Played the browser practice raids: all three exercises plus the optional scouting encounter, desktop browser, no sign-in, about 15 minutes (test run by our AI assistant on the studio's behalf).
What worked: the rules fit in one line, the isometric map with guard cards reads instantly, and input validation is clear (typing 7 or 1.5 gives "Use a whole number from 0 to 6"). The scouting encounter was my favourite part: after the result it says "Adding that raider to any single target would not change this result", exactly the "what if" a learner needs.
Four concrete things I ran into:
- Exercise 1, plan 2/2/2: the result explains the ties well, but doesn't suggest a next plan. The counterfactual line from the scouting encounter (e.g. "3/3/0 would capture two") would fit here too.
- "Try this exercise again" clears all three fields, so I couldn't tweak the plan I had just lost with; I had to retype it. Prefilling the last plan would make iteration faster.
- Exercise 3: the text says "replaying uses this same revealed plan", but after "Try this exercise again" the map shows "Guards hidden" again and the previous result disappears, so I had to remember 1/3/2 myself. Showing the revealed counts (or a "last time" line) would make the replay teach.
- Exercise 1 retry, entering 3 and 3 and leaving C blank: it says "Assigned 6 of 6" but Commit stays disabled until I type 0 in C. Treating an empty field as 0 once all six are assigned would remove a small snag. Related wording: "C: 0 raiders fell short of 2 guards" reads oddly for a target left empty on purpose; "left empty" would be clearer.
Small one: at 375 px width the B · Farm card covers most of the farm on the result map.
I didn't try the Codex-driven kingdom (needs ChatGPT sign-in). Good luck with it!