2026-09-27
QualityForge Lab #10: AUTH-T01–T04 in numbers
The first four HIGH-risk SDD v2 tasks in numbers: passes, corrections, human interventions, findings, and honest comparison limits.
02 / BLOG
Notes on quality and engineering.
Practical observations on automation, process, security, and building software.
↗
2026-09-27
QualityForge Lab #9: zero human interventions — what actually happened in AUTH-T03
What zero human interventions really meant in HIGH-risk AUTH-T03, and why that is not the same as full autonomy.
2026-09-27
QualityForge Lab #8: where the agent must stop
Three security decisions from AUTH-T01, T02, and T04 that required a human gate instead of agent improvisation.
2026-09-27
QualityForge Lab #7: SDD v2 — Orchestrator, Builder, and Reviewer
How risk-based SDD v2 separated Orchestrator, Builder, and Reviewer, bounded correction cycles, and enabled honest validation reuse.
2026-09-27
QualityForge Lab #6: Severity ≠ blast radius
Why finding severity determines importance but does not determine the scope of correction, validation, or repeated review.
2026-09-27
QualityForge Lab #5: when SDD became bureaucracy
How hosted CI, portability evidence, and repeated reviews turned useful SDD v1 controls into substantial process overhead.
2026-09-27
QualityForge Lab #4: Green tests, broken contract
How PLAT-T05 and PLAT-T06 passed automated checks while independent review still found contract violations.
2026-09-27
QualityForge Lab #3: from specification to acceptance — our first SDD workflow
Our first strict SDD baseline: a canonical specification, human gates, independent acceptance review, and evidence traceability.
2026-09-27
QualityForge Lab #2: local LLMs, Ollama, OpenCode, and Spec Kit — our first SDD traps
How a local-first SDD setup with Ollama, OpenCode, and Spec Kit turned model infrastructure into an experiment of its own.
2026-09-21
QualityForge Lab: why a QA engineer needs a testing playground
Why a QA portfolio should demonstrate more than code and tests by tracing each risk to a verifiable engineering result.