Rubrics

Rubrics define the scoring criteria used to evaluate TA performance during office hour simulations. Each rubric organizes standards into groups, assigns point values, and sets passing thresholds — giving evaluators a structured framework to assess how well a TA handled a student interaction.

What is a Rubric?

A rubric in Glow is a hierarchical scoring instrument composed of:

• Standard Groups — categories of competency (e.g., “Communication Skills”, “Policy Knowledge”, “De-escalation”)
• Standards — individual criteria within each group (e.g., “Active Listening”, “Empathy”, “Clarity of Explanations”)
• Points — the total point value and passing threshold for the rubric

Rubrics connect to three parts of the platform:

1. Agents through the agent_rubrics_junction — defines what the agent is scored on
2. Simulations through simulation_scenario_rubrics_junction — attaches rubrics to specific scenarios
3. Evals through eval_model_rubrics_junction — applies rubrics during automated evaluation runs

How it Connects

Rubric
  |
  +-- Standard Groups
  |     +-- Standards (with point values)
  |
  +-- Agent (via agent_rubrics_junction)
  +-- Simulation Scenario (via simulation_scenario_rubrics_junction)
  +-- Eval (via eval_model_rubrics_junction)

• Agents are scored against rubrics attached through the junction table. The rubric defines what is measured; the agent’s behavior during the simulation provides the data.
• Simulation Scenarios can have per-scenario rubrics, allowing different scenarios to emphasize different competencies (e.g., a grade dispute scenario uses the “De-escalation” rubric while a conceptual question uses “Clarity of Explanations”).
• Evals reference rubrics through model_rubric_ids to apply scoring criteria during automated evaluation runs.

Create a rubric

Via the CLI

Calls below use $GLOW_INSTANCE_URL + $GLOW_TOKEN — see Authentication to export them once.

glow rubrics create --body '{
  "rubrics": [{
    "name": "Communication Skills",
    "description": "Evaluates TA ability to communicate clearly and empathetically during student interactions"
  }]
}'

Via the API

curl -X POST $GLOW_INSTANCE_URL/rubric/create \
  -H "Authorization: Bearer $GLOW_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "rubrics": [{
      "name": "De-escalation",
      "description": "Measures TA effectiveness at calming frustrated students and redirecting conversations productively"
    }]
  }'

Each entry returns the new rubric_id, a generated draft_id, and the initial version.

Standard groups and standards

A well-structured rubric organizes standards into logical groups. Here is an example for a TA training program:

Rubric: “Communication Skills” (Total: 30 points, Pass: 21 points / 70%)

Managing standards via API

Standards and standard groups are managed through their respective resource IDs on the rubric:

• standard_group_ids — attach standard groups to the rubric
• standard_ids — attach individual standards
• point_ids — configure point values and passing thresholds

The search response includes denormalized standard_groups and standards arrays alongside the rubric list, so you can display the full rubric structure without additional API calls.

Each standard group in the response includes:

Each standard includes:

Points and passing thresholds

Rubrics define scoring at multiple levels. The rubric list response includes points, pass_points, and pass_percentage for each rubric.

# Search rubrics and inspect scoring
glow rubrics search

The response for each rubric entry includes:

The draft cycle

Rubrics support a draft workflow for staging changes. This is especially important because modifying a rubric can affect scoring across active simulations and evaluations.

# Create or update a rubric draft
glow rubrics draft --body '{
  "name": "Communication Skills v2",
  "description": "Added empathy standards for improved scoring",
  "standard_group_ids": ["ACTIVE_LISTENING_UUID", "EMPATHY_UUID", "CLARITY_UUID"],
  "standard_ids": ["PARAPHRASE_UUID", "CLARIFY_UUID", "ACKNOWLEDGE_UUID"],
  "point_ids": ["POINTS_CONFIG_UUID"]
}'
 
# List your drafts
glow rubrics list

The draft endpoint uses PATCH semantics with expected_version for optimistic concurrency. The response includes draft_id, new_version, and a form_state containing the resolved standard_group_ids, standard_ids, point_ids, and other fields.

Search & filter

glow rubrics search --body '{"search": "communication"}'
 
glow rubrics search --body '{
  "filter_department_ids": ["CS_DEPT_UUID"],
  "page_size": 25
}'

Bulk operations

Bulk delete and update follow the canonical all-matching shape — explicit IDs, or all: true with flat filter fields plus optional excluded_ids and a patch body. The persona surface is the prove-out; rubrics follow the same shape.

Delete by explicit IDs:

glow rubrics delete --body '{"rubric_ids": ["rubric-1", "rubric-2"]}'

Delete all matching a filter (with exclusions):

glow rubrics delete --body '{
  "all": true,
  "search": "v1-deprecated",
  "excluded_ids": ["rubric-still-in-use"]
}'

Bulk update via patch:

glow rubrics update --body '{
  "all": true,
  "filter_department_ids": ["dept-retired"],
  "patch": { "archived": true }
}'

Common Operations

Related

• Rubrics API Reference — full endpoint and type documentation
• Rubrics CLI Reference — all CLI commands
• Evals Guide — use rubrics within automated evaluation runs
• Tools Guide — tools that agents use alongside rubric-scored behavior