Challenge 2 — Solution

Task: Explain, using this chapter's own finding about the duplicated
"HTML Course Header Comment Rule" memory, what the ideal fix would
actually be — not just noticing the duplication, but what should
happen to the memory file itself.

The finding itself is that a feedback memory ("HTML Course Header
Comment Rule") states the same rule, word for word in substance, as
P1 in permanent_rules.md - a real rules file this course's own Chapter
3 already covered in depth. The memory almost certainly predates P1
actually being written as a formal rule - it likely captured the very
first time this convention was requested, before it graduated into a
permanent, always-loaded rule.

The ideal fix isn't simply deleting the memory silently, since the
memory file itself preserves real context worth keeping (when this was
first requested, that it originated as feedback rather than being
designed upfront). The better fix is retiring the memory's own role
as the active source of truth for this rule, since that role now
correctly belongs to P1 - while still leaving a trace of its own
history rather than erasing it outright. Concretely: the memory's own
content could be updated to note explicitly that it has been
superseded by P1 in permanent_rules.md, and the MEMORY.md index
entry's own one-line description updated to reflect that it's now
historical rather than an active, independently-relevant instruction -
so a future session doesn't waste effort treating two sources as
equally authoritative for the same rule.

Notes:
- This mirrors exactly the kind of transparent supersession Chapter 1
  found in this project's own CLAUDE.md - the real note stating "Path
  correction (supersedes R1 in recovery_rules.md)" rather than quietly
  deleting the old R1 rule with no trace. The same discipline applies
  here: mark the old source as superseded, don't just delete it and
  lose the history.
- The general principle this illustrates: once a piece of feedback
  memory's own content becomes formalized as a permanent rule, the
  memory's job is done - it shouldn't keep acting as an independent,
  parallel source of truth for the same instruction going forward.