Jump to content

Help:Abstract Data guide

From Abstract Wikipedia

This page explains how to use Abstract Data, a Toolforge tool that turns the state of Abstract Wikipedia into a worklist: which articles already render, which ones do not, and which edits would unblock the ones that do not.

  • Tool: https://abstract-data.toolforge.org
  • Source: toolforge-repos/abstract-data
  • Data: recomputed nightly from the Abstract Wikipedia and Wikifunctions APIs and from Wikidata (query service and API), with the database replicas used only for revision history and article creation dates. The sidebar shows how old the current snapshot is ("Last sync").

The tool never asks you to log in and never edits anything. Every action button hands you off to Wikidata or Wikifunctions, where you make the edit.

What this tool is

[edit]

Abstract Data is a side project built by one member of the Abstract Wikipedia team, mostly by prompting an AI coding assistant ("vibe coded"), to give contributors a view into Abstract Wikipedia that didn't exist anywhere else. It's not an official team deliverable, and it wasn't built with engineering review or production support. It's a proof of concept: the numbers are a best-effort reconstruction of a complex system, so please treat anything surprising as a lead to verify against Wikidata, Wikifunctions, or the rendered article. Bug fixes for this tool will happen on a best-effort basis. If the useful parts prove their worth, the goal is to fold them into Abstract Wikipedia properly at a later stage.

Before you start: five words the tool uses

[edit]

An abstract article is not written as text. It is a set of function calls, and a language only gets a sentence when every piece those calls need exists in that language. Five pieces come up constantly:

  • Label: an item's name in your language on Wikidata (for example, Q90 is "Paris" in English, "París" in Spanish). Without it, the sentence either borrows another language's word or prints the bare QID.
  • Lexeme: a real word on Wikidata, with its grammar, so a sentence can inflect it (plural, case, gender). Some renderers need a lexeme, not just a label, and there is nothing to borrow if it is missing.
  • Configuration: a Wikifunctions object that says, for one sentence pattern, which function to use per language. If your language is not listed, the pattern either falls back to generic default wording or produces nothing.
  • Implementation: the actual code behind a function on Wikifunctions.
  • Renders / ready: an article is ready in a language when all of the above exist for that language, so the article can be shown as prose there.

"Via default" is worth a second look: the sentence does appear, but in wording nobody wrote for your language. The tool counts that as work still to do, because it is.

The landing page: two ways in

[edit]
The landing page asks two questions. Answer either one.

The tool opens by asking two questions, and they correspond to the two ways people actually arrive:

  1. "Which language do you want to contribute to?" You speak a language and want to know what it needs. Follow Path 1.
  2. "Which article are you trying to improve?" You care about one topic and want to know which languages it is blocked in. Follow Path 2.

Everything else in the tool is reachable from the left sidebar: Overview, Abstract Articles, Functions, Languages and Historical Stats. The search box at the top right (or ⌘+K / Ctrl+K) finds articles, functions and languages by name or by ID from anywhere.

Path 1: start from your language

[edit]

Step 1. Pick the language

[edit]
Languages that have at least one Wikifunctions configuration, with the number of articles that currently render in each.

Open the dropdown and pick your language. Type to filter; the number on the right is how many abstract articles already render in that language.

Your choice becomes the working language for the whole tool. It is carried in the URL (?lang=es) and shown in the top bar as "Viewing es Spanish", so every page you visit afterwards (articles, functions, languages) is framed for that language until you change it. To go back to an unscoped view, choose "No language (show all)" in the top-bar picker.

Step 2. Read the top of the Overview

[edit]
The headline sentence and the six counters for the working language.

The Overview opens with one sentence: how many articles render in your language today, and how many of those have no article on your Wikipedia yet. That second number is the gap Abstract Wikipedia can fill right now.

The six counters underneath are all scoped to your language:

CounterWhat it counts
RenderableArticles that can be shown in your language today.
Missing from WikipediaOf those, the ones your Wikipedia does not have as an article. High-value publishing targets.
Pending labelsWikidata items used by articles that have no name in your language.
Pending lexemesItems whose renderer needs a word form in your language that nobody has created.
On the config defaultArticles that do produce a sentence, but in generic default wording rather than yours.
Vital knowledgeHow many of the ~1000 "articles every Wikipedia should have" render in your language via Abstract Wikipedia.

