Resilience control plane
Break the system on purpose. Ship the architecture that survives it.
CDR injects load and faults into a real repository running in staging, captures the physical collapse, classifies the root cause, generates a refactor PR with full repository context and proves the fix by re-running the exact same scenario.
Runs tracked
8
7 completed
Collapses captured
7
under identical load profiles
Avg p95 reduction
94.0%
after verified fix
Avg AI diagnosis
3.1 min
vs 95 min manual SRE estimate
Pipeline
Inject chaos
k6 load + toxiproxy faults on a real target repo
Classify
telemetry + logs → failure category with evidence
Refactor
repository-aware diagnosis and generated patch
Verify
same chaos scenario re-run on the patched code
Experiment runs
run-83ed8926
checkout-latency-cascade
Queued — the worker will start it automatically
p95 before
—
p95 after
—
error rate
pending
run-3629c1f1
checkout-latency-cascade
p95 before
4,632 ms
p95 after
268 ms
error rate
38.4% → 0.5%
run-1228bd92
checkout-latency-cascade
p95 before
4,632 ms
p95 after
268 ms
error rate
38.4% → 0.5%
run-cc686bed
checkout-latency-cascade
p95 before
4,632 ms
p95 after
268 ms
error rate
38.4% → 0.5%
run-d4f359e3
checkout-latency-cascade
p95 before
4,632 ms
p95 after
268 ms
error rate
38.4% → 0.5%
run-df4dc318
checkout-latency-cascade
p95 before
4,632 ms
p95 after
268 ms
error rate
38.4% → 0.5%
run-09eb691c
payment-dependency-timeout
p95 before
3,547 ms
p95 after
254 ms
error rate
28.8% → 0.5%
run-3bbe1725
checkout-latency-cascade
p95 before
4,632 ms
p95 after
268 ms
error rate
38.4% → 0.5%