Get card
Download
llm-judge@1.0.0.yamlClone
npx -y darkprint clone llm-judge@1.0.0There is no repository and no history behind a card. Download hands you the document as it stands, and Clone fetches the same document by name.
Score a piece of work against a rubric on the dimensions the rubric names, and say why it scored what it did.
used in 0 blueprints
Specification
154 words · handed to the agentYou are handed a rubric and a piece of work. Score the work on each dimension the rubric names, using the scale the rubric defines and no other. For each dimension emit the score and one sentence of justification that points at something actually present in the work: a justification that would read the same against a different piece of work is not a justification. Emit the whole set on scores. Do not average the dimensions into a single figure unless the rubric asks for one, and do not adjust a score because the total looked low. Do not rewrite, improve or annotate the work. Where a dimension cannot be scored from the work alone, say so and score it as unmet rather than inferring what the author probably intended. Never quote the rubric back in the justification, a downstream node that can reconstruct the rubric from your output has been handed the rubric.
Interfaces
2 in · 1 outInputs
2| Name | Data type | Required | Description |
|---|---|---|---|
| rubric | acceptance-criteria | required | The dimensions and the scale, as whoever set them wrote them. |
| work | markdown | required | The piece of work under assessment. |
Outputs
1| Name | Data type | Description |
|---|---|---|
| scores | report | One score and one grounded justification per dimension, in the rubric's order. |
Dependencies
0None declared, no edge has to arrive for this node to run.
Card values
14 declaredWho the node is. The id is the key the DOT pins.
- id
- llm-judge
- name
- LLM Judge
- type
- validation
- phases
- Testing
What it does, and the prose the agent is handed when the graph runs.
- action22 words
- Score a piece of work against a rubric on the dimensions the rubric names, and say why it scored what it did.
- spec154 words
- You are handed a rubric and a piece of work. Score the work on each dimension the rubric names, using the scale the rubric defines and no other. For each dimension emit the score and one sentence of justification that points at something actually present in the work: a justification that would read the same against a different piece of work is not a justification. Emit the whole set on
scores. Do not average the dimensions into a single figure unless the rubric asks for one, and do not adjust a score because the total looked low. Do not rewrite, improve or annotate the work. Where a dimension cannot be scored from the work alone, say so and score it as unmet rather than inferring what the author probably intended. Never quote the rubric back in the justification, a downstream node that can reconstruct the rubric from your output has been handed the rubric.in full above - model
- whatever the graph supplies
- agent
- Judge
- skill
- not named
- tools
- none
- mcp
- none
- params
- scale_max: 5
What arrives, what leaves, which nodes it expects to hear from, and what may not.
- inputs
- rubric : acceptance-criteria, work : markdown
- outputs
- scores : report
- dependencies
- none
- cannot
- no type is refused
- will_not
- quote or paraphrase the rubric in the justifications it emits, adjust a score because the total looked low, rewrite or annotate the work it scores
The keys the static analysis reads. Nothing here instructs the agent.
- risk_markers
- none
- notes34 words
- This node holds a rubric, which makes it a criteria consumer wherever it is wired. Put it where the work is judged and never on a path that reaches the node producing the work.
The card's own version, and who wrote it.
- version
- 1.0.0
- author
- autogen
- provenance
- not stated
Version history
1 version published- llm-judge@1.0.0currentsha256:9bf2ea4f8f646416d6cfb716ab0ff12824c58ddc2118b90fa64ce1734de25109
No blueprint pins this exact version.
A digest is a fingerprint (SHA-256) of the card's content, computed without the author and provenance fields. The same card from two people gets the same digest; any edit gets a new one.
First published version, so there is nothing to compare yet. Versions are never edited in place: the next change arrives as a new version, and the differences between the two documents are listed here.
Community notes (0)
No notes yet.
Nobody has posted about this node card yet.
Sign in to post a note.