The clang -M include pass is unaffected by a warm AST cache — confirmed at --repeat 3

Index: Known gaps

Status (re-verified 2026-10-09 at 456989f): OPEN. header_include_memo.py is a process-wide memo with no persistent/AST-cache tier, so a new run still recomputes the include graph whatever the warm AST cache holds.

Re-confirmed after the cache-lifecycle fix below, now that a multi-repetition run actually works: across three repetitions of the cold/warm sequence the header extraction count goes 2 → 0 (the AST cache serves it) while include_pass stays at 1 every time. The include graph is recomputed on every run regardless of cache state, which is the same finding as the per-header scaling above seen from the cache side.