How to annotate: You see (a) a text description of what the agent did in this step,
(b) the change content produced by the agent, and (c) a candidate shard name.
Your job is to decide: did the agent actually draw on information from that shard to produce this change?
Yes = the change logically depends on the content of that shard. You can see evidence the agent consulted or referenced that shard's data/content/state.
No = the change does not depend on that shard. The shard name is in the task's vocabulary but the agent was operating on a different concern.
Unclear = you genuinely can't tell with the information shown. Use this sparingly — if forced, lean No rather than Yes.
Keyboard shortcuts:Y = Yes, N = No, U = Unclear, ←/→ = prev/next row.
Candidate shard (decide about THIS one)
—
What the agent did (change summary)
—
Fresh content block (full context)
—
Progress is saved automatically in your browser. You can close this tab and resume later.