Writing

Swap on a Modern Linux Server: zram, a Swapfile, and When to Use Which

October 2, 2026

Cloud images ship with no swap, so processes get OOM-killed under transient pressure. Here is a layered fix: compressed zram, a disk swapfile fallback, and server-appropriate swappiness.

From DPIA to FRIA to CRA Risk Assessment: One Risk-Assessment Engine for Three EU Regimes

October 1, 2026

GDPR DPIAs, AI Act FRIAs and CRA product risk assessments share most of their structure. One risk-assessment engine can capture the common inputs once and emit the specific artefact each regime expects.

Embedded Insurance APIs: Governing the Distribution You No Longer Directly Control

October 1, 2026

Selling insurance through third-party platforms via API pushes IDD conduct, disclosure and data duties to a distribution edge you no longer directly control. How to make the control model travel with the API.

Installing Python CLI Tools the Sane Way With pipx (and When uv tool Replaces It)

October 1, 2026

Stop installing command-line Python apps into your global interpreter. pipx gives each tool its own venv, and uv tool does the same job faster — here is how to run both, and which to standardise on.

AI Sandboxes Become a National Obligation: What an AI Act Regulatory Sandbox Offers Your Deployment

October 1, 2026

Every member state must run an AI regulatory sandbox by 2 August 2026. Here is what an AI Act sandbox actually gives a novel high-risk system, and when it is worth applying.

Compliance Convergence 2026-2027: A Single Control Fabric for AI Act, CRA, NIS2 and the Data Act

October 1, 2026

The AI Act, CRA, NIS2 and Data Act arrive with heavily overlapping controls. Run one control catalogue mapped to every obligation rather than four siloed programmes.

mTLS Between Services: Mutual TLS for Internal APIs Without a Service Mesh

October 1, 2026

You want two internal services to authenticate each other, but a full service mesh is overkill. mTLS with a small CA and nginx gets you there without the sidecar tax.

The AI-BOM: A Bill of Materials for Models, Weights and Training Data

October 1, 2026

A software SBOM captures your libraries but says nothing about the model, its weights or its training-data provenance. Here is how to build a machine-readable AI bill of materials that answers what model, trained on what data, is running where.

The sshd_config Hardening Reference: Every Directive That Matters, With the Value and the Why

September 30, 2026

A directive-by-directive reference for a hardened OpenSSH 9.6 sshd_config: auth, crypto, access scope, brute-force limits and forwarding, each with a value and the why.

LLM Evaluation Gates: Promptfoo Regression Testing for Prompts and Chains

September 30, 2026

In most firms, prompt and chain changes ship with no regression testing at all. An LLM eval gate built with Promptfoo scores every change in CI and blocks the quality drop before a customer sees it.

Previous Next