Measure the compute cost, and prove it

Every pipeline run can report what it cost in energy and CO2, and seal that report so a reader can check it was not edited. The tracker follows the CodeCarbon methodology (the formulas and their sources are in Compute emissions); the seal is a bricklayer capsule.

Try it on your machine

morie emissions --seconds 5 --country CAN
rmorie emissions --seconds 5 --country CAN

Emissions:         1.76e-06 kg CO2eq over 5.0 s
Energy:            1.03e-05 kWh (CPU 4.77e-06, RAM 5.56e-06)
CPU:               Intel(R) Core(TM) i7-1185G7, TDP-scaled 8.6 W at 25.1% utilisation (machine mode)
Carbon intensity:  0.17 kg/kWh (CAN)
Capsule:           emissions/emissions_manifest.json + signed emissions/capsule_bundle.json

--country picks the grid (ISO-3); without it the packages use MORIE_COUNTRY_ISO, then a geolocation lookup, then the world average, and the report says which. MORIE_EMISSIONS_OFFLINE=1 skips the lookup.

Track a pipeline run

morie pipeline --all -y --output-dir out/        # ends with: Pipeline CO2 emissions: 0.000123 kg CO2eq
rmorie pipeline --all --output-dir out/          # same, from R; --no-carbon turns it off

Both write out/emissions/emissions.csv (one row per run, the CodeCarbon column layout, so rows from both languages can be concatenated) and the capsule next to it.

From code

from morie.emissions import EmissionsTracker
from morie.modules import run_module

def run_everything():
    run_module("power-design", output_dir="out/power-design")

with EmissionsTracker(project_name="my-analysis", output_dir="out/emissions") as t:
    run_everything()
print(t.capsule)        # {'manifest': ..., 'bundle': ..., 'signed': True/False}

run_everything <- function() rmorie::morie_run_morie_module("power-design", output_dir = "out/power-design")
r <- rmorie::morie_emissions_track(run_everything(), project_name = "my-analysis",
                                   output_dir = "out/emissions")
r$emissions_kg; r$capsule
rmorie::morie_emissions_verify("out/emissions")$ok

The capsule

emissions_manifest.json is a bricklayer manifest: the measurements, the method and its sources, the environment (interpreter, platform, packages). capsule_bundle.json holds the SHA-256 of the CSV and the manifest under an ML-DSA-44 signature; rmorie::morie_emissions_verify() or rmoriebricklayer::capsule_bundle_verify() re-hashes the files and checks it, so a changed byte in either file fails. The Python package signs through R when Rscript with rmoriebricklayer is installed and otherwise writes the manifest only and says “unsigned”.

The signing key is generated per run unless you pass your own (key = rmoriebricklayer::fips_keygen() in R), so by default the capsule proves the files are unchanged since sealing, not who sealed them.