nx_collision_plan.nx

nx_collision_plan.nx -- the RESOLUTION PROPOSER: mechanizes the careful merge method for each shadow collision so the team can resolve sprawl without hand-judging every pair (and without ever losing a capability). For a shadowed basename it reads BOTH copies (runtime/<n> = A, runtime/_hdl_build/<n> = B), extracts each copy's FUNCTION-NAME SET, and from the set relationship proposes the resolution: shared==0 -> CODE 0 RENAME_DISTINCT : disjoint = two DIFFERENT capabilities sharing a name -> rename both to clear names (GROWTH, like nx_extract vs nx_func_extract). shared>0, onlyA==0, onlyB==0 -> CODE 1 KEEP_NEWER_GEN : same func-set, different bytes = the SAME capability in two GENERATIONS -> keep the _hdl_build (probe-first) canon, retire the old runtime gen. shared>0, onlyA==0, onlyB>0 -> CODE 2 RETIRE_A_SUBSET : B is the SUPERSET and B already RUNS (probe-first) = correct -> retire the runtime subset A. shared>0, onlyA>0, onlyB==0 -> CODE 3 CAPABILITY-LOSS!: A (runtime) is the SUPERSET but the LESS-capable B shadows it -> we are SILENTLY LOSING A's extra funcs -> must merge A's extras into B (or promote A). The highest-value catch. shared>0, onlyA>0, onlyB>0 -> CODE 4 MERGE_UNION : each copy has unique funcs -> merge the union into one. Toolchain-critical organs (nx_link = the ELF linker, never-brick #26) are flagged REVIEW regardless -- the proposer NEVER auto-executes; it emits a plan the warden/operator gates. READ-ONLY (never-brick by construction). RESIDUAL CAVEAT (honest): the program-vs-library signal is "has a top-level main". A LIBRARY that also carries a self-test main (e.g. nx_research_ledger, prints GREEN x2) looks like a program, so a library-vs-report pair can still slip through as MERGE_UNION. Full disambiguation needs IMPORT-ANALYSIS (does anything import it and call its exported funcs?) -- the next rung; for now the benchmark/human layer catches that edge. license_tier: ORIGINAL