2 Critical findings
The weak baseline exposed gaps around AI controls, prompt injection, sensitive data, public-service ownership, and operational handoff.
SYNTHETIC REHEARSAL CASE STUDIES
See Janus reason at two very different scales.
These case studies are fictional/synthetic demonstrations of Janus Shield's reasoning. They are not real customer engagements, production assessments, certifications, or independently verified customer outcomes.
Small businessBiscuit Barn benchmarkEnterpriseDQ-Scale benchmarkSame engineNo customer-specific scoring shortcuts
CASE STUDY 01 · BISCUIT BARN
Can Janus recognize meaningful improvement instead of repeating the same alarm?
Biscuit Barn is a fictional two-location business used as a small-business regression benchmark. We first loaded an intentionally weak Digital Twin, then documented a remediated control state and rehearsed the environment again.
17 → 88 modeled readiness.
The baseline produced modeled Critical exposure with 2 Critical findings. After the synthetic Twin documented the remediated controls, Janus resolved all five modeled findings as controlled and moved the environment to modeled Low exposure with 0 Critical findings.
5 documented safeguards
The remediated Twin documented the controls Janus needed to distinguish an existing safeguard from an open gap.
0 open modeled gaps
Janus stopped recommending controls already documented as present and shifted the roadmap from remediation to validation and assurance.
CASE STUDY 02 · DQ-SCALE HEALTH BENEFITS ENTERPRISE
Can the same reasoning stay conservative inside a much larger enterprise model?
DQ-Scale is a synthetic health-benefits enterprise benchmark. It is not a DentaQuest customer engagement. The model combines synthetic enterprise structure with explicitly identified safe public observations to stress-test Janus at scale.
46/100 · Modeled High Exposure.
Janus identified 0 modeled Critical findings, 2 open control gaps, and 3 partial controls with Medium evidence confidence. Instead of forcing a favorable score, Janus kept unsupported controls open and separated modeled risk from evidence confidence.
ENTERPRISE REHEARSAL
46/100
Modeled readiness · Medium evidence confidence5 findings · 0 Critical · 2 High · 2 Medium · 1 Low
Relevant systems, not arbitrary inventory
Sensitive-data reasoning selected clinical history, identity, analytics, cloud, eligibility, payment, backup, API, and provider workflows.
Work went to the right teams
AI, Data/Privacy, Application/Public-Service, and Incident/Operations ownership stayed separated by the objective of each finding.
“Test this” is not “this exists”
Janus rejected rehearsal instructions as proof while still recognizing declarative human-approval, rollback, MFA, rate-limiting, and handoff controls where documented.
WHAT THE TWO BENCHMARKS PROVE
Different environments. The same evidence discipline.
Load the Twin
Janus starts with the authorized environment description and keeps synthetic, client-stated, externally observed, and verified evidence distinct.
Rehearse the path
Findings connect controls and dependencies to plausible business consequences without claiming a real attacker used the path.
Resolve control state
Documented safeguards lower modeled risk; instructions, desired states, and test objectives do not masquerade as implemented controls.
Make a decision
Leadership receives priorities, accountable owners, verification guidance, rollback-safe Courses of Action, and a roadmap appropriate to the remaining gaps.
EVIDENCE MODEL
Claim only what the evidence can support.
Client-stated / Twin evidence
Information supplied in the authorized Digital Twin can support modeled control state while remaining clearly distinguished from independent verification.
Safe external verification
Finding-applicable checks such as authorized DNS, TLS, and HTTP observations can support only what those checks actually demonstrate.
Finding-specific evidence
Authorized internal evidence, synthetic workflow tests, or owner attestation can strengthen a specific finding without contaminating unrelated conclusions.
WHAT JANUS LEAVES BEHIND
A decision package, not just a score.
Representative artifacts include executive findings, control-state evidence, attack-path context, operational handoff, approval decisions, rollback guidance, and the run manifest.
janus_executive_findings.md control_state.json janus_evidence_layers.json janus_operational_handoff_tree.json janus_client_coa_decisions.json rollback_vault.json run_manifest.jsonExplore Deliverables
What would Janus find in your environment?
Build the authorized Digital Twin, rehearse the likely path, and turn modeled exposure into owner-ready decisions.