K11 Software Solutions
Research Preview · 2026

Carbon-Aware Testing
for Agentic CI/CD

The first scheduler to jointly optimise defect detection and grid carbon footprint — reducing CI pipeline emissions by 38.7% with zero quality loss.

🌍 World's first carbon-aware scheduler for agentic CI/CD pipelines
38.7%
CI carbon
reduction
52.4%
Grid intensity
reduction
0.931
F1 score
maintained
197
Tests · 5 phases
CASSET CASSET — Carbon-Aware Scheduling System for Eco-Tier Testing
The Problem

CI/CD pipelines have a carbon footprint problem

Full-suite agents run for every PR at any grid intensity, any time of day. No scheduler addresses both testing cost and carbon.

⚡

Energy-intensive by default

Full-suite CI agents run for every PR regardless of risk level, burning compute at peak grid carbon intensity.

~45 W TDP per agent
🕐

Timing is never considered

Grid carbon intensity varies 3× throughout the day. The window matters as much as the workload itself.

420 vs 199 g CO₂/kWh
📋

Regulatory gap is widening

EU CSRD and the GSF Software Carbon Intensity spec require software emission measurement. CI/CD is unaddressed.

CSRD 2024 mandate
The Solution

Two mechanisms. One scheduler.

CASSET wraps your existing agentic QA pipeline via a zero-modification adapter. No pipeline code changes required.

Mechanism 01

Risk-Proportionate Tier Routing

PR risk score maps to an agent tier. Low-risk PRs run only cheap, fast agents. Full-suite reserved for high-risk.

38.7%
CI carbon reduction across 180 PRs
Mechanism 02

Grid-Intensity Deferral

HIGH-tier agents query the GSF Carbon Aware SDK and defer to the lowest-carbon 30-minute window within 6 hours.

52.4%
Grid intensity reduction via deferral
Mechanism 01

Risk-Proportionate Tier Routing

A 5-feature risk score (lines changed, cyclomatic complexity, history, dependencies, test coverage) gates which agents run.

LOW <0.40
−52.5%
−52.5% CO₂
MED <0.70
−35.8%
−35.8% CO₂
HIGH <0.90
−38.4%
−38.4% CO₂
CRIT ≥0.90
−28.4%
−28.4% CO₂
CRITICAL PRs always run immediately — safety overrides sustainability. DeferEngine exempts CRITICAL from scheduling delays.
Mechanism 02

Grid-Intensity Deferral

HIGH-tier agents query the Green Software Foundation Carbon Aware SDK. APScheduler executes at the cleanest available window.

24-hour grid carbon intensity — eastus zone
00:0006:0012:0018:0024:00
420
g CO₂/kWh
at submission
199
g CO₂/kWh
at deferred window
2.1 h
median deferral
latency
Empirical Evaluation

180 real PRs. Zero quality compromise.

Five-phase verification: 197 tests across unit, integration, agentic, E2E, and carbon accounting layers — all passing.

38.7%
CI carbon reduction
via tier routing
53.52 g → 32.82 g CO₂
52.4%
Grid intensity reduction
via deferral
31.4% of 53.52 g baseline
0.931
F1 score maintained
zero quality impact
Identical to full-suite
197
Verification tests
all green
5-phase test suite
Architecture

Zero-modification adapter pattern

🔀
PR Event
CI trigger
→
🔬
RiskRouter
Score → Tier
→
🌿
CarbonClient
GSF SDK query
→
⏰
DeferEngine
APScheduler
→
✅
QA Pipeline
Adapter call
Tech Stack

Python · FastAPI · APScheduler · Redis · Docker · GitHub Actions

Standards

GSF Carbon Aware SDK · SCI Specification · EU CSRD · Apache 2.0

Evaluation

180 PRs · 5-phase test suite · 197 tests · eastus grid zone

K11
Open Source · Apache 2.0

Make your CI pipeline carbon-conscious

Drop-in adapter for any agentic QA pipeline. Zero modification required. Immediate carbon reduction.

🐍 Python · FastAPI 🌿 GSF Carbon Aware SDK 📋 EU CSRD Ready ✅ 197 Tests Passing
⭐ View on GitHub
Kavita A. Jadhav · K11 Software Solutions LLC · Texas, USA
1 / 8