MediaKit knowledge application
fix(applications): add dated Correction callouts to 13 of 14 applications/ articles — worst defect rate of any category swept (fabricated kill-switch UI, WordPress admin panel, C2PA badges, collab feature x3, stale Phase 4/category-count claims, wrong catchment radius, misattributed crates); 1 verified clean
@@ -46,6 +46,8 @@ references: url: "https://en.wikipedia.org/wiki/Wikipedia:Manual_of_Style/Layout" --- **Correction (2026-08-02):** three claims below don't match the real crate. (1) **Stale by weeks, not absent**: the article frames Phase 4 (history/blame/diff, MCP server, git smart-HTTP remote) as "planned... awaiting operator clearance" — but git history shows these shipped *before* this article's own `last_edited: 2026-05-25` date (history/blame 2026-05-03, MCP server 2026-05-12), and live routes confirm all of them (`/history/{*slug}`, `/blame/{*slug}`, `/diff/{*slug}`, `POST /mcp`, `git-upload-pack`) are deployed today. (2) **Fabricated**: the "Real-time collaborative editing" section (Yjs, `tokio::sync::broadcast` relay, `cm-collab.bundle.js`) and "squiggle linting" have zero code footprint anywhere in the crate — same class of finding as `patterns/collab-via-passthrough-relay.md`. (3) **Wrong location and content**: the "8 inventions" list is attributed to an `INVENTIONS.md` "at the crate root" — the real file is at `docs/INVENTIONS.md` and lists 5 differently-named inventions (Substrate-enforced AI grounding, Content-addressed federated AI adapters, etc.), `status: thinking`, not the 8 named here. **Flagged, not resolved.** `app-mediakit-knowledge` is the single-binary Rust wiki engine that serves PointSav's engineering documentation at `https://documentation.pointsav.com`. The engine combines an `axum` HTTP server, a `comrak` CommonMark renderer[^1][^2] with platform-specific extensions for wikilinks, footnotes, table of contents, and section anchors, a `tantivy` full-text search backend[^3], and a `maud` templating layer with four article templates. The engine reads markdown files from a content directory the operator names at startup, renders them on demand into HTML, and returns them with caching headers tuned for a documentation audience. The engine is a *view* over a markdown tree, not a content repository. The markdown tree is canonical; the running binary is a view that any number of operators can stand up over the same content tree, or different content trees, with no shared mutable state on the binary side. This source-of-truth inversion is the single most important design choice and is treated in detail in the next section.