the supporting detail alongside — summary, dimension scores, strengths and risks, each read under several alternative key names.
- 4Emit chart data ready for Radar Chart, plus the same thing as text.
What you wire
Inputs
- data *object · accepts several edges
Outputs
- chart_dataobject
- textstring
Configuration
| Key | Type | Default & options | What it does |
|---|---|---|---|
| score_key | string | default "final_score" | Key to extract the numeric score from each section |
| max_score | number | default 10 | Maximum value for the radar axes |
What usually goes wrong
Watch for this
A source that does not carry the configured score_key is skipped silently — it simply does not become an axis. A chart with fewer axes than you expected is a key-name mismatch, not a scoring failure; check what the LLM actually returned before changing the prompt.
Any _LLM suffix is stripped from the axis label, so name the source nodes for how they should read on the chart.
Behaviour & provenance