Step 3. Work the "Highest-leverage fixes"

[edit]

This is the heart of the tool. Instead of listing broken articles, it lists the missing pieces, ranked by how many articles each one would unblock. Fix the top row and every article waiting on it improves at once.

Two of the four cards. The number on each row is how many articles need that piece; the button takes you where the edit happens.
The other two cards. "Missing implementations" needs a Wikifunctions editor rather than a Wikidata edit.

The four cards, in the order most people can act on them:

  • Missing labels → "Add label" opens the item on Wikidata. The quickest possible contribution: one name, in your language, which can unblock several articles at once.
  • Missing lexemes → "Create lexeme" opens Special:NewLexeme with the lemma prefilled. Needs a little grammar knowledge of your language.
  • Add your language to a configuration → "Add language" opens the configuration object on Wikifunctions. Rows tagged via default already make a sentence in generic wording; adding your language is what makes the sentence your language's own.
  • Missing implementations → the function your language dispatches to has no code yet. This one needs someone comfortable writing a Wikifunctions implementation.

Each card has a How-to guide button. It opens a step-by-step drawer inside the tool, screenshots included, and ends with a button that takes you to the right page on Wikidata or Wikifunctions. If you have never added a label or created a lexeme before, start there.

The in-tool how-to drawer for adding a label. There are equivalent guides for lexemes and for adding a language to a configuration.

Step 4. Hunt specific gaps in "Abstract Articles"

[edit]

The Overview answers "what should I fix first?". The Abstract Articles page answers "show me exactly the articles with this problem".

Abstract Articles, scoped to Spanish, filtered to articles with a missing label.
The filter bar. Issue filters combine with scope filters and the free-text box.
  • Issues: tick Missing label, Missing lexeme, Missing impl or No implementation to keep only articles with that kind of gap. Tick several to widen the net.
  • Scope: ★ Vital 1000 / ★ Vital 10000 restrict to the "articles every Wikipedia should have" lists; ✚ Missing from your Wikipedia restricts to articles that render in your language but have no article on your Wikipedia yet.
  • Filter articles…: free text over QID, title and the functions an article uses.
  • Columns are sortable: Languages (how many of the configured languages the article is ready in), Root fns (how big the article is, in top-level calls), Created.
  • The Issues column tags each row, and the tag counts tell you how much work one article needs. Click any row to open the article.

Step 5 (optional). The language detail page

[edit]
One language, in full: articles your Wikipedia is missing, every pending fix, and the configurations that do not list your language.

Languages → click your language for the complete picture: the articles renderable in your language that your Wikipedia lacks (with a link to create each one), a paginated table of every pending fix with its blocker type and an action button, and the list of configurations that exist for other languages but not yours. This is the exhaustive version of the Overview's top-ten lists.

Path 2: start from an article

[edit]

Step 1. Search for the article

[edit]
Searching by name or QID. Exact matches come first.

Type the article's name or its QID in "Which article are you trying to improve?" on the landing page (or in the top-bar search from anywhere). Suggestions show the QID and the item's name; pick one to open its article page.

Step 2. See the article across all languages

[edit]
Q90 Paris with no language selected: what it would take to support more languages at all.

With no language selected, the article page is the "all languages" view. The banner says whether the article is ready anywhere, and the sections below list what stands between this article and more languages:

  • Add languages: the configuration objects this article depends on. Editing one of these on Wikifunctions is what extends the article to a new language.
  • Add Wikidata labels: the items this article names, with a chip per language saying what that language currently gets instead: another language's word, the language-neutral mul one, or the bare QID.
  • Add Wikidata lexeme: items that need a word form, and the languages where it is missing.
  • Article metrics and Function tree: how the article is built, in distinct functions, root calls, depth, and the call tree itself. Useful when you want to understand why a dependency exists.

Step 3. Switch language in the "Viewing" dropdown

[edit]
The "Viewing" dropdown reframes the page for one language at a time.

This is the move that makes Path 2 work: pick a language in the top-bar Viewing dropdown and the same article is re-read for that language alone. Switch through the languages you know to see where you personally can help.

Step 4. Read "To unblock this article"

[edit]
The same article, scoped to Croatian: everything standing between it and a Croatian rendering.

