nx_rebuild_plan.nx

nx_rebuild_plan.nx -- DEPENDENCY-AWARE STALENESS: which deployed organs must be rebuilt, and WHY. THE WOUND (measured 2026-07-30, debt 1785446507 sev8): 667 of 722 deployed organs run binaries older than what their source+deps produce. Only 222 of those are stale by their OWN source. The other 445 are stale because a SHARED runtime moved underneath them -- `nx_syscalls.nx` is imported by 14,172 files, so one edit to it invalidates almost every statically linked elf in the ecosystem WITHOUT touching a single line of their own .nx. Nothing in the ecosystem could compute that. The existing instruments each answer a smaller question: `nx_stale_check <target>` rebuilds ONE target and byte-compares (correct but O(one build) and says nothing about WHY), and an own-source mtime census misses the dominant 445 entirely. ★★★THE POINT IS THE TRIGGER, NOT THE VERDICT. "stale" is useless on its own -- 667 rows of "stale" is not a work order, it is noise. Naming the ONE file whose mtime invalidated each organ turns the census into a plan: 445 organs all triggered by nx_syscalls.nx is a SINGLE decision (rebuild the dependents of one file), not 445 investigations. A tool that reports a problem without naming its cause makes the operator redo the diagnosis the tool already did. METHOD: resolve each deployed <t>.elf to its source, walk the TRANSITIVE import closure (lazily, from the deployed organs as roots -- never the whole 17.5k-file tree), memoise closure_max_mtime + argmax per node, then compare against the deployed elf's mtime. Cycles are coloured and broken, so a cyclic import cannot hang the walk or silently drop a subtree. nx_rebuild_plan [elfdir] [srcroot] [maxrows] -> per-organ rows {organ, verdict, trigger, age_min} + a summary keyed by TRIGGER verdicts: CURRENT | STALE-BY-OWN | STALE-BY-DEP | STALE-BY-TOOLCHAIN | NOSOURCE Exit 0 always (a census is not a gate; it reports, it does not refuse). license_tier: ORIGINAL No hw writes (Rule 26). expect_exit: 0