Skip to content

CHANGELOG.md

Latest commit

 

History

History
31 lines (24 loc) · 2.96 KB

CHANGELOG.md

File metadata and controls

31 lines (24 loc) · 2.96 KB

Changelog

v2.0 — Multi-tool leaderboard (June 2026)

viberank evolved from a Claude Code leaderboard into the leaderboard for all AI coding usage — Claude Code, Codex, Gemini CLI, Copilot, OpenCode and every other tool ccusage tracks.

Fixed

  • Submissions were failing with Invalid date format: undefined — ccusage v20's default report keys daily rows by period (not date). All ccusage report shapes are now normalized server-side at a single chokepoint (src/lib/ccusage.ts), so old and new output both work. (#49)
  • "Token totals don't match" rejections for Gemini/Codex users — reasoning/thinking tokens are counted in totalTokens but not broken out by ccusage. The token check is now one-sided (total >= components) with a cost/token ratio guard as the anti-inflation defense. (#48)
  • Merge button / web upload RLS failures: all writes (claim/merge, upload, admin flag) now run through authenticated server routes with the service-role client. (#42, #47)
  • Token stats now roll over to trillions (2.3T, previously 2305.1B).

Added

  • Multi-tool support (#45): submissions record which tools contributed (submissions.tools[], daily_breakdowns.agents[], migration 002); tool chips on every row; an "All tools / Claude / Codex / …" filter.
  • Per-tool leaderboards at /tool/claude, /tool/codex, /tool/gemini, /tool/copilot, /tool/opencode — server-rendered with FAQ + structured data.
  • Server rendering everywhere: homepage (first page + stats, ISR 5 min), profiles (single cached DB read shared with metadata), tool boards (ISR hourly). Structured data: FAQPage, ProfilePage, BreadcrumbList, BlogPosting.
  • Global rank on profile pages.
  • Full redesign: editorial scrolling layout with hero, top-3 podium, sticky filter bar, homepage FAQ, site footer; flat dark theme (no gradients).
  • Blog: three data-backed posts (tool cost comparison, what Claude Code costs, cutting your AI coding bill) + Tailwind Typography (post formatting was previously broken).
  • pnpm test: zero-dependency test harness for ccusage normalization/validation (test/ccusage.test.mts).

Changed

  • npm packages renamed (the old names belong to the original author): CLI is now viberank-cli (v1.1.0, pinned to ccusage daily --json), MCP server is viberank-mcp.
  • Branding broadened additively: "Claude Code, Codex & AI Coding Leaderboard" (the Claude Code keyword is preserved everywhere for SEO; no URL changes).
  • Historical submissions backfilled with tools derived from their models_used.

Removed

  • The dormant Convex backend — viberank is Supabase-only. The Vercel build no longer wraps convex deploy.

Known limitations

  • Multi-machine submissions with overlapping dates overwrite rather than sum daily data (#43) — ccusage exposes no machine identifier; a CLI-supplied machine ID is the planned fix.