# Wiki Articles An AI editorial room where ideas are challenged before they become copy ## Project brief - **Problem:** Valuable ideas disappear in compressed conversations or reach publication after their reasoning, source context, and uncertainty have been flattened. - **System:** A Git-backed editorial operating system separates source capture, argument review, writing, metadata, localization, and publication into traceable stages. - **Input:** Raw dialogue, screenshots, claims, objections, drafts, project evidence, editorial feedback, and publication-contract changes. - **Output:** Approved articles and project stories as versioned English, Russian, and Latvian Markdown snapshots for Portfolio. - **Evidence:** The current workspace maintains 10 published articles and 8 project stories, with reviewed Russian and Latvian editions for every article, stable source identities, export validation, and live feedback returning to canonical sources. - **Status:** Working internal system for one author. Multi-user editorial governance, universal CMS use, and fully automated semantic verification remain unvalidated. - **Full story:** /projects/wiki-articles/story/ ## Full story Wiki Articles began with a practical limitation of chat-based work. A valuable conversation contained not only a conclusion, but also doubts, objections, revisions, and the context that made the conclusion meaningful. When chat history was compressed, the answer could survive while the reasoning behind it disappeared. The project was built as an editorial operating system between live dialogue and public publication. It preserves source material, separates argument review from writing, and turns approved evidence into versioned articles and project stories. ## From conversation to governed source Every material begins with evidence: dialogue, screenshots, claims, objections, drafts, or verified project state. The original source is preserved before editorial work starts. It can then become a readable transcript, a research note, a stable idea, or a structured project snapshot. Canonical records retain identity, dates, workflow status, relationships, and publication history. A polished article is produced only after the source has been understood and the relevant claims can be traced back to it. This order prevents confident prose from replacing incomplete evidence. It also keeps corrections possible because the published result never becomes the only remaining version of the idea. ## Reasoning, writing, and publication stay separate Wiki Articles divides editorial work into distinct responsibilities. Dialogist tests premises, exposes weak reasoning, and checks whether an attractive formulation is hiding an unsupported claim. Article Copywriter turns an argument that has survived review into clear public prose. Markdown Knowledge Operator maintains structure, metadata, relationships, localization state, and publication contracts. Final approval remains human. The roles provide different forms of pressure and control, but they do not decide which interpretation becomes the author's public position. The separation matters because a fluent model can challenge an idea, rewrite it, structure it, and publish it in one pass. That speed is useful, but without boundaries it can also flatten uncertainty and make an unfinished argument look complete. ## Portfolio is the quality gate A finished Markdown file is not necessarily a finished public artifact. Rendering the material on a real website exposes problems that remain easy to miss in a repository: weak titles, repeated summaries, internal terminology, broken rhythm, misleading dates, unsupported claims, or metadata that only makes sense inside the source project. Those defects return to the canonical source. Repeated corrections become publication rules rather than one-off edits. Public titles follow a shared contract, original dates survive export, internal paths and unfinished links are blocked, and short descriptions are kept distinct from the Full Story. Portfolio therefore acts as a quality gate, not as the owner of project meaning. Source projects provide approved facts and evidence. Wiki Articles develops the public narrative. Portfolio owns rendering, navigation, and presentation. ## One source supports several readers An approved project snapshot can produce a short landing page, a detailed Full Story, and a compact context document for AI assistants. The same source also controls English, Russian, and Latvian editions. These surfaces serve different readers without creating separate interpretations of the project. A visitor can scan the landing page, a technical reviewer can inspect the complete story, and an AI system can retrieve bounded context from the machine-readable endpoint. Stable source identities connect every exported edition to its canonical material. Localization records preserve review state and source dates, while validation checks structure, metadata, unsupported Markdown, and missing language coverage before delivery. ## The system expanded beyond articles The first use case was preserving a conversation and developing it into an article. The same workflow later proved useful for project stories. Agent infrastructure, operational knowledge systems, design systems, simulation tools, and mobile applications all contain more evidence than a conventional portfolio card can show. Wiki Articles turns that internal material into a readable account of the problem, architecture, decisions, failures, working capabilities, and current limits without taking ownership away from the source project. The current workspace maintains ten published articles and eight project stories. Every article has reviewed Russian and Latvian editions, and project localization is expanding under the same contract. The important output is not the number of Markdown files. It is the preserved chain from unfinished thought to public claim: where an idea came from, how it was tested, which evidence supports it, what was approved, what became public, and how later corrections return to the source. ## Current boundary Wiki Articles is a working internal system for one author. It has not been validated as a multi-user editorial platform, a universal CMS, or an autonomous publisher. Source quality and semantic accuracy cannot yet be verified completely by automation. Delivery to Portfolio remains controlled, and a human still determines which claims are accurate, which uncertainty must stay visible, and when a story is ready for publication. The system does not replace editorial judgment. It preserves the evidence, structure, and review history that allow that judgment to remain visible from the first conversation to the final public page.