Assessment Studio · PA Toolkit

ASSESSMENT SETUP

assessment/v1

Locale Question label Points label Result label Score label

QUESTIONS

Open answer is graded for real; this form just cannot edit it yet. Deploy compiles it to a text input plus gpt_and_save_int_answer against the rubric. Add it via Import JSON with rubric and max_points. The preview scores it with the deterministic offline stub, never the model.

RESULT & SUBMISSION

Reveal rules

Submit button label

Question 14 points

Answers remain editable until the explicit submit event.No score is shown before submit.

TOOLKIT DRY-RUN

Native authoring graph

Exact objects projected by pa_toolkit.ix.compiler.dry_run.

Inspector

Select a node to inspect its exact JSON.

EXISTING DOPOMOGAI ASSISTANT

Plan with the agent, apply with a human gate

This embeds the real /lab/deck chat. Assessment Studio does not create a second agent loop or call a model itself.

Not loaded

1Prepare intent

MCP-aware

Target What should change?

Prepared assistant prompt

Paste the copied prompt into the embedded chat. The existing assistant asks concise clarifying questions and uses the PA MCP toolkit. A model call occurs only inside that host when it is configured.

No live model available?Use a deterministic demo proposal to exercise validation and human gates. It is explicitly not an AI response.

2Validate proposal

Waiting for JSON

Assistant-produced assessment/v1 JSON

Nothing has been validated. Apply remains locked.

3Human apply gate

Applying creates a new browser-local assessment. It never overwrites the original.

4Separate DEV deploy gate

I explicitly want a guarded DEV deployment payload

This button only copies JSON. The local Studio never calls PitchAvatar or deploys anything.