# Pricing fix evidence

Issue [#25414](https://git.cafe/pingdotgg/t3code/issues/25414); base [184d915d87](https://git.cafe/pingdotgg/t3code/commit/184d915d87), head [f5a43a82bb](https://git.cafe/pingdotgg/t3code/commit/f5a43a82bb). Local HEAD independently confirmed as f5a43a82bb3c5ab4e4343bd66357b71638a8cf5b.

Capture matrix: `[]`. The base-to-head diff contains only server usage pricing and tests; no changed render paths. No new runtime or browser was launched for this evidence pass. Existing real runtime captures were supplied by the lead from explicit `--home-dir /tmp/issue-25414/t3-home`.

## Runtime and regression proof

- [Before runtime JSON](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/a01bca4c3abe2745/before-runtime.json)
- [After runtime JSON](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/0d117b26e1fc1e4a/after-runtime.json)
- [Reproduction and red-test transcript](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/b7dd10b7e71e4bb4/reproduction.txt)
- [Isolated fixture seed](https://gh-file-drop-api-prod-mi5fy3sowv63ufte.pinglabs.workers.dev/f/838426408dd15817/seed-usage.py)

Independent Python assertions passed against the local JSON artifacts: both responses Success; identical day, provider, model, source path, token totals, record/session counts, cost source and pricing metadata. Model gpt-6-sol, provider codex, one record, one session, zero unpriced records. Session identity itself is not exposed in these summary responses; the seed identifies session issue-25414 and turn issue-25414-turn.

Matching totals: 20,000 uncached input + 280,000 cached input, zero cache creation, 1,000 output, zero reasoning. The 300,000-token prompt exceeds 272K. Cost changes from $0.106 to $0.207 (JSON floating-point representation 0.20700000000000002). After categories: input $0.080, cache read $0.112, output $0.015, cache write $0; sum $0.207. Cache savings $1.008. Assertions used absolute tolerance 1e-12.

The supplied red transcript records the Codex test receiving $0.106 instead of $0.207, plus Claude and priority-tier pricing failures. It is a summarized transcript, not a complete raw red-test log; tests were not rerun in this capture-only pass.

## Validation and gaps

Local check-changed.log confirms 3 test files / 68 tests passed; format, lint, GraphQL and one knip phase passed. Overall check:changed was NOT green: typecheck failed resolving TypeScript 7.0.2 bin/tsc; the second knip phase failed loading apps/mobile/metro.config.js and reported PASTED_TEXT_ATTACHMENT_THRESHOLD_BYTES as unused. The implementer reports 121 full usage-suite tests passed; no raw log for that run was supplied or independently verified here. Runtime evidence uses synthetic usage/rate fixtures through the real server, not a paid provider request.

## FRICTION

- Supplied description called the check log green; inspection found the typecheck/knip failures above.
- Read-only shell tool responses were delayed briefly; no retry runtimes, repo edits, pushes, merges, watches, delegates or checkouts were created.