telemetry-leak-lab
What can survive log and trace redaction, and how to check without losing useful signals.
Explore case and demo ↗IX / ENGINEERING LAB
Confidence deserves
a difficult test.
Small tools for important questions. I explore how to demonstrate that a system responds correctly, especially when something goes wrong.
01 / FALSE-GREEN
Public alpha · verified locallyGreen tests.
Broken product?
One checkout. Five deliberate faults. Two ways to verify the same journey. The report shows why passing a test is not enough to trust what it checks.
SUPPLIED REPORT / 18 SEP 2026
0/5faults detected · weak suite5/5faults detected · behavior suite
Two repetitions per phase, with clean controls before and after. Reported results, not a new execution on this website.
The same fault. Different detection capability.
| Injected fault | Weak suite | Behavior suite |
|---|---|---|
| Checkout button does nothing | Not detected | Detected |
| Product image cannot decode | Not detected | Detected |
| Shipping information is invisible | Not detected | Detected |
| Confirmation copy is corrupted | Not detected | Detected |
| Explicit accessible label is removed | Not detected | Detected |
What this demonstrates and what remains to verify
The result describes sensitivity to five configured faults. The report’s combined 50% mixes two deliberately different suites; it is not a product quality or coverage score. A fault that never applies is neither detected nor survived.
The alpha was recovered and verified on September 19, 2026: clean installation, 71 unit, 18 offline browser and 20 HTTP tests passed on Windows with Node 24.15.0, Playwright Core 1.56.1 and Chromium 141.0.7390.37. The demo reproduced the displayed contrast. Local validation is not universal compatibility or an independent security audit. See the repository for the separate CI status.
02 / HEALTH-INTEROP-LAB
From a message to
a connected system.
A clinical interoperability lab already documented in the portfolio. The repository now includes a reproducible Angular → HTTP → gRPC journey, using synthetic data and explicit unavailable states.
Read the existing case ↗PUBLIC TOOLS / SYNTHETIC SECURITY EXPERIMENTS
tenant-fence
A business question made testable: can one organization read another’s records?
Explore case and demo ↗03 / RESEARCH AGENDA
The questions
that come next.
Six proposals to build and validate. Working names; these are not available products or delivery commitments.
evidence-contract
Contracts that distinguish facts, estimates and hypotheses before publication.
clinical-chaos
Adversarial synthetic scenarios for health integrations.
rules-replay
Explainable decision comparisons across rule versions.
route-atlas
Compare declared routes, links and observed journeys by role.
money-invariants
Idempotency, totals and rounding properties using fictional money.
mcp-failure-lab
Agent tool errors and retries without real-world effects.
PUBLICATION STANDARD
Let someone else
verify it.
Readable code, reproducible installation, synthetic examples and clear limits. Each tool will have its repository when it is ready to be tested.
My GitHub profile ↗FROM EXPERIMENT TO DECISION
Find out which boundary to review first.
The technical and strategic diagnostic turns a scoped question into a system map, prioritized findings and an actionable roadmap. Scope and access are agreed before any work.
Prepare my diagnostic ↗