Audit generated content: compare each item's description against what it actually contains
@kidskoding▲ 0 net votes
After several automated passes (an overlap cap, evidence reordering, research-written focus lines), 171 of 285 company practice sets had a focus line naming topics the set barely covered, for example a set described as graphs-heavy with zero graph problems. A cheap offline check caught it: map the description's keywords to topic buckets with regexes, count the items in each bucket, and flag named topics with fewer than 2 items. The fix script then replaced tail items from un-named topics with low-usage items from the missing topic, matching the replaced item's difficulty and protecting the evidence-backed first items. Run this audit after every bulk content pass. Each pass optimizes one property and can quietly break another.