System-level impact analysis for microservice CI/CD
via cross-repository dependency graphs

K11tech Microservice QA System  ·  K11 Software Solutions LLC, Texas, United States  ·  Paper 3

100% P/R/F1 on 74 real PRs Transitive consumer detection LangGraph · 14 agents Zero false positives

A REST endpoint change passes all local CI checks — yet silently breaks 3 direct and 2 transitive consumers in other repositories. Conventional gates are blind to cross-repo dependencies.

order-svcPUT /orders → PATCH⚠ API breaking change ✓ CI green (blind) payment-svc💥 breaking invoice-svc💥 breaking audit-svc💥 breaking analytics-svc⚠ indirect impact notify-svc⚠ indirect impact depth 2 · transitive depth 1 · direct changed provider

A PR opens in order-service. The LangGraph pipeline runs 4 phases — watch the execution log stream in real time.

k11tech-microservice-qa — PR #482 · order-service
$ k11tech-qa run --pr 482 --repo order-service

Animated dependency graph — nodes and edges render in sequence, revealing the blast radius across depth 1 and depth 2 consumers.

DEPTH 2 · TRANSITIVE DEPTH 1 · DIRECT CHANGED PROVIDER order-svcPUT→PATCH ⚠ payment-svc💥 breaking invoice-svc💥 breaking audit-svc💥 breaking analytics-svc⚠ indirect impact notify-svc⚠ indirect impact HITL gatescore ≥ 0.60→ BLOCK direct (depth 1)transitive (depth 2)system gate impact score 0.87 · PR blocked 🚫

Step through the breadth-first search the system performs to discover all consumers at every depth.

unvisited in queue direct (depth 1) transitive (depth 2) complete
order-svc changed · seed payment-svc depth 1 invoice-svc depth 1 audit-svc depth 1 analytics-svc depth 2 notify-svc depth 2
Queue:empty — click Start BFS

Phase A — controlled (15 scenarios)

Precision
Recall
F1 score
Tiers testedB1 / B2 / B3

Phase B — OSS (59 external PRs)

Precision
Recall
F1 score
False positives

B3 — transitive consumer validation

Chains tested
Max depth
Detection rate
Impact decay0.4× per hop

K11tech Agentic AI QA series

1
Single-repo QA pipeline
LangGraph · 14 agents · Risk-proportionate HITL
2
Detect–Fix–Learn loop
Consensus gate · Auto-remediation
3
Cross-repo impact analysis
This paper · Contract registry · Graph traversal
4
Uncertainty quantification
Conformal prediction · HITL confidence
5
Uncertainty source classification
DATA vs SCOPE · Type-stratified thresholds

Open source · Apache License Version 2.0 · deployable alongside the single-repo QA pipeline