What Needs Grading, and What Doesn't
Most Step types need nothing from you. Content steps complete on their own, answer-checked steps mark themselves against the answers you set, and survey-style steps collect responses with no right or wrong. Challenges grade themselves too. A Game Master's actual review time goes on Open Text steps and the written submissions they produce.
How a Grade Decides Rewards
Three settings decide what grading does:
- Validate on the milestone ("Require validation before awarding") — the milestone's rewards wait until you validate the submission.
- Grade scale type, in Adventure Settings (Resource Mechanics) — None, Percentage or GPA (letters). With None you validate or invalidate. With a scale you also give a grade.
- Assign Resources, on the same panel — Before Grading or After Grading. After Grading holds a milestone's rewards until it has a grade above zero, then pays its BLOO in proportion to the grade: a 75 pays 75% of the BLOO.
Setting the Bar Before They Submit
An Open Text step can require a minimum word count (Min Words), so a one-line answer simply can't be sent. The player is told how many words they need and how many they wrote. You never see the half-effort version, and the player gets the feedback immediately rather than waiting for you to bounce it back.
Turning On AI-Assisted Feedback
With a Claude API key on the Adventure (see Setting Up AI Grading), every Open Text step has three A.I. settings:
- Check the player's answer with A.I. before they continue — the instant check. When it's on, the answer goes to Claude when the player submits it. If it falls short, the player sees why and revises before moving on.
- A.I. Strictness — Lenient (only reject blank or unrelated answers), Standard (a genuine, on-topic attempt passes) or Strict (every part of the prompt must be engaged with).
- A.I. Reviewer Context — your private rubric. Say what a good answer to this question must include: key ideas, facts, what to reject. Players never see it, and the A.I. is told never to quote it back. The milestone's own content is sent automatically, so write only what's specific to the question. It's optional: without it, the A.I. judges whether the answer is a genuine attempt at the prompt, not whether it's right.
One rubric, two readers. The same strictness and reviewer context drive the player's instant check and your second opinion on the review screen, so what a player is held to and what the reviewer is told can't drift apart. Both are used even when the instant check is off, so you can still ask for a second opinion on a finished submission. On the review screen, when a submission covers several Open Text questions, the strictest one sets the bar.
The check never strands a player. If it can't run — no key, a rejected key, an unreadable reply — the answer passes. That also means a broken key looks exactly like a pass. Staff who play the milestone see an amber note under their answer saying why the check was skipped, and the skip is written to the activity log. Players never see either. Press Test Key after saving a key to be sure.
Your Review Screen
Open it from the milestone's Review link on Manage Content → Journey, or from Review Milestone in the milestone builder. Each submission is a card with the player's name and email, when it was published and last modified, and the full answer, plus:
- A Grade field (0–100), when the Adventure uses a grade scale. On a milestone with Validate on, a grade above zero also validates the submission.
- On a milestone with Validate on, one button showing the action available: Validate or Invalidate. Validating keeps any grade already given. With nothing graded yet, it records 100, and invalidating records 0.
- Validate with A.I., when the Adventure has a key.
- GM Feedback (visible to player) — a written note for the player. See below.
For a Challenge, the review screen lists attempts instead: status, grade, correct answers and date.
Work through submissions per milestone rather than per player. It's easier to stay consistent when you're comparing several answers to the same prompt back to back.
A Second Opinion From the A.I.
Validate with A.I. sends one submission to Claude with the milestone's content, each question's prompt and your reviewer context, at the step's strictness. The verdict panel says whether it looks like a genuine, complete answer and why. It also shows a suggested grade on your Adventure's scale and a suggested comment for the player. You decide what happens:
- Validate with this feedback
- Invalidate with this feedback
- Add the feedback without validating
All three are offered whatever the verdict: the A.I. is a second opinion, not the decision. If the player edits their answer after the check, applying the verdict is refused and you're asked to run the check again, so a grade the A.I. never gave can't be written. Every verdict is saved, and a later Validate All reuses it rather than paying for it twice.
Written Feedback
Type in GM Feedback (visible to player) and it saves when you click away. A.I. feedback you accept lands in the same box. The player sees it as "Feedback from your Game Master" when they reopen the milestone after it's been graded or validated. At the moment, that note only appears on milestones submitted as a single written answer, with no steps. On step-based milestones the feedback is saved, and travels in the review CSV, but isn't shown to the player yet.
Who Can Grade
GMs and admins. An NPC can open the review screen and read every submission and verdict, but can't grade, validate or run the A.I.
Where to go next: Reviewing & Grading Submissions in Bulk, once you're grading at cohort scale.