Fix Claude Fable 5 cost totals

[steipete]

Summary

• add bundled pricing for Anthropic's claude-fable-5
• preserve request timestamps and price historical Claude usage with the tariff active at request time
• apply long-context thresholds per request instead of cumulative session totals
• account for native Claude 1-hour cache-write tokens
• invalidate stale Claude and Pi cost caches after pricing changes

Pricing

Anthropic pricing reference:
https://platform.claude.com/docs/en/about-claude/pricing

Pi session JSONL does not record Anthropic cache retention, so Pi usage keeps its persisted default cache-write tariff rather than guessing historical retention.

Proof

• make check
• focused pricing/scanner/cache suite: 53 tests passed
• full suite: 3,417 tests in 392 suites passed
• autoreview, whole branch against origin/main: clean, no accepted/actionable findings

Fixes #1368

[clawsweeper]

Codex review: needs changes before merge. Reviewed June 10, 2026, 2:22 AM ET / 06:22 UTC.

Summary
Adds Fable 5 Claude pricing, historical request-level Claude repricing, 1-hour cache-write accounting, cost-cache invalidation, parser hash updates, and pricing/scanner tests.

Reproducibility: yes. from source inspection: Vertex-style claude-fable-5@... names are recognized by the scanner, but the fallback pricing normalizer does not strip the @... suffix before the new dictionary lookup.

Review metrics: 2 noteworthy metrics.

• Changed surface: 13 files; 901 additions, 117 deletions. The PR spans pricing logic, scanners, cache schema versions, generated parser hash, changelog, and tests.
• Cache invalidations: 2 artifact-version bumps. Claude/Vertex cost cache and Pi cost cache versions change, so maintainers should notice the intentional derived-cache refresh.

Merge readiness
Overall: 🦐 gold shrimp
Proof: 🌊 off-meta tidepool
Patch quality: 🦐 gold shrimp
Result: ready for maintainer review.

Overall follows the weaker of proof and patch quality, so missing proof can cap an otherwise strong patch.

Rank-up moves:

• [P2] Normalize @default and @YYYYMMDD Claude aliases for the bundled Fable fallback and add an empty-cache regression test.

Risk before merge

• [P1] Vertex-formatted Fable model IDs such as claude-fable-5@default or claude-fable-5@20260609 can still be unpriced when the models.dev cache is stale or missing.

Maintainer options:

1. Decide the mitigation before merge
   Land the pricing/cache fix after adding bundled fallback support and regression coverage for Claude @... model aliases.
2. Pause or close
   Do not merge this PR until maintainers decide whether the risk is worth taking.

Next step before merge

• [P2] A narrow automated repair can add Claude @... alias normalization for bundled fallback pricing plus regression coverage.

Security
Cleared: The diff changes local pricing, scanner, cache, and test code; I found no new dependency, secret, network, CI, or code-execution surface.

Review findings

• [P2] Price Vertex-formatted Fable aliases in the fallback — Sources/CodexBarCore/Vendored/CostUsage/CostUsagePricing.swift:216

Review details

Best possible solution:

Land the pricing/cache fix after adding bundled fallback support and regression coverage for Claude @... model aliases.

Do we have a high-confidence way to reproduce the issue?

Yes, from source inspection: Vertex-style claude-fable-5@... names are recognized by the scanner, but the fallback pricing normalizer does not strip the @... suffix before the new dictionary lookup.

Is this the best way to solve the issue?

No. The overall pricing/cache direction is maintainable, but the patch should cover the existing Vertex alias surface before merge.

Full review comments:

• [P2] Price Vertex-formatted Fable aliases in the fallback — Sources/CodexBarCore/Vendored/CostUsage/CostUsagePricing.swift:216
  The new built-in claude-fable-5 entry only matches the bare normalized ID. For Vertex-style Claude logs such as claude-fable-5@default or claude-fable-5@20260609, normalizeClaudeModel leaves the @... suffix intact after a stale/missing models.dev lookup, so the scanner can still record an unpriced or zero Fable cost. Please normalize those aliases or add explicit fallback aliases with a regression test.
  Confidence: 0.9

Overall correctness: patch is incorrect
Overall confidence: 0.88

AGENTS.md: found and applied where relevant.

Codex review notes: model gpt-5.5, reasoning high; reviewed against 6f6cb097dc58.

Label changes

Label changes:

• add P2: This is a normal-priority provider cost-calculation fix with limited blast radius, though one alias pricing path remains incomplete.
• add rating: 🦐 gold shrimp: Overall readiness is 🦐 gold shrimp; proof is 🌊 off-meta tidepool and patch quality is 🦐 gold shrimp.
• add status: ⏳ waiting on author: ClawSweeper has contributor-facing work open and is waiting for author action. Not applicable: The PR is owner-authored, so the external-contributor real behavior proof gate does not apply; the body lists make check, focused pricing/scanner/cache tests, and the full suite.
• remove status: 👀 ready for maintainer look: Current PR status label is status: ⏳ waiting on author.
• remove rating: 🐚 platinum hermit: Current PR rating is rating: 🦐 gold shrimp, so this older rating label is no longer current.

Label justifications:

• P2: This is a normal-priority provider cost-calculation fix with limited blast radius, though one alias pricing path remains incomplete.
• rating: 🦐 gold shrimp: Overall readiness is 🦐 gold shrimp; proof is 🌊 off-meta tidepool and patch quality is 🦐 gold shrimp.
• status: ⏳ waiting on author: ClawSweeper has contributor-facing work open and is waiting for author action. Not applicable: The PR is owner-authored, so the external-contributor real behavior proof gate does not apply; the body lists make check, focused pricing/scanner/cache tests, and the full suite.

Evidence reviewed

Acceptance criteria:

• [P1] swift test --filter CostUsagePricingTests.
• [P1] swift test --filter CostUsageScannerClaudeFableTests.
• [P1] make check.

What I checked:

• Current main still lacks Fable pricing: A current-main search finds no Fable pricing or cache-retention support, so this PR is not obsolete on main. (6f6cb097dc58)
• PR adds the bundled Fable fallback: The PR head adds a built-in claude-fable-5 entry to the Claude pricing table. (Sources/CodexBarCore/Vendored/CostUsage/CostUsagePricing.swift:216, 91364f2009ac)
• Bundled fallback still misses @ aliases: claudeCostUSD falls back to the dictionary key returned by normalizeClaudeModel, but that normalizer strips dotted, -vN:N, and -YYYYMMDD suffixes only; it does not strip @default or @YYYYMