Origin brief
MDL Wiki exists for the gap between model performance and model evidence.
Most AI projects produce a strange mix of polished demos and scattered judgement. The demo shows what is possible; the scattered judgement explains whether it should be trusted, maintained, extended, or slowed down. MDL Wiki was shaped for that second layer. It collects the language of model documentation, dataset review, learning workflow, evaluation design, and operational caution in one concise reference space.
Plain first
Entries start with ordinary language before they move into technical distinctions that matter.
Evidence aware
Claims are framed around what would need to be checked, measured, sampled, or logged.
Decision useful
A definition should help a team compare, review, escalate, pause, or ask a better question.
MDL stands here for model, data, and learning because those three concerns rarely stay separate in real systems. A model reflects the data it absorbed. A dataset reflects the assumptions and labor that shaped it. A learning process reflects the incentives of the people tuning, evaluating, and shipping it. When teams talk about only one layer, the other two usually return later as incidents, rework, or confusing stakeholder debates.
The site is intentionally compact. It does not try to replace textbooks, research papers, vendor manuals, or formal governance frameworks. Instead, it offers connective tissue: short explanations, naming discipline, review prompts, and careful summaries that make technical conversations less brittle. A reader should be able to leave with a sentence they can use in a meeting and a sharper sense of what evidence is still missing.