With a language selected, the page becomes a checklist for exactly that pair (this article, this language). Every row is one missing piece with the action that resolves it, grouped by kind: language not configured, missing label, missing lexeme, missing implementation. Rows also say what happens today: "via default function" for generic wording, "shows the bare QID" for nothing at all.

Clear the list and the article renders in that language. For a small article that can be two or three edits.

Step 5. What "ready" looks like

[edit]
A green banner, and no work left for this language.

A green banner and "No language-specific issues" means every requirement is met in the language itself: nothing borrowed, nothing defaulted. "View on Abstract Wikipedia in language" opens the rendered article so you can read the result and check the wording.

Appendix A: Historical Stats

[edit]
Growth and edit activity, weekly or monthly.

Historical Stats is computed from revision history, and answers questions about the project rather than about your language:

  • Article growth: cumulative articles over time, and new articles per period. Toggle Week / Month top right.
  • Article creation: how many articles were created in each period; useful for spotting campaigns and slowdowns.
  • Edit volume: initial versus follow-up edits. A high follow-up share means articles are being revised, not just created.
  • Multi-contributor articles: how many articles more than one person has touched, as a share of all articles.
  • Growth by month: the same data as a table, with period-over-period change, if you want the numbers rather than the shape.

Appendix B: the Functions pages

[edit]
The 50 most-used functions, with type and status.

Functions ranks Wikifunctions ZIDs by how often abstract articles use them. It is the page to open before changing a function, and the counters at the top are a quick health check: how many functions are tracked, and how many have no implementation at all.

  • Direct: how many articles call this function themselves. Transitive: how many reach it through other functions. A function with a small direct count and a large transitive one is load-bearing infrastructure.
  • Languages: for configurable functions, how many languages the configuration lists. Click through to see which.
  • Type: Configurable by language (behaviour is chosen per language), Terminal (a code implementation), Built-in (native).
  • Filter by type with the chips above the table; sort by any column.
One function: usage, signature, and the call tree beneath it.

A function's own page adds:

  • Articles using it: the impact number to check before an edit, with the article list further down the page.
  • Signature: inputs (name, key, type) and return type, so you can read a call without opening Wikifunctions.
  • Call tree: what this function calls, directly and transitively, up to four levels, colour-coded by type. "Expand all" opens the whole tree. A red "no implementation" node anywhere in the tree explains a failure that looks mysterious at the top.
  • Tests: whether testers are failing, when that information is available.

Appendix C: the Languages pages

[edit]
All tracked languages, sortable by any column.

Languages is the cross-language comparison: one row per language with at least one configuration.

  • Renderable: abstract articles that render in that language.
  • Not in Wikipedia and Gap: how many of those the language's Wikipedia lacks, in absolute terms and as a share. A high gap means Abstract Wikipedia has a lot to offer that community right now.
  • Missing labels / lexemes / impl.: the size of each backlog.
  • Vital 1K / 10K, on wiki vs via AW: of the vital lists, how many the local Wikipedia already has, and how many Abstract Wikipedia can render. The difference is the argument for deploying to that language.
  • The chips above the table give quick cuts: ≥ 500 articles, High gap (≥ 40%), Low gap (< 15%). Click any row for the language detail page described in Path 1, Step 5.

Notes and caveats

[edit]
  • Freshness. Everything is a nightly snapshot. An edit you make on Wikidata or Wikifunctions shows up in the tool after the next recompute, not immediately.
  • Renderability is computed, not executed. The tool derives what an article needs by reading stored Wikifunctions definitions, never by running the functions. That is deliberate, and it keeps the tool fast and harmless, but it means a few dependencies that only appear at run time (inside higher-order or language-dispatched code) are approximated from a curated map rather than observed. Treat a "ready" verdict as a strong signal, not a guarantee, and read the rendered article when it matters.
  • "Via default" is not "done". A sentence in the configuration default's wording is a sentence nobody wrote for your language. The tool counts it as pending on purpose.
  • Only an own-language label counts. A label reached through the fallback chain, or the language-neutral mul one, is reported as a gap: the sentence is carrying a word from somewhere else.
  • Found a bug, or want a number the tool does not show? File a task on Phabricator under #Tool-abstract-data, or open an issue at toolforge-repos/abstract-data.