{% extends "base.html" %} {% block title %}Starter packs — FreeWeight{% endblock %} {% block content %}
Four goals ship with FreeWeight, complete with tasks, criteria, proposed rules and a worked set of graded calibration samples. They are here to be read: down this page, the share of weight scored deterministically rises from 40 % to 90 %, and that is the single most useful thing to internalise about writing a measurable rubric — the better you understand what you want, the less of it needs a judge.
unforked in the UI, in its results and in its exports. A voice
measured on somebody else's prompts is not your voice.
{{ error }}
{{ pack.intent }}
Free, exact, and never disagrees with you.
This is what calibration has to measure.
{{ pack.task_count }} starter tasks, all replaceable.
graded samples, reproducing κw {{ '%.2f' % calibration.summary.weighted_kappa_w }} · validity {{ '%.2f' % calibration.summary.judge_validity_factor }}
{{ pack.carries }}
{% if calibration %}| Criterion | κw | ρ | MAE | Bias | nholdout | Band |
|---|---|---|---|---|---|---|
| {{ criterion }} | {{ '%.3f' % figures['kappa_w'] }} | {{ '%.3f' % figures['rho'] }} | {{ '%.2f' % figures['mae'] }} | {{ '%+.2f' % figures['bias'] }} | {{ figures['n_holdout'] }} | {{ figures['band'] | replace('_', ' ') }} |