# Engineering practice from systems in production Articles on the parts of building software that are rarely documented well: how failures arise and how to prevent them, the design decisions behind a system and their trade-offs, and what it costs to operate once real users depend on it. - Written by the people who built it - Every claim from a system in production - The trade-off, not just the choice The system most of these come from: [Building Jversity: a multi-tenant educational LMS platform](https://matiforge.com/work/building-jversity) ## Failure modes and how to prevent them Where systems tend to break, how the fault is found, and the structural change that prevents it. - [Enforcing tenant isolation beyond the database](https://matiforge.com/insights/tenant-isolation-by-construction) — Cross-tenant leaks tend to sit in caches, channels and queues, so isolation has to be enforced by the system in every shared layer. (2026-08-14, Architecture, 3 min read) ## Design decisions and their trade-offs Choices with a real trade-off: the options, the criteria that separate them, and a recommendation. - [Payments architecture for multi-tenant platforms](https://matiforge.com/insights/multi-tenant-payments-architecture) — Tenant-owned gateway accounts keep the platform out of the flow of funds, at the price of a provider-neutral core and disciplined reconciliation. (2026-09-09, Payments, 5 min read) - [Selecting a video provider for live classes](https://matiforge.com/insights/choosing-a-live-class-provider) — Select per class format and cost per participant-hour, and settle participant roles before designing the interface. (2026-08-31, Video, 3 min read) - [Designing AI features with cost controls and human oversight](https://matiforge.com/insights/ai-features-with-a-cost-ceiling) — Model the cost before writing the prompt, and keep a person accountable for every decision about a person. (2026-08-26, AI, 3 min read) ## Operating in production Cost, standards and practices that matter once software is live and people depend on it. - [Cost structure of a multi-tenant education platform](https://matiforge.com/insights/what-an-edtech-platform-costs-to-run) — Bundle what scales with storage and meter what scales with participant time; the arithmetic decides whether a subscription can work. (2026-09-02, Operations, 3 min read) - [Accessibility checklist for public web pages (WCAG 2.2 AA)](https://matiforge.com/insights/accessibility-checklist-for-public-pages) — Automated scanners cover only part of WCAG 2.2 AA; the rest needs a manual review and rules enforced on every page. (2026-08-22, Accessibility, 3 min read)