fix: apply PUE to reported power too

[davidberenstein1957]

What changed

_do_measurements now scales the measured power by self._pue at the same point where it already scales energy (codecarbon/emissions_tracker.py:1177). Power.__mul__ already exists, so every hardware branch below picks up the scaled value automatically.

Why

PUE was applied to energy (and, through it, to water) but not to the power values reported in the same record. With pue != 1 the CSV row was internally inconsistent: energy_consumed did not match (cpu_power + gpu_power + ram_power) x duration, and nothing in the row explained the gap. PUE is a facility property, so it applies equally to instantaneous power and integrated energy.

Verification

Added tests/test_emissions_tracker_constant.py::TestCarbonTrackerConstant::test_carbon_tracker_offline_constant_pue, which runs the offline tracker with a forced CPU power and pue=2.0, asserts the reported cpu_power includes the PUE, and asserts energy_consumed is reconstructible from the reported power columns and duration. The test fails on master and passes with this change. Full file: uv run pytest tests/test_emissions_tracker_constant.py -q -> 7 passed.

Note on visible numbers

This changes reported power values for users running with pue != 1 — cpu_power, gpu_power and ram_power (and the per-hardware logger.info lines) are now facility-level. energy_consumed and emissions are unchanged. The default pue=1.0 means the overwhelming majority of runs see no difference, but the change is not silent and warrants a changelog entry. Anyone manually multiplying power by PUE downstream would now double-apply it.

Follow-up, out of scope here: EmissionCreate (codecarbon/core/api_client.py) carries wue but no pue, so the server cannot tell whether an incoming cpu_power is device- or facility-level.

Closes #1320

🤖 Generated with Claude Code

Changelog / release note

CHANGELOG.md is not on master yet (it arrives with docs/traction-batch), so the note lives in the docs instead: docs/reference/output.md now carries a "PUE and the power columns" admonition. When the changelog lands, copy this entry:

Changed — the pue multiplier is now applied to the reported power columns (cpu_power, gpu_power, ram_power) as well as to the energy columns, so energy_consumed stays reconstructible from the reported power. Runs with pue != 1 will see larger power values; energy and emissions are unchanged.

Consequence worth spelling out: pue also scales forced values, so force_cpu_power=100 with pue=1.5 now reports a 150 W CPU. That is consistent with treating the columns as facility-level, but it is surprising for a parameter named "force" — documented in docs/reference/output.md and in both pue docstrings.

Because this changes user-visible numbers, it should ride a minor release (3.4.0), not a patch.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 91.43%. Comparing base (065d0e6) to head (1342fca).

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1332      +/-   ##
==========================================
+ Coverage   91.39%   91.43%   +0.04%     
==========================================
  Files          49       49              
  Lines        5056     5057       +1     
==========================================
+ Hits         4621     4624       +3     
+ Misses        435      433       -2     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.