{% extends "base.html" %} {% block title %}Grade — FreeWeight{% endblock %} {% block content %}

{{ goal.pack.name }}

{% include "partials/_wizard_steps.html" %}

5. {{ step.heading }}

Blinded and shuffled: which model wrote which of these is not shown and is not fetched, and the order is not the order they were produced in. Every grade is saved the moment you submit it, so you can stop, refresh, restart the server and come back — nothing here is held in the page.

{% if error %} {% endif %}
Graded
{{ progress.recorded }} / {{ progress.expected }}

{{ progress.samples }} samples × {{ progress.judged_criteria }} judged criteria.

Minimum samples
{{ progress.min_samples }}

Below this the answer is "not enough grades", not "poor agreement".

Complete
{{ 'yes' if progress.complete else 'not yet' }}

{% if progress.complete %} Run the calibration to see the agreement. {% else %} Keep going — or stop, and come back to exactly here. {% endif %}

{% if progress.complete %}

{% endif %} {% if samples %} {% for sample in samples %}

Sample {{ loop.index }} of {{ samples | length }}

{{ sample.content }}
{% for criterion in criteria %} {% set existing = sample.grades.get(criterion.key) %} {% endfor %}
Grade this sample on each of your judged criteria. Notes are what the disagreement diagnostics quote back to you, so they are worth writing.
Criterion Your grade Note Save
{{ criterion.name }} {% if criterion.scale %}
what the points mean
{% for point, text in criterion.scale.descriptors | dictsort(reverse=true) %}
{{ point }}
{{ text }}
{% endfor %}
{% endif %}
{% include "partials/_csrf.html" %}
Grade {% for point in range(1, (criterion.scale.points if criterion.scale else 5) + 1) %} {% endfor %}
{% if existing %} graded {{ existing.grade }} {% endif %}
{% endfor %} {% else %}

Nothing to grade yet

This goal has no calibration samples. Add some — run a spread of models over your tasks, or paste in text you already have:

freeweight goals calibrate {{ goal.slug }} --generate

A calibration set that is all excellent or all terrible cannot produce a meaningful agreement figure. There has to be something to disagree about.

{% endif %} {% endblock %}