Changelog

1. 2026-07-29

   API

   v1.1.0

   One credit is one row

   #

   Breaking change

   Credits are rows now, not hundreds of rows, and they are sold by the amount you type rather than in packs. Existing balances were multiplied by 100 so nobody lost anything.

   Added

   • GET /account/v1/rate — the whole pricing model as data: the rate per currency, the step, the bounds and worked examples. It answers whether or not payments are switched on, because a pricing page that 503s is a marketing outage.

   Changed

   • A credit is one observation row. GET /v1/data now costs max(1, rows) — the number of credits is the number of lines in the response — where it used to cost max(1, ceil(rows / 100)). Nothing else moved: rankings and country detail are still a flat 1, and metadata is still free.
   • Every existing balance was multiplied by 100 in the same change, so a balance bought under the old unit buys exactly the same data as it did before. It appears in your ledger as one admin_adjust line reading Credit unit change — a restatement, not a grant. Entries either side of it are denominated in different units, which is why that line is there to find.
   • Nothing rounds any more. One request of 1,200 rows and twelve of 100 now cost the same 1,200 credits, so chunk size is an engineering decision rather than a pricing one.
   • Packs are withdrawn; you top up by amount instead. Type what you want to spend — any whole multiple of ₹100 or $1, up to ₹500,000 or $5,000 — and it buys credits at a flat rate with no tiers and no volume discount: ₹1 = 1,000 credits, $1 = 100,000 credits.
   • Prices include tax. The amount you type is the amount charged is the amount on the invoice. Where GST applies it is extracted from that amount rather than added to it — ₹100 is ₹84.75 of taxable value plus ₹15.25 of GST — and a purchase from outside India is a zero-rated export. Credits granted follow the amount paid and never the tax split.
   • The free grant at email verification is unchanged at 2,000 credits, which is now 2,000 rows.
   • The documentation moved with the model: /docs/credits carries the rate, the per-endpoint cost table, the worked arithmetic and the X-Credits-* headers, and the pricing, refund and terms pages no longer describe packs.
2. 2026-07-28

   API

   v1.0.0

   Accounts, API keys and credits

   #

   worldgovdata becomes a developer platform. Create an account, mint a key, and call the same data the portal itself runs on.

   Added

   • Accounts: email sign-up with verification, password reset, and a settings page that owns your profile, your consent, your export and your deletion.
   • A public metered API at https://api.worldgovdata.com/v1 — countries, indicators, topics, rankings, observations and metadata, in the same response shapes the portal reads.
   • API keys shaped wgd_live_…, up to ten live at once. The full key is shown once at creation and never again, and any key can be revoked on the spot.
   • Credits, priced in rows: one credit covers about 100 observation rows, so GET /v1/data costs max(1, ceil(rows / 100)). Verifying your email grants 2,000 of them.
   • The catalogue stays free and keyless. /v1/countries, /v1/indicators, /v1/topics and /v1/meta/* cost nothing and answer without credentials, so browsing to find out what to buy never costs money.
   • Published rate limits — 30 requests a minute anonymously, 120 with a verified account, 600 once an account has bought credits — with X-RateLimit-* and X-Credits-* headers on every metered response.
   • A documentation site at /docs: quickstart, authentication, credits, rate limits, errors, pagination, licensing, a page per endpoint, and copy-paste clients in cURL, JavaScript, Python, R and Go.
   • Pricing at /pricing: three one-off credit packs — starter 25,000, growth 120,000, scale 500,000. No subscription, no auto-renew, no stored cards, and credits do not expire.
   • Your usage, in your account: balance, a daily chart, a per-endpoint breakdown, and the credit ledger in full — every grant and every debit, with the balance after each one.
   • This changelog, filterable by category, with an RSS feed at /changelog/rss.xml.

   Changed

   • The legal pages are rewritten and expanded: full terms, a GDPR-grade privacy policy, cancellation and refunds, a delivery statement, acceptable use, cookies, named subprocessors, and a data-processing addendum for API customers.
   • Checkout is not open yet. The packs and their prices are real and published, but card payments switch on shortly — until then credits are granted manually on request.
   • The licence gate is unchanged and still absolute. Indicators held under display-only terms are never exported, and buying credits does not unlock a licence we do not hold.
3. 2026-07-28

   Fixes

   v0.6.1

   Reliability: deploys, indexes, and one empty card

   #

   Four quiet failures, three of them invisible from the outside. The invisible ones are the expensive ones.

