Founder & engineer
Adeshina Emmanuel
Founder and engineer of RELIASTRA, Adeshina works across cloud identity, access control, Kubernetes, and AI systems. He applies that security-engineering experience to measuring external service behavior and preserving verifiable observation records.
Maintainer
Founder & engineer
Adeshina Emmanuel
AI and Cloud Infrastructure Security Engineer
His work spans cloud identity and access control, Kubernetes security, and the security of AI systems. His public work includes the Cloud Identity Security Engineering research series and open-source security tools. At RELIASTRA, he leads engineering of the observation pipeline, incident-confirmation rules, attribution methods, and verifiable evidence records.
Areas of practice
- Dependency observability and evidence integrity
- Cloud identity: IAM evaluation, trust boundaries, least privilege
- Security of AI infrastructure: model APIs, routing, agent boundaries
Research subjects
- Measurement integrity
- Distributed systems
- Kubernetes security
- Cloud access control
- Reproducible research
Profiles
Contact
- support@reliastra.com - product and support
- security@reliastra.com - vulnerability reports
Research & open-source work
Selected research and tools.
- chimeraA closed-loop causal reasoning engine for offensive security work: observe, model, hypothesise, interrogate, test, update, decide, remember. Python, no third-party runtime dependencies, 30 test modules covering the loop.
- headershieldA security-header audit tool that follows redirects, classifies each missing header by the risk it actually introduces, and reports HIGH, MEDIUM or LOW confidence per finding rather than a single score.
- CloudVitalsA Go cloud-security scanner that checks public storage exposure, open security groups, unencrypted volumes, root MFA and audit-log coverage, and prints only what is misconfigured.
- go-systems-labSystems software built in public to keep Go sharp: TCP behaviour, log analysis and the networking internals worth reading the source for.
Why it exists
External-service observations for incident review.
Organizations depend on external services they do not operate. During an incident, a provider’s status report and a customer’s measurements may differ because they describe different views of the service. A separately maintained observation record can help clarify what was seen, and when.
RELIASTRA focuses on recording HTTP behavior observed by its probes, retaining the timestamps and outcomes, and applying documented rules to confirm failures and assess attribution. These records provide evidence for review; they do not, by themselves, establish the cause of an incident.
A probe measures a route from a particular observation point. Measurements from one point cannot establish a provider-wide outage, and an attribution score is not proof of cause. Those boundaries are reflected in the product's methodology and in how results are reported.
Measurement principles
Clear findings, explicit limits.
Probes are issued from 1 observation point today, on a 300-second interval by default. Every page that reports a measurement says which window it covers and how many observations stand behind it.
Attribution is not causation
The attribution score compares timelines using five weighted signals. It does not establish that a provider caused an outage.
Measurements describe their observation path
A single observation point measures one route. It does not establish a provider-wide state or independent regional corroboration.
Availability includes its sample count
Availability figures are reported with the observation count. With no observations, the result is reported as insufficient data—not as 100%.
Unmeasured values remain unavailable
A field without a measurement is represented as unavailable, not as zero or as a mark that could be mistaken for zero.
Gaps in observation remain visible
Missed probes are not backfilled. Records disclose when their observation coverage is limited.
Product status
Available today and in development.
Available today
- Scheduled probes of external endpoints, with every observation retained.
- Deterministic fault confirmation, replayable from the stored rows.
- Deterministic attribution with published weights and a methodology version.
- Compiled evidence records with payload and document checksums, verifiable by a third party without an account.
- A public observatory that reports what it measured and states what it did not.
- A documented REST API, a working CLI, and webhooks.
In development
- Additional independently operated observation points to support agreement-based confirmation and identify the applicable rule in each record.
- Application-side signals (SDK, OpenTelemetry, webhook ingestion) so a dependency’s behaviour can be described from the caller’s side as well.
- Cross-application correlation: the same dependency degrading for unrelated applications at the same time.
- A dependency graph built from recorded observations, with attribution informed by available vantage-point coverage.
These are development goals, not current capabilities or live network measurements.
Research, methods, and implementation.
Read the technical work behind RELIASTRA, including its measurement methodology, measurement audits, and research on cloud identity and infrastructure security.
Prefer to read the source first? The measurement methodology is documented in the methodology guide, and the paper behind the detector is How RELIASTRA measures vendor reliability. Start observing if you would rather test it than read about it.