verification date: 2026-10-28
main: 8e76f945d2 (github/main)
pr: 15842d9d5f5ee63ff474a65b3e3eaf7f5cc79478

setup:
  git worktree add ../main github/main
  timeout 600 vp i (main and pr separately, both passed)
  main dev ports: web 8693 / server 16733, isolated .t3
  main stopped by captured process-group PID before starting pr
  pr dev ports: web 5733 / server 13773, isolated .t3
  browser: repository Playwright 1.60.0 + /usr/bin/google-chrome
  provider: real Codex gpt-6.1-sol through local CLIProxyAPI
  provider home: /home/user/verify-codex; config copied, no history seeded

main UI:
  added /home/user/t3code as a project through Add project > Local folder
  selected GPT-6.1-Sol and submitted a real provider turn
  provider response: usage verification beta
  native token_count: input=19008, cached_input=9216, output=7, total=19015
  usage > tokens > gpt-6.1-sol: 19K tokens, $0.02, $1.08 per 1M, 48.5% cache hit
  usage > all environments > model prices:
    mapped gpt-6.1-sol to verification-unpriced (no known rate)
  model dialog: Unpriced, 19K tokens have no known price

pr UI + actual WebSocket response:
  same real provider history + same saved model mapping:
  bucket model=verification-unpriced, provider=codex
  totals={uncachedInputTokens:9792,cachedInputTokens:9216,cacheCreationTokens:0,outputTokens:7,reasoningTokens:0}
  records=1, unpricedRecords=1, unpricedTokens=19015, costSource=unpriced
  dialog matches main: 19K tokens have no known price, no About prefix
  created project/thread through UI and submitted another real GPT-6.1-Sol turn
  provider response: usage verification gamma.
  second native token_count: input=18918, cached_input=9216, output=8, total=18926
  final actual WebSocket response:
    totals={uncachedInputTokens:19494,cachedInputTokens:18432,cacheCreationTokens:0,outputTokens:15,reasoningTokens:0}
    records=2, unpricedRecords=2, unpricedTokens=37941, sessions=2
  native total: 19015 + 18926 = 37941, exact RPC match

neighboring controls through UI:
  Set price opened custom model prices for verification-unpriced
  entered USD/million: input=1, output=2, cache read=0.5, cache write=1
  saved: cost=$0.03, per 1M=$0.76, tokens=37.9K, cache hit=48.6%, no unpriced footer
  calculated cost from native usage: $0.02874, effective rate=$0.75749/million
  reset price to automatic and saved: Unpriced, 37.9K tokens have no known price
  reload: RPC still unpricedTokens=37941, matching both native sessions

checks via blacksmith-check, exact PR head:
  vp i: passed
  vp test run apps/server/src/usage/usageAggregation.test.ts packages/shared/src/usageMerge.test.ts apps/web/src/components/usage/usageBreakdown.test.ts
  3 test files passed; 30 tests passed
  existing PR CI: 27 successful, 0 failing, 0 pending

limitation:
  could not reproduce the original mixed-priced/unpriced bucket on main with real Codex.
  codexUsage.ts always emits reportedCostUsd=null and the same model uses the same rate.
  mapping or changing a custom price reprices all records, so neither creates a mixed bucket.
  no synthetic history, intercepted responses, mocks, fixtures or demo data were used.
  mixed-record corrected effective rate and legacy mixed-record estimate labels remain unverified in the real app.
  Codex limits also reports JSON-RPC -32600 through this proxy on both main and PR; token history scanning works.
  native mobile and packaged desktop were not run on this Linux VM.