Skip to main content
Performance evidence

WattSutra AI

Record time and memory. Leave watts, carbon, and money empty until you supply them.

Offline performance-evidence CLI. npm [email protected] published 2026-11-15. The README rule is that CPU time is not energy. No sensor is read in 0.2.

  • Offline
  • 20 skills
  • Node 20+
GitHub version
0.2.1
Licence
Apache-2.0
Language
JavaScript
Weekly downloads
440
Overall downloads
440
GitHub stars
4

Live npm weekly downloads, npm overall downloads, and the latest npm version, plus GitHub stars and the package.json version, fetched 2026-11-17 19:07 UTC. Cached 10 minutes. Weekly downloads cover 2026-11-10 to 2026-11-16. Overall downloads cover 2026-10-01 to 2026-11-17. npm does not publish an earlier total. Stars and downloads show public use. They are not a quality score.

Use it when

  • 1

    You want a benchmark where every number is labelled measured, imported, modeled, or not measured.

  • 2

    You want a comparison that fails when the result changed or the candidate is slower beyond the interval.

  • 3

    You do not want CPU time converted into energy or CO2.

What it does, and what it does not

Read this before the guide. A limit here is a limit of the tool.

Does

  • Labels wall time, process CPU time, and process RSS as measured.
  • Leaves energy, carbon, and cost empty until you import a meter reading or supply explicit model inputs. Those values are then labelled imported or modeled.
  • compare gates: FAIL, REVIEW_REQUIRED, PASS, or NOT_RUN. The interval is a seeded percentile bootstrap, not a significance test.
  • The built-in workload is a fixed prime-sum function with two algorithms. The README says arbitrary code and live inference benchmarks are planned, not shipped.
  • 20 skills: 17 core and 3 advisory (gpu-scout, finops-auditor, optimization-planner). Live IDE activation is not verified.
  • The README self-audit expects PASS with 17 of 18 verified, 0 failed, and 1 not-run. That not-run check is the hardware energy instrument. Timings vary by machine.

Does not

  • It never turns wall time, CPU time, or memory into watts, kWh, CO2e, or money.
  • No sensor is read in 0.2. RAPL, NVML, powermetrics, Scaphandre, and Kepler are roadmap. doctor only reports detected-not-used.
  • A hash proves a file was not edited afterwards. The README says it does not prove the numbers were honest. Signing is roadmap.
  • Process RSS is whole-process memory, not per-workload allocation.
  • Carbon uses only the intensity you supply. No regional defaults. No SCI conformance is claimed.
  • The audit is a self-audit, not a third-party certificate. 0 runtime dependencies. Node.js 20 or newer.

Claims above are from the project pages and are attributed as stated there. They are not independently verified here.

How a run moves

From the WattSutra AI README, npm 0.2.0, published 2026-11-15. CPU time is not energy. No meter is read in this version.

  1. 1. Run

    A fixed workload

    prime-sum, two algorithms. Arbitrary code and live inference are not shipped.

  2. 2. Measure

    Time and memory

    Wall time, process CPU time, and process RSS are measured. Watts are not derived from them.

  3. 3. Label

    Four kinds

    measured, imported-meter-observation, modeled, or not measured. A model uses the watts you state.

  4. 4. Compare

    A gate

    FAIL, REVIEW_REQUIRED, PASS, or NOT_RUN for energy, carbon, and cost when no meter or model was given.

How to read the result

Result It means It does not mean
measured Wall time, CPU time, or process RSS from this process. Not watts, kWh, CO2e, or money.
not-run The energy instrument check in the self-audit. The README says that is correct because no meter is read. Not a green energy result.
PASS on compare Same quality, same environment, enough samples, and the interval excludes no change. Not a significance test. The bootstrap is seeded and reproducible.

Wording is from the project page or README and is attributed as stated there. It is not independently verified here.

Commands and notes

  • npx [email protected] audit --out wattsutra-audit self-audit of the tool, not a scan of your app.
  • wattsutra bench --manifest fixtures/baseline.json --out run.json
  • wattsutra compare --baseline a.json --candidate b.json
  • wattsutra estimate --run run.json --assumptions fixtures/assumptions.json
  • Exit codes in the README: 0 ok, 1 gate FAIL or verification failed, 2 usage or validation error.

Guide

How to use the result. GitHub is the source repository. npm is the install.

What gets measured

From the WattSutra AI README, npm 0.2.0, published 2026-11-15. The rule is that CPU time is not energy.

  • Wall time, process CPU time, and process RSS are labelled measured.
  • Energy, carbon, and cost stay empty until you import a meter reading or supply explicit model inputs. Those values are then labelled imported or modeled.
  • compare gates: FAIL, REVIEW_REQUIRED, PASS, or NOT_RUN. The interval is a seeded percentile bootstrap, not a significance test.
  • The built-in workload is a fixed prime-sum function with two algorithms. Arbitrary code and live inference benchmarks are planned, not shipped.

What 0.2 does not read

  • No sensor is read. RAPL, NVML, powermetrics, Scaphandre, and Kepler are roadmap. doctor only reports detected-not-used.
  • The self-audit expects PASS with 17 of 18 verified and 1 not-run. That not-run check is the hardware energy instrument. Timings vary by machine.
  • A hash proves a file was not edited afterwards. The README says it does not prove the numbers were honest. No regional carbon defaults. No SCI conformance is claimed.
  • 20 skills: 17 core and 3 advisory. Live IDE activation is not verified. npx wattsutra audit is a self-audit of the tool, not a scan of your app. Node.js 20 or newer.

Questions

Does WattSutra calculate carbon from CPU time?

No. The README says CPU time is not energy. Watts, kWh, CO2e, and money stay empty until you import a meter reading or supply explicit model inputs.

Does the audit command benchmark my application?

No. npx wattsutra audit is the tool self-audit. The built-in workload is a fixed prime-sum function. The README says arbitrary code benchmarks are not shipped.

WhatsApp