   Fixed

   • Five topics no longer render an empty card on every country page. They had no featured indicator to show, and the placeholder read as a stalled fetch rather than as a gap in curation.
   • Deploying the ingest image now means the published tables match the transformation models inside that image, rather than whatever the last successful build happened to leave behind.
   • The boot-time build says whether it passed. A failure used to be indistinguishable from a success.
   • Index definitions on the fact tables apply on every rebuild again — they had been silently doing nothing, which is the kind of silence that shows up as a slow page months later.
4. 2026-07-28

   Data

   v0.6.0

   Projections

   #

   The forecast half of the catalogue is now drawn as forecast: separated from observed history, labelled, and never counted as fact.

   Added

   • Projection series are drawn as projections — held apart from observed history in charts and tables, and labelled, instead of being blended into one continuous line.

   Fixed

   • A projected value can no longer be presented as a country’s or an indicator’s latest data.
   • A single scenario series can no longer set the freshness date for an entire provider.
5. 2026-07-23

   Fixes

   v0.5.1

   Faster pages, and a cache that tells the truth

   #

   A country page was making roughly fifty requests to render four panels, and a momentary API blip could be remembered as a permanent 404.

   Changed

   • Country pages fetch 4 rankings instead of about 50, cap their sparkline fan-out at 12 series, and prefetch those in a single batch.
   • Country pages are incrementally regenerated again rather than rendered from scratch on every request.
   • The API connection pool is sized explicitly at 20, up from a default of 4.

   Fixed

   • A transient API failure could be cached as a permanent 404. It cannot any more.
   • Fact-table indexes that a warehouse rebuild silently dropped are restored.
   • Topic classification was re-audited end to end and the misfiled indicators reclassified.
6. 2026-07-23

   Platform

   v0.5.0

   SPINE: a new design system, and a clearer map of the site

   #

   The whole site moved onto one type system, one accent and a hairline ledger aesthetic, and the navigation was redrawn around what people actually look for.

   Added

   • A dedicated country browser at /data/countries: every country with its region and income group, searchable and filterable.
   • Sortable columns and a free-text filter on every data table — click any header to sort, click again to reverse it.
   • A trend-line view on rankings, a donut option on composition charts, and axis names on every cartesian chart.
   • A source column and a sort direction on the indicator listing.

   Changed

   • Navigation is now Data · Countries · Analysis · Sources. /data is the catalogue itself: the old hub is gone, /data/rankings redirects there with your query intact, and indicator pages keep their /data/rankings/{id} URLs.
   • Compare moved inside Analysis as its seventh tool. /data/compare redirects to it.
   • The homepage was rebuilt around the inventory wall — one tile per country, shaded by how much data we hold for it, every tile a link to that country.

   Fixed

   • Six reported defects: compare deselection, workbench view switching, sortable tables, and the rank visualisation.
   • A mobile sweep — comparison overflow, tables as the default view on phones, and taller controls.
   • Value columns fall back to a plain Value header when an indicator carries no unit.
7. 2026-07-22

   Platform

   v0.4.0

   Analysis: a suite of six tools

   #

   A lite data studio over the whole catalogue: transform, correlate, weight, compute, rank and pivot — all of it shareable as a URL.

   Added

   • ChartKit: seven fixed chart shapes, so no page picks a chart type by hand, plus linked selection — click a table row and the matching chart series and map country light up with it.
   • An Analysis hub with six tools, each on its own URL and its own state: explore, correlate, index-builder, formula, ranks and pivot.
   • The composite index builder: choose up to eight indicators, weight them with sliders, pick a normalisation, and rank every country by the result.
   • The formula lab: computed series such as A / B or (A − B) / C * 100, read by a safe expression parser rather than evaluated.
   • The correlation lab: a pairwise matrix over shared country-years, lagged correlation, and one click from any cell to that pair as a scatter.
   • Transforms that compose: per-capita, index = 100 at a chosen year, year-on-year change, z-score, rank, and log scale.
   • Export the transformed table as CSV or JSON, and share the entire workbench state as a link.
   • A command palette on ⌘K / Ctrl-K over countries, indicators, topics and pages.
   • Light and dark themes, an edge-to-edge frame, and a mobile tab bar.
8. 2026-07-22

   API

   v0.3.1

   Portal-first: the public API withdrawn

   #

   Breaking change

   The public API route was taken off the edge while licence terms were settled. Downloads kept working; the base URL did not.

