Port upstream 0.67.0: count Mistral plan-covered tokens

[Finesssee]

Summary

Mistral token totals now count consumed units, while spend keeps counting billed units, so plan-covered usage (Le Chat, Vibe Code, included API allowance) is no longer reported as zero tokens.

• BillingResponse gains chat and vibe_code.completion (JSON key vibe_code).
• Tokens and cost aggregate across completion, chat, and vibe_code.completion; each model entry counts once toward model_count.
• Per entry, tokens use value.or(value_paid); cost uses value_paid.or(value) times the price index.
• Checked overflow still applies: TokenCounts::add_lane / add errors surface across all three categories.
• ocr, connectors, audio, libraries, and fine-tuning stay cost-only.

Upstream reference

• Release: CodexBar v0.67.0, Mistral plan-covered token counting (item 24 in the port audit).
• Tag-pinned (v0.67.0) sources: Sources/CodexBarCore/Providers/Mistral/MistralModels.swift (chat, vibeCode, MistralVibeCodeUsageCategory), MistralUsageFetcher.swift (parseResponse, aggregateModel), docs/mistral.md; tests MistralTokenArithmeticTests.swift.

Ported / Deferred

• Ported: the summary-level assertions of the upstream tests (plan-covered consumption across completion/chat/vibe_code with the fixture model at price 0.25, value/value_paid fallback, paid-zero not hiding overflow, checked overflow across API/chat/Vibe categories, cost-only categories not summing token lanes).
• Deferred: the daily-bucket assertions (daily, last30DaysTokens); Win-CodexBar's Mistral provider has no daily buckets.
• model_count follows the upstream loop (one increment per model entry per category), so a model name that appears in two categories is counted twice.

Validation

• cargo +1.98.0 fmt --all: clean
• cargo +1.98.0 test -p codexbar mistral -- --test-threads=4: 17 passed, 0 failed (6 new)
• cargo +1.98.0 clippy -p codexbar --all-targets -- -D warnings: clean
• cargo +1.98.0 clippy --workspace --all-targets -- -D warnings: not completed. The Tauri crate build script failed on a stale plugin-permissions path outside this worktree (...\cargo-target\cursor-cost-403\...\app_hide.toml); the Tauri crate is not touched by this PR.

Affected areas

• Rust backend / provider (rust/src/providers/mistral/mod.rs)
• Tauri shell
• Frontend
• Settings / bridge types

UI proof

Not applicable

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Advanced

Run ID: 3fed7ccb-2152-4e90-b704-b60ea9fda3b5

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[Finesssee]

Thermo-nuclear review

Reviewed by Codex gpt-6-luna (xhigh); verified and validated by Claude

Head reviewed: cc07049

Findings: none. The change matches the upstream 0.67.0 "mistral-plan-tokens" item: Le Chat (chat) and Vibe Code (vibe_code.completion) are aggregated with API completions; tokens count consumed units (value, falling back to value_paid) while spend counts billed units (value_paid, falling back to value); token sums stay checked for overflow; cost-only categories (ocr, connectors, audio, libraries) do not add to token lanes. The added tests cover each category, the fallbacks, cross-category aggregation, and the overflow path.

Nothing needed fixing, so no follow-up commit was pushed. Validation commands were not re-run because no files changed.

[Finesssee]

Validated 2026-10-02 at cc07049 (PR head):

• Reviewed against the 0.67.0.md PR 17 spec (Mistral plan-covered tokens): the diff matches — chat + vibe_code.completion categories added to BillingResponse; token totals count consumed units (value falling back to value_paid) across completion/chat/vibe_code.completion; spend keeps counting billed units (value_paid falling back to value); model_count counts each model entry once; other categories (ocr, connectors, audio, libraries, fine-tuning) stay cost-only; checked-overflow (TokenCounts::add_lane) applies across all three token categories. Upstream MistralTokenArithmeticTests fixtures ported (input 1000 / output 500 / cached 300 / model 1 / cost 50; value-missing fallback 20 tokens cost 5; overflow across categories). No defects found; no changes needed.
• cargo fmt --all --check — clean.
• cargo test -p codexbar — 2166 passed / 0 failed / 1 ignored (mistral focused 17/17).
• cargo test -p codexbar-desktop-tauri — 459 passed / 1 failed = the documented Isolate bootstrap payload test from real settings #684 bootstrap env baseline (79 vs 78, real-settings dependent; hermetic fix Make the bootstrap catalog test hermetic (#684) #711 is not on this base; also fails on unmodified base).
• cargo clippy --all-targets -- -D warnings — the only three errors are the pre-existing main-history lint sites (alibabatokenplan/cli.rs:163 manual_range_contains, kiro/usage_limits.rs:315 + openai/subscription.rs:206 nonminimal_bool), all outside this diff (documented in QUEUE INTEGRATION FINDINGS; rustc 1.96 lint drift). The diff itself is clippy-clean.
• No push needed (head cc07049 already on origin, verified via ls-remote); no fixes required.

Chain note: this PR gates #657 (Mistral plan detail) and through it #702/#717/#718; after #657 merges this head the chain is unblocked.