Skip to main content

Lab Results

Stave's control catalog and evaluation engine verified against four independent vulnerable-infrastructure lab suites. Every documented attack path is detected from a static configuration snapshot.

Summary

VendorLabScenariosFindingsChainsDetection rate
Rhino SecurityCloudGoat105312100%
Bishop FoxIAM Vulnerable33 users30100%
NCC GroupSadCloud12 services573100%
DatadogPathfinding Labs2 chains7100%

Zero false positives across all four suites.

CloudGoat — 10 scenarios

ScenarioKey findingChains
iam_privesc_by_attachmentATTACHUSERPOLICY.001iam_privesc_by_attachment
iam_privesc_by_rollbackCREATEPOLICYVERSION.001iam_privesc_by_rollback
lambda_privescASSUMEROLE.001 + PASSROLE.CREATEFUNCTION.001lambda_privesc
cloud_breach_s3IMDSV2.001 + PUBLIC.001ec2_public_credential_exposure
codebuild_secretsSECRETS.001 + ENCRYPT.001
ecs_efs_attackIMDSV2.001 + ENCRYPT.001ec2_public_credential_exposure (x2)
glue_privescPASSROLE.CREATEJOB.001
sns_secretsPOLICY.PUBLIC.001 + ENCRYPT.001sns_data_exposure
sqs_flag_shopASSUMEROLE.001lambda_privesc
vulnerable_cognitoCOGNITO.MFA.001

What the labs proved

  1. The control catalog covers real attack paths — not theoretical misconfigurations, but the exact configurations deployed by professional red-team lab vendors
  2. Compound chains assemble correctly — multi-hop escalation paths (user → role → Lambda → admin) are detected as chains, not just individual findings
  3. Three reasoning engines agree — CEL, Soufflé, and Z3 independently verify the same facts on the IAM escalation scenarios
  4. The collector pattern works — scenario-specific collectors capture only the relevant assets, producing clean, focused observations

What the labs surfaced

  • 3 engine bugs fixed during Lab 1 (Inconclusive exposure, asset-type gate, assessment cache)
  • 1 new control authored during Lab 7 (PASSROLE.CREATEJOB.001 for Glue job escalation — the catalog had no Glue CreateJob control)
  • 2 S3 policy intersection controls authored after the labs (SHADOW.ALLOW.001, DENY.BYPASS.001)
  • Collector enrichments accumulated across labs: service wildcards, group-inherited policies, compound escalation actions, role escalation properties

Every bug, gap, and enrichment was found by running the labs, not by auditing the code. The labs are the test suite.