Lab / An engineering notebook
Questions under investigation, the implementation evidence available, and what remains untested. These topics are exploratory.
AI / Automation
What can an automated finding establish, and which decisions still belong to an analyst?
Evidence: AegisGraph validates bounded AI claims against incident evidence. CreditOps exposes recorded proposals and advisories for human review. The AegisGraph article preserves its original local live-provider checks and their limits.
Open question: How well do those boundaries hold across repeated model runs and a broader set of incidents?
Security
When should related signals become an incident, and what context makes that decision inspectable?
Evidence: The Detection Engineering case study documents tested Windows behavior rules, correlation, simulated triage, and scoped tuning. AegisGraph adds decision receipts for both formed incidents and unmet correlation conditions.
Open question: How should thresholds and correlation rules be evaluated against more varied, representative telemetry?
Fintech / Test engineering
What makes a passing test meaningful when requests, sessions, and order state can change concurrently?
Evidence: TradeFlow uses test-scoped sessions, exact-record assertions, independent OpenAPI validation, and controlled stale-response regressions across three browser engines.
Open question: Which additional asynchronous failure cases would reveal gaps beyond the lab's finite synthetic workflows?