diff --git a/docs/compare.md b/docs/compare.md index 77e028f..41024d1 100644 --- a/docs/compare.md +++ b/docs/compare.md @@ -24,7 +24,7 @@ Sources: [Help:Page history](https://en.wikipedia.org/wiki/Help:Page_history), | **Find when a sentence first appeared** | Not built in — on English Wikipedia, the history page's "Find addition/removal" link opens WikiBlame | `refract claim "Page" --text "sentence"` → exact revision + timestamp | | **Track a sentence across its entire lifecycle** | Manual — follow the page history | Automatic — first seen, modified, removed, reintroduced, all timestamped | | **Detect citation swapping** | Manual — compare each diff's reference section | `citation_replaced` event — `before`/`after` show the old and new source | -| **Detect edit wars** | Partly — undos, rollbacks and manual reverts are tagged, and so are the edits they revert; spotting back-and-forth is manual | `revert_detected` + `edit_cluster_detected` — automatic structural detection | +| **Detect edit wars** | Partly — undos, rollbacks and manual reverts are tagged, as are the edits they revert; spotting back-and-forth is manual | `revert_detected` + `edit_cluster_detected` — automatic structural detection | | **Correlate article edits with talk page discussion** | Manual — check Talk tab separately | `talk_page_correlated` — Refract checks 7 days before / 3 days after each edit | | **Compare the same topic across language editions** | Manual — open each wiki separately | `refract diff` — cross-wiki comparison with z-score outlier detection | | **Query with SQL** | No | DuckDB: `SELECT "eventType", count(*) FROM 'events.jsonl' GROUP BY 1` | @@ -37,8 +37,8 @@ Sources: [Help:Page history](https://en.wikipedia.org/wiki/Help:Page_history), | Tool | What it does | Refract's difference | |---|---|---| | **[XTools](https://www.mediawiki.org/wiki/XTools)** | [Page History](https://www.mediawiki.org/wiki/XTools/Page_History): statistics about a page's edits, such as top editors and edits per year and month. [Authorship](https://www.mediawiki.org/wiki/XTools/Authorship): each editor's share of the current text, by character count. [Blame](https://www.mediawiki.org/wiki/XTools/Blame): the edits that added a given text. Authorship and Blame use WikiWho | Refract reports each change between revisions as a typed event and does not attribute text to editors | -| **[WikiBlame](https://en.wikipedia.org/wiki/User:Flominator/WikiBlame)** | Searches a page's revisions, by binary or linear search, for the one in which a given text was inserted or removed. Works on MediaWiki wikis | `refract claim` follows one sentence through every revision: first seen, modified, removed, reintroduced | -| **[Who Wrote That?](https://www.mediawiki.org/wiki/Who_Wrote_That%3F)** | Browser extension for Chrome and Firefox that highlights who wrote each part of an article's current text, using WikiWho | Refract does not identify editors; it reports what changed at each revision | +| **[WikiBlame](https://en.wikipedia.org/wiki/User:Flominator/WikiBlame)** | Uses binary or linear search over a page's revisions to find the one in which a given text was inserted or removed. Works on MediaWiki wikis | `refract claim` follows one sentence through every revision: first seen, modified, removed, reintroduced | +| **[Who Wrote That?](https://www.mediawiki.org/wiki/Who_Wrote_That%3F)** | Browser extension for Chrome and Firefox that uses WikiWho to highlight who wrote each part of an article's current text | Refract does not identify editors; it reports what changed at each revision | | **[WikiWho](https://www.mediawiki.org/wiki/WikiWho)** | Token-level provenance for about 70 Wikipedia editions: who added, removed or reinserted each token, and in which revision | Refract emits events for sentences, citations, templates, reverts and sections rather than tokens, and does not attribute editors | | **[WhoColor](https://github.com/wikiwho/WhoColor)** | Userscript that colors article text by author, with conflict and age views, built on WikiWho | Refract structures the data for querying, not just viewing | | **[Wikimedia Enterprise](https://meta.wikimedia.org/wiki/Wikimedia_Enterprise/FAQ)** | APIs that deliver Wikimedia project content: daily project snapshots, single articles on demand, and a real-time stream of updates. Free access, and paid plans with service-level agreements | Refract analyzes a page's past revisions, is open-source, and runs locally | diff --git a/docs/mcp.md b/docs/mcp.md index a62b8b0..d4a54ef 100644 --- a/docs/mcp.md +++ b/docs/mcp.md @@ -1,6 +1,6 @@ # Wikipedia & MediaWiki revision-history MCP server -`refract mcp` starts a Model Context Protocol server over stdio that lets AI agents analyze page revision histories from Wikipedia or any MediaWiki API. +`refract mcp` starts a stdio Model Context Protocol server that lets AI agents analyze page revision histories from Wikipedia or any MediaWiki API. The MCP server supports **sampling** — it can request the host's LLM to interpret events at any BYO-inference boundary without managing API keys. diff --git a/docs/tutorials/wikipedia-history.md b/docs/tutorials/wikipedia-history.md index 5c49642..2b83495 100644 --- a/docs/tutorials/wikipedia-history.md +++ b/docs/tutorials/wikipedia-history.md @@ -3,8 +3,9 @@ ## Goal Use Refract to analyze the full revision history of a Wikipedia page and understand what -changed, when, and what kind of change it was — all deterministically, byte-for-byte -reproducible. +changed, when, and what kind of change it was. + +The analysis is deterministic, and its output is reproducible byte for byte. ## Steps