[Borisserz]

Summary

Adds an optional Claude subscription leverage row for #200, scoped to Claude only as requested in review. Rebuilt from current main.

When the default Claude login is Max / Pro / Team and the user enters a monthly plan price in Settings, the popover shows one line under this week / this month:

Claude plan $100.00/mo · API-equiv $610.00 · 6.1×

How the numerator is computed (record-level provenance):

• UsageStore.claudeMonthAPIEquivalentCost reads the Claude snapshot's month usageCost. It is looked up by id, the same way the existing claudeActiveBlock lookup works.
• That total is already folded per record by LocalUsageReader.Bucket:
  • Claude's source-reported costUSD (from cost-state) counts as .source.
  • A price-table amount counts as .estimate.
  • An unpriced record counts as .unavailable: it adds nothing to the amount and marks the total partial.
• Both .source and .estimate Claude amounts are API-rate figures, so every known amount counts as API-equivalent.
• An unpriced record never inflates the ratio. The row's tooltip reuses the existing cost explanation, so a partial month says so.
• There is no provider-level costIsEstimate flag any more. Other providers (Gemini/Codex estimates, Grok/OpenCode/Hermes bills) never enter the numerator. The combined month header still sums every provider, unlabeled.

Gate: all of the following must hold, or the row is hidden:

• LimitStatus.isFlatRateSubscription (max / pro / team, any case).
• Plan price > 0.
• Claude month $ > 0.
• Exactly one Claude login. With several logins the month total is machine-wide while the plan belongs to one account, so the ratio would mix plans. Per-account leverage can follow separately.

The menu bar and all cost rows are unchanged: bare $, no "API-equiv" caption anywhere except the leverage row (#224).

Support for other providers is intentionally out of scope and can be added later with their own plan inputs.

Type of change

• Bug fix
• New feature
• Refactor / cleanup
• Documentation
• Other:

UI changes

Before	After
Popover: week / month totals only.	Same totals. For a single Max/Pro/Team Claude login with a plan price set, one extra caption line: Claude plan $X/mo · API-equiv $Y · Z×. Its tooltip explains the source, estimate, or partial coverage.
Settings → General: no plan price.	Optional Claude monthly plan price ($) field (default 0 = off). It drafts locally and commits on submit, blur, or leaving Settings (same pattern as #187), so UserDefaults isn't written on every keystroke. Negative, NaN, or infinite values load and save as 0.
Menu bar cost: $12.3.	Unchanged.

Validation

• ./scripts/test-gate.sh: 1395 tests, 0 failures (17 skipped); logic-core line coverage 93.39% (≥ 75%).
• New tests:
  • UsageStoreTests:
    • testLeverageUsesOnlyClaudeMonthCost: Claude $610 + Gemini estimate $200 + Grok bill $50, plan $100. Header = $860, leverage = 6.1×.
    • testOtherProvidersNeverFeedLeverage.
    • testLeverageKeepsRecordLevelCoverage: mixed source/estimate/unavailable gives the partial hint.
    • testLeverageGates: price unset or 0, Free, no plan, no credential, unpriced-only month, non-costing snapshot.
    • testLeverageOpensForProAndTeam.
    • testLeverageHiddenWithSeveralClaudeAccounts.
    • testMonthlyPlanPricePersistsAndSanitizes.
    • testLeverageStaysOutOfTheMenuBar.
  • ModelDecodingTests: testIsFlatRateSubscription, testMultiplierFormat, testLeverageUiShape (no store binding; draft and onDisappear commit).
  • LocalizationInterpolationTests: leverage placeholders survive in all 7 languages.
• Fault injection: I removed each guard in turn and the targeted tests failed.
  • Dropping the Claude-id filter: other providers fed a 4× ratio.
  • Dropping the multi-account gate.
  • Dropping the finite-price check.
• llvm-cov --show-regions shows no ^0 regions in the new UsageStore code.
• Docs: the defect-log.md UI entry covers the record-level vs. provider-level provenance and the multi-account gate. provider-extension.md notes that the leverage lookup is an explicit provider-specific branch, like claudeActiveBlock.

Checklist

• swift build and swift test pass locally
• PR title and description are written in English
• UI changes are described above (before/after — images optional)
• No copyrighted assets, secrets, or private tooling references are committed (see CONTRIBUTING)
• Tests were added or updated for this change

Related

• Issue: Plan-aware cost: label API-equivalent cost and show subscription leverage #200
• Closed attempt: feat: label popover cost as API-equivalent on Max/Pro/Team (#200) #224

[shawnla90]

Built and ran this same feature on my own machine last week (branch linked in #200), so two notes from that experience:

1. The multiple mixes subscriptions. monthAPIEquivalentCost sums every costIsEstimate provider (Claude, Codex, Gemini, ...) but subscriptionLeverage gates on the Claude plan and divides by one plan price. A Max user who also runs Codex CLI (covered by a ChatGPT subscription, not the Claude plan) gets an inflated 'leverage' number: on my machine, August is ~$2,783 of Claude API-equivalent, and Codex estimates on top would push the displayed multiple past what the Max plan actually delivered. Two clean fixes: scope the numerator to claude_code (the plan the gate reads), or keep the mixed sum and make the row's copy say what it covers. I'd do the former — the hint text says the price is for the Max/Pro/Team row, so users will type their Claude plan price.

2. The price field writes defaults per keystroke. TextField("0", value: $store.monthlyPlanPrice, ...) binds straight to the store, and the didSet writes to UserDefaults on every edit. The scan-roots field from #187 established the draft + commit-on-submit/blur pattern for exactly this reason, and CustomScanRootsTests.testSettingsUsesProviderPickerAndCommitsOnSubmit locks it there ('direct TextField binding to the store writes on every keystroke'). Worth matching that pattern here so the convention holds.

Neither blocks the shape — per-provider costIsEstimate is the right structure and matches the #224 close. Happy to test the branch on my dual-login setup once these land.

[Borisserz]

Landed in c8d7c71.

1. Mixed subscriptions. $Y is now Claude's month estimate only (costIsEstimate && providerID == claude_code). Combined month header still sums everyone unlabeled. Fixture: Claude $610 + Gemini $200 estimate + Grok $50 bill → header $900, leverage 6.1× (not 8.1×). Settings hint now says the field is the Claude Max/Pro/Team price.

2. Per-keystroke defaults. Plan price uses a local draft and commits on submit, blur, and leaving Settings — same pattern as custom scan roots (#187). $store.monthlyPlanPrice is no longer bound to the TextField.

Happy to have you run the dual-login setup against this commit.

[chattymin]

Please keep this PR scoped to Claude subscription leverage for now and align the PR description, tests, and UI copy with that behavior. The calculation should use record-level cost provenance and include only the Claude API-equivalent cost, rather than classifying an entire provider with one estimate flag. Please rebase onto current main. Support for additional providers can be added separately later.

[Borisserz]

Reworked per your review and rebased onto current main (f0420a3):

• Claude-only scope. The numerator is the Claude snapshot's month usageCost, looked up by id like claudeActiveBlock. No other provider can enter it; the combined header still sums everyone.
• Record-level provenance. The provider-level costIsEstimate flag is gone. The amount comes from the per-record CostCoverage that Bucket already merges (source-reported costUSD / price-table estimate / unavailable). Unpriced records add nothing and mark the month partial, which the row's tooltip explains.
• Multiple Claude logins hide the row, since the month is machine-wide but the plan is one account's.
• Description, tests and copy ("Claude plan … · API-equiv … · Z×", "Claude monthly plan price") now match this behavior. Gate: 1395 tests green, 93.39% coverage.