   Added

   • Per-indicator access tiers — open and display — recorded through the pipeline and carried in every payload, so a client can see what it is allowed to do before it asks.

   Changed

   • The public API base was removed from the edge. The portal reaches the API over a private internal call instead, and every download is proxied through it.
   • Indicators recorded as display-only stopped being exportable: data and CSV requests for them return 403, and a redistribution we do not hold the right to make is never served.
9. 2026-07-22

   Data

   v0.3.0

   The catalogue fills out

   #

   Three ingestion waves, at full depth — every year of history a provider publishes, not just the recent slice.

   Added

   • Wave two: nine providers, 138 indicators, 46 curated indicator links, and a cadence-driven scheduler that imports each source on its own publication rhythm.
   • Wave three took the total to 42 providers — WHO GHO, the UN SDG database, UNESCO UIS, UN Comtrade, DHS, IMF, OECD, BIS, ECB, the World Inequality Database, Maddison, Gapminder, FAOSTAT, EDGAR, the Global Carbon Budget, IRENA, ClimateTRACE, Aqueduct, ND-GAIN, V-Dem, Freedom House, Transparency International, International IDEA, the Fragile States Index and the IPU among them.
   • Wave four added ten more sources, including Women, Business and the Law and the Open Budget Index through Data360, the Climate Change Knowledge Portal, AMECO, UN World Urbanization Prospects, further OECD dataflows, World Bank ASPIRE, Gender and Findex, and OpenAlex.
   • An events layer: GDACS, USGS and EONET hazard feeds, upserted and pruned on a schedule.
   • Sources of sources — every provider now names the upstream origins it is built on, drawn as a lineage board at /data/sources.
   • A 13-topic taxonomy across the whole catalogue, applied by a three-layer classifier: provider defaults, then deterministic code and name rules, then a curated override file. The 1,104 indicators the first two layers left over were classified by hand; exactly one stayed in other.
   • Human-readable names for 659 indicators that arrived as bare provider codes.
   • Tokenised multi-word search, sort options, and provider, region and income filters on the API.
   • Related indicators on every indicator page, merged from four tiers — curated links, same-concept groups, same topic, and computed correlations — each labelled with the tier it came from.

   Changed

   • Correlation candidates are bounded to featured indicators plus the 300 best-covered per topic. That bound is what keeps the correlation mart buildable at this size.

   Fixed

   • NaN and Infinity values are dropped during transformation. JSON cannot represent them, and they were turning into 500s.
   • Rate-limited providers are retried honouring their own Retry-After rather than a guess.
   • Provider metadata is corrected on the way out — INFORM’s Phisical typo, embedded series codes, and stray No breakdown labels.
10. 2026-07-22

    Platform

    v0.2.0

    The portal opens

    #

    The public site: a catalogue, a ranking page for every indicator, a profile for every country, and charts that read correctly in both themes.

    Added

    • The homepage, the indicator catalogue, ranking pages with a choropleth and a league table, country profiles, and an about page.
    • CSV and JSON downloads on every ranking page, carrying the source and its licence alongside the numbers.
    • Open Graph images, a sitemap, and CDN cache headers.
    • Cookieless product analytics: no cookies, no persistent identifier, Do Not Track honoured — and nothing collected at all where it is switched off.
    • Continuous integration on every push: pipeline and API tests, a warehouse parse, and a web build.

    Fixed

    • Choropleth geometry crossing the antimeridian is normalised, and Antarctica is dropped from the map.
    • Every page and fixture was aligned to the live API shapes, and rendering made null-safe throughout.
11. 2026-07-21

    Platform

    v0.1.0

    Foundations

    #

    The first build: an ingestion pipeline, a transformed warehouse, and a read API over it. Nothing public yet.

    Added

    • An ingestion pipeline with resumable loads, a scheduler, pending and ingest-due commands, and nightly backups.
    • A warehouse: staging per provider, one country spine, a single observation fact table, and the ranking marts on top of it.
    • A read API over the marts, and an admin console for running and inspecting imports.

Entries are written from what actually shipped. Version numbers are editorial: the platform ships continuously and carries no release tags, so a release here is a grouping of work, dated by the day it landed. Breaking changes are marked as such. Corrections and questions: get in touch.