[binlecode]

Summary

docs/TODO-layering-cleanup-2026-07-02.md drove LC-1 → LC-3 (shipped v1.2.4–v1.3.0: SystemSnapshot as the sole frame contract, ProcessSample through L2, and analytics.py taking AlertEngine / throttling / session energy out of the widget) — but it lived only on the unmerged lc-2-process-sample branch. The code shipped while its rationale stayed unpublished.

This lands it as the design record, then deletes that branch.

§§1–9 are kept verbatim as the July plan, so their line references point at July code — flagged in the header rather than silently rewritten, since rewriting them would destroy the record of what was actually decided.

Status verified against main @ v1.6.0

Confirmed shipped: MetricsUpdated carries exactly the snapshot (§8-3), and every internal §8-5 lists is gone from tui/ — analytics.py holds attribute_power, domain_throttling, bandwidth_percent, package_power_percent, AlertFrame, AlertEngine.

New §10 records the only open scope:

Item	Status
§10.1 — LC-4 RollingStats	Open, low priority. Watt/GB-s deques + _avg_max still in HardwareDashboard.
§10.2 — Export parity	Open, highest value here. Processes / alerts / session energy never reach --json / --serve.
§10.3 — §8-1's literal grep	Does not pass; should be reworded, not "fixed".

§10.2 is the one that matters

export.py emits nothing for three data points the TUI has:

• Processes — on the snapshot, but the export loops build Monitor(interval_s, subsamples) without include_processes=True, so the list is always empty.
• Alerts / throttle flags and session energy — produced by analytics.AlertEngine, which only the dashboard widget ever drives.

Someone profiling a local inference run through the export backends gets watts and utilization but cannot answer which process drew them, did the chip throttle, or what did the run cost in energy — the questions that profiling session exists to ask. Both candidate designs are recorded (opt-in export flags vs. moving AlertEngine behind Monitor), with a recommendation and the per-PID Prometheus cardinality constraint that keeps process data NDJSON-only either way.

§10.3 — why not just fix the import

grep "from actop.utils import" actop/tui/*.py still matches: tui/app.py imports get_soc_info for one construction-time call, create_dashboard_config(args, get_soc_info()). The criterion was written to kill per-frame L1 acquisition in the view, and that is gone — nothing in the render loop touches L1. Routing a single startup call through another module to satisfy a grep would add indirection for no layering gain, so the doc says reword the criterion.

Roadmap

Both open items promoted into docs/TODO-architecture-roadmap.md, which is where that file says open work is tracked (it explicitly excludes completed work).

Validation

Docs only — no code change.

.venv/bin/ruff check .            # All checks passed
.venv/bin/ruff format --check .   # 64 files already formatted
.venv/bin/pytest -q -m "not local"  # 95 passed, 44 deselected

🤖 Generated with Claude Code