{
  "intent": "Split three of the largest remaining linux_configuration scripts, and defer a fourth whose seam cannot be made lint-clean without a suppression.",
  "scope": [
    "single_use/utils/organize_downloads.sh (513 -> 72) + lib/download_classify.sh, download_sizes.sh, download_media.sh",
    "single_use/features/setup_kcd2_dice_solver.sh (497 -> 112) + lib/kcd2_serve.sh, kcd2_expose.sh, kcd2_autorebuild.sh",
    "single_use/fixes/fix_ubuntu_performance.sh (494 -> 150) + lib/ubuntu_perf_fixes.sh, ubuntu_perf_more.sh",
    "Deferred and fully restored: single_use/fixes/diagnose_pacman_hook_stall.sh (493)",
    "Non-goal: no script executed"
  ],
  "changes": [
    "Three cuts each for organize_downloads.sh and setup_kcd2_dice_solver.sh, two for fix_ubuntu_performance.sh. These are the first 490+ line scripts done, and every one needed more than one seam.",
    "Repeated splitting chained the libs - each new lib sourced the next, because the tool inserts its source line relative to the file it just wrote. Flattened so organize_downloads.sh sources all three directly and no lib sources a sibling, and removed a leftover SCRIPT_DIR the chain had left in a lib.",
    "diagnose_pacman_hook_stall.sh was reverted to HEAD. Its transaction runner assigns LAST_ELAPSED which the caller reads, and reads PACMAN_BIN which the caller defines; linted as a fragment that is SC2034 plus SC2153, and the pre-commit hook lints each lib in isolation. Neither is a real bug, which is exactly why suppressing them would be wrong - the honest answer is that this seam does not exist."
  ],
  "verification": [
    {
      "command": "meta/scripts/verify_shell_split.sh per script",
      "result": "fail then pass",
      "evidence": "organize_downloads.sh first reported SC1091 for a lib sourcing a sibling; after flattening, all three report 'libs sourced, functions defined'."
    },
    {
      "command": "pre-commit run shellcheck --files <all touched>",
      "result": "fail then pass",
      "evidence": "Failed on the pacman lib's SC2034/SC2153. This is the check that matters here and is stricter than shellcheck -x on the entry script, which passed on the same code - the hook lints each lib alone, with no view of the caller's globals. Passes now that the split is reverted."
    },
    {
      "command": "systemctl cat media-organizer.service, test -f its ExecStart, bash -n that file",
      "result": "pass",
      "evidence": "Resolves to organize_downloads.sh and parses. This is one of only two units pointing into the repo, and it is the script that shrank most in this batch (513 -> 72)."
    },
    {
      "command": "git checkout HEAD -- diagnose_pacman_hook_stall.sh, then shellcheck -x",
      "result": "pass",
      "evidence": "Restored to 493 lines and clean. Note the first `git checkout` without HEAD restored the staged split rather than the original - checked with wc -l, not assumed."
    },
    {
      "command": "bash ~/utils/scripts/check_file_length.sh --all",
      "result": "pass",
      "evidence": "None of the three entry scripts or seven libs is listed."
    }
  ],
  "risks": [
    "None of the three was run. organize_downloads.sh moves files in ~/Downloads and runs from a live systemd timer; verifying it by running would be exactly the mutation decision 6 forbids. The unit-path check and bash -n are the strongest available evidence.",
    "The lib-chaining bug now has two instances (20-dump-stock.sh, organize_downloads.sh). Any script split more than twice needs its source lines grepped afterwards; the tool does not do this itself.",
    "diagnose_pacman_hook_stall.sh remains at 493 lines, the largest deferred file. Splitting it needs the LAST_ELAPSED/PACMAN_BIN state passed explicitly rather than shared."
  ],
  "rollback": [
    "git revert the commit; the three scripts return to their previous sizes and the seven libs disappear.",
    "After rollback, re-run bash -n and shellcheck on the restored scripts."
  ]
}