WattSutra AI documentation
Everything a new reader, an auditor or a future maintainer needs, so nobody has to ask. Each page exists as Markdown (source) and responsive HTML (generated by npm run build:docs; CI fails if it is stale).
Reading order
| Audience | Start here |
|---|---|
| Executive, buyer, reviewer | Executive summary |
| Product owner, tester | Functional specification |
| Engineer, contributor | Technical specification, then Architecture |
| User, CI owner | User guide |
| Security, legal, procurement | Security, privacy and legal |
| Auditor, skeptic | Independent audit and evidence, Observed test results |
One-paragraph summary
WattSutra AI benchmarks a fixed workload, verifies the answer stayed correct, compares a baseline and a candidate with fairness gates and writes JSON, Markdown and HTML reports. Every figure carries an evidence kind: measured, imported meter observation, modeled or not measured. It never converts CPU time into energy, carbon or money. Version 0.2.0 reads no hardware sensor.
Evidence files
Under evidence/: test-results.json (machine-written test run), samples/audit.{json,md,html} (a real self-audit), samples/measurement.{md,html} (a real run report), samples/comparison.json and samples/estimate.json.
How to keep these docs true
- Change code, then run
npm test,npm run test:report,npm run audit. - Copy the new audit into
evidence/samples/(node packages/cli/index.mjs audit --out documents/WattSutra-AI/evidence/samples). - Edit the Markdown, run
npm run build:docs, commit the HTML too.
Created by Vatsal Shah. Apache-2.0.