status-translations

Translation-progress dashboard for the QuantEcon lecture series — public data store + GitHub Pages site tracking coverage, freshness, automation and review state for every language edition.

Live dashboard: https://quantecon.github.io/status-translations/

This repo follows the org's status-* pattern (sibling of status-lectures, which tracks build/environment configuration): data lives in-repo as JSON, the dashboard is a static page on GitHub Pages, and no credentials beyond GITHUB_TOKEN are needed — every fact is computed from public GitHub state.

What it answers

Translation health is two-dimensional: coverage (how much of the source series is translated) and freshness (how current those translations are against the moving English source). An edition can be 100% translated and rotting, or 65% and perfectly synced. The dashboard shows, per source series → edition:

• coverage (lectures translated / total, with per-lecture detail)
• last sync and upstream drift (source PRs merged since)
• automation wiring (sync / AI review / rebase workflows)
• human-review backlog (translation-review issues)
• a synthesised status (healthy / drifting / stalled / planned), plus the language-expansion pipeline

How it works

1. collector/collect.py blobless-clones each source repo and its editions, compares lectures/*.md file sets and per-file last-commit dates, checks workflow wiring, and counts sync-era PRs and review issues.
2. It writes data/latest.json and a dated snapshot in data/history/ — history accumulates from day one, so trend views come free later.
3. The site (static, no dependencies) fetches data/latest.json and renders three pages — Overview (stat tiles, series × edition table, language pipeline), Rollout (per-edition lifecycle steppers: scaffolded → seeded → automated → published → reviewed), and Sync detail (per-lecture cells + orphan files). The publish workflow deploys site/ + data/ to Pages on every push, and the collect workflow redeploys after committing fresh data. Every page carries a freshness badge (green ≤ 7 days since collection, orange past a week, red past two) and a light/dark theme toggle (system preference by default).

Editorial facts the collector can't compute (series titles, phase labels, review-campaign notes, the language pipeline) live in collector/config.json. To add a language or series, extend that file.

Layout

data/
  latest.json         # current snapshot (the dashboard reads this)
  history/            # dated snapshots, append-only
site/
  index.html          # Overview page (static, client-side rendering)
  rollout.html        # Rollout tracker: per-edition lifecycle steppers
  sync.html           # Sync detail: per-lecture cells + orphan files
  style.css           # shared design system (light/dark themes)
  app.js              # shared chrome: nav, freshness badge, theme toggle, data loading
collector/
  collect.py          # computes the numbers (git + gh + python stdlib)
  config.json         # repo pairs + editorial fields
.github/workflows/
  publish.yml         # deploy site + data to Pages on push
  collect.yml         # nightly data collection (cron, 03:17 UTC) + redeploy

Counting rule

A "lecture" is a top-level lectures/*.md file in the source repo, excluding README and status (the build-status page). Coverage counts files present in both source and edition; edition-only files are reported as orphans (e.g. pre-rename leftovers). Program docs have historically counted raw .md files including status, so figures here can sit one lower (48/51 vs 49/52) — this repo's rule is the canonical one for reporting.

Status & verification

• The collector was verified on 2026-07-18 against the hand-seeded 2026-07-09 snapshot: coverage and orphan lists reproduced exactly; the computed per-lecture staleness supersedes the seed's hand-set values. The nightly cron in collect.yml has been on since.
• Two measurement caveats, stated in the data itself (last_sync_note, drift_approx): the per-edition activity date is the newest commit touching a matched lecture file in the edition, so edition-local maintenance moves it too (read it as activity, not sync — per-lecture staleness is the reliable currency signal); and drift-PR counts are approximate by design (all merged source PRs since last activity, not only lecture-touching ones). The translation-sync-metadata contract is the upgrade path to exact per-file drift.

Placement (provisional)

This repo is the dashboard's current home, not a permanent commitment. How QuantEcon organises project and dashboard reporting is still being worked out (see the reporting-strategy discussions at meta#332 and meta#321); if a central reporting hub emerges (e.g. under QuantEcon/dashboard), this dashboard is designed to relocate cheaply — the static site and the versioned data contract (data/latest.json + data/history/) are self-contained, so moving means moving files and updating one URL.

Related repos

• action-translation — the translation engine (sync / review / rebase); its sync metadata (#66) is this dashboard's upgrade path for exact drift.
• project-translation — private program workspace: roadmap, decisions, research; this dashboard replaces its hand-maintained coverage tables as the live status layer.
• status-lectures — build/environment-configuration dashboard for the same repo family.
• dashboard — curated presentation hub; links here for translation progress.