{
  "intent": "Split usage_report.py (426 lines, over the 250 cap) into an entry point plus three concern-shaped modules, verified by the 100%-coverage tests from the preceding commit.",
  "scope": [
    "usage_report.py 426 -> 101 (argument parser, mode dispatch, main)",
    "New _usage_report_logs.py 117 (log discovery, the report-state file, per-day segment planning, log descriptions)",
    "New _usage_report_emit.py 68 (preflight checks, stdout emission, clipboard)",
    "New _usage_report_run.py 195 (the two run paths and the aggregation they drive)",
    "Test files rewired to name the owning module for every moved symbol",
    "Non-goal: any behaviour change"
  ],
  "changes": [
    "Extracted three modules along concern boundaries: where the logs are and which slices to read; how the finished report leaves the process; and the two runs that tie those together. usage_report.py keeps only _is_single_day_mode, _build_parser and main.",
    "_Segment moved to the logs module and _Aggregates to the run module, each next to the code that builds it.",
    "_should_advance_state and _build_parser were relocated mid-split when the first placement created a cycle: _run_since needs the former, and the latter belongs with main. Both are AST-identical to HEAD wherever they landed.",
    "Rewired every test reference to name the owning module. This was the bulk of the work and the source of every failure along the way: monkeypatch.setattr must target the module whose namespace resolves the name, which is the caller's module for an imported function. Patching run._run_since instead of usage_report._run_since made main() call the real function, which hit the machine's actual atop logs and hung the suite - fixed by patching where main resolves it."
  ],
  "verification": [
    {
      "command": "AST comparison of every top-level def and class against HEAD",
      "result": "pass",
      "evidence": "'compared: 22, mismatches: 0'. Every function and both dataclasses are AST-identical to their pre-split form across the four modules, so nothing was rewritten while being moved."
    },
    {
      "command": "pytest linux_configuration/tests/ --cov of all four modules --cov-branch",
      "result": "pass",
      "evidence": "usage_report.py 25 stmts / 2 branches, _usage_report_logs.py 56 / 8, _usage_report_emit.py 29 / 10, _usage_report_run.py 91 / 24 - all 100%, zero missed statements and zero partial branches. 204 passed, the identical count to before the split."
    },
    {
      "command": "python usage_report.py --date <today> and --since <today>",
      "result": "pass",
      "evidence": "Both entry points produced full reports from the machine's real logs (35 atop samples, 11988 pmon samples). Running both matters here because the split separated the two run paths into a new module; the unit tests stub their collaborators, so only a real run proves the import graph resolves at runtime."
    },
    {
      "command": "bash ~/utils/scripts/check_file_length.sh --all",
      "result": "pass",
      "evidence": "38 over-cap files before, 37 after. usage_report.py is off the list and all four modules are well under 250 (101/117/68/195)."
    },
    {
      "command": "pre-commit run --files <all changed source and test files>",
      "result": "pass",
      "evidence": "All hooks pass after ruff auto-fixed 14 import-ordering findings created by the mechanical move. Coverage was re-measured after those fixes and still reports 100% on all four modules with 204 passing."
    }
  ],
  "risks": [
    "This split moved far more than the previous two and took several corrective passes: two names had to be relocated after their first placement created an import cycle, and four modules' worth of monkeypatch targets had to be re-pointed. Each step was verified, but a smaller first cut would have been cheaper - worth remembering for optimize_vscode.py, which is the next 498-line target.",
    "Three source-level bugs were introduced and caught by the tests during the split: Path and _fmt_h left under TYPE_CHECKING while used at runtime, and dataclass missing where _Aggregates landed. All three were import-only and would have failed at first call; the 100% coverage from the previous commit is what surfaced them immediately.",
    "_usage_report_run.py at 195 lines is the largest of the four and holds both run paths plus the aggregation helpers. If it grows, _aggregate_segments and _Aggregates are the natural next seam."
  ],
  "rollback": [
    "git revert the commit. usage_report.py returns to 426 lines with everything in it, the three new modules disappear, and the test files return to referencing usage_report for every symbol.",
    "After rollback, confirm pytest linux_configuration/tests reports 204 passed and that both `usage_report.py --date <today>` and `--since <today>` still produce reports."
  ]
}