One language for cost and usage.

Unreleased

Added

• Crawlers, search engines and AI agents can discover every public page: the site serves robots.txt and a sitemap, each page names its own path as canonical (shared query-string views canonicalize to the base page), every page has a description, and the About and spec column pages publish JSON-LD; a local smoke checks all sitemap pages over HTTP.

Fixed

• The About page changelog shows plain version labels instead of links to private repository compare pages; bundling fails if a private-repository link would reach public content.
• Links in prose and notices now stay underlined in light and dark themes; a local accessibility smoke checks key pages for serious WCAG violations and horizontal overflow.
• Snapshot generation now uses consistent read-only transactions, shared source fingerprints and atomic replacement; incomplete or mismatched snapshot pairs and stale catalog defaults stop bundling. Static results have deterministic ordering and serialization, with an explicit read-only drift check.
• Forecast artifacts now validate dates, uncertainty bands, anomalies, projections, model metrics and recorded snapshot ranges before bundling; malformed or stale data stops development and builds with a specific diagnostic.
• Failed explorer and catalog queries now keep their inputs and recovery controls, show safe classified messages, and expose query-state markers for semantic monitoring.
• Explorer and catalog inputs now validate identifiers, calendar dates, UTC timestamps, integer pages and value lengths before SQL; normalized default parameters keep precomputed results.
• Explorer untagged drills now select missing tag values, keep literal (untagged) values separate, and preserve the missing filter in shared URLs.
• Live queries now stop waiting when database connections or responses stall, and recover safely from dropped idle connections.

Security

• Database connections from the app, snapshot tools, forecast and psql scripts now verify the Supabase certificate chain and hostname against a pinned root CA; URL sslmode parameters can no longer disable verification.
• Responses now carry explicit browser hardening headers (no-sniff, referrer, permissions, framing denial) and an enforced Content Security Policy that limits scripts, styles, images, fonts and connections to the site itself and frames to focus.finops.org; preview deployments also allow the Vercel Toolbar. npm run test:headers checks real responses, zero violations in Chrome and refused framing.

1.3.0 - 2026-10-10

Added

• A 14-day cost forecast, built offline with Prophet from daily Usage cost:
  • A Forecast panel on the dashboard, featuring the series with the most recent data.
  • A new Forecast page under Data, with Total, AWS, Microsoft and Oracle series, 95% and 80% expected ranges, a projected 14-day total, full-screen charts and data tables.
  • A model card for each series, with the training window, settings and a final-week backtest comparing daily and hourly models.
• Anomaly flags: any day outside the 95% expected range is marked as a spike or a drop, with its deviation and severity, plus a guide to how the forecast flags anomalies.
• Forecast tooling (tools/forecast/) and npm run precompute:all, which regenerates the dashboard snapshot and the forecast in one step.

Changed

• The sample data now reads as recent: charge timestamps move forward 78 weeks (keeping weekdays) and billing periods 18 months (keeping calendar months), so most charges fall in March 2026. A repeatable migration applies the shift, and the sample loader re-applies it.
• Query defaults, explorer ranges and documentation use the shifted dates.

Fixed

• Use case capabilities and reference text show "&" and other characters instead of HTML codes such as &.
• The column profile precompute runs in smaller batches so each query stays within the 5-second read-only limit.

1.2.0 - 2026-10-10

Changed

• The menu is now grouped into collapsible sections:
  • Data: Dashboard, Explore, Column profile.
  • Use case library: Use cases, Queries.
  • Specification: versions 1.4 and 1.0.
  • About: on its own.
• The section containing the current page opens automatically, and each section shows how many pages it holds.

1.1.0 - 2026-10-10

Added

• A full-screen control on every chart:
  • Uses true full screen on desktop, Android and iPad, and a full-viewport view on iPhone.
  • Redraws for portrait or landscape. On Android, it can lock the screen to landscape.
  • In full screen, you can zoom into part of the axis, show or hide series from the legend, and switch to the data table.

Changed

• New look based on the AI Agent Skills resource library design system:
  • Navy, paper, mint and coral palette, with Georgia headings and square corners.
  • Each page opens with a "data receipt" showing source, data, status, mode and snapshot date.
  • Headline numbers are a ruled ledger.
  • Query results are labelled "Precomputed" or "Live".
• Accessibility:
  • A contrasting inner focus ring.
  • Tap targets of at least 44px.
  • Scrolling SQL blocks can be reached with the keyboard.
  • Long legend labels wrap.
• Full-screen charts fit short landscape screens: the legend sits beside the plot and the controls fold away.

1.0.0 - 2026-10-09

The first public release of FOCUS Explorer, live at https://finops-focus-jwilkes.vercel.app.

Added

• Cost dashboard over the FOCUS 1.0 sample data (100,000 rows from AWS, Microsoft, Oracle and Google Cloud), with:
  • KPIs for effective, billed and list cost, savings versus list, and tag coverage.
  • Charts for daily usage cost by provider, service categories, top services, regions, charge categories and tag keys.
• Data explorer:
  • Filter by provider, service category, service, region, charge category, sub account, date range, tag and text.
  • Group by any dimension or tag key and pick a measure.
  • Drill into any group.
  • See daily trends and paged charge rows.
  • Share any view by its URL.
• Column profile showing the fill rate, distinct values and most common values of every column.
• All 118 FOCUS use case queries, translated to Postgres for the v1.0 sample: 33 run directly, 49 are adapted with documented column substitutions, and 36 are marked unsupported with the reason. Each runnable query has editable parameters, a chart and a result table.
• Use case library (115 use cases grouped by FinOps capability) and FOCUS spec column pages for v1.4 and v1.0, from a snapshot of the FOCUS MCP Server.
• About page with:
  • A link to and embedded preview of focus.finops.org.
  • The FinOps Certified FOCUS Analyst credential.
  • A QR code for the public site.
  • This changelog.
• Menu drawer for navigation on every page and screen size.
• Precomputed results for the dashboard, column profile, filter lists, explorer landing view and every query's default parameters, so those load instantly. Only custom filters and parameters query the database.
• Tooling:
  • FOCUS MCP Server crawler.
  • Catalog validator that runs every query.
  • Sample data loader.
  • Precompute script.
  • One-time tailnet form for saving the database connection string.
  • Least-privilege database role setup for Vercel.

Security

• The app queries Supabase with a dedicated focus_app login. That login can only read, through the select-only focus_reader role, in read-only transactions with a 5 second statement timeout.
• The sample table lives in a schema that isn't exposed through the Supabase Data API, with row-level security enabled.
• Explorer column names come only from fixed allowlists, and every user-supplied value is a bound parameter.
• Deployments upload only the app and its data (a .vercelignore allowlist), never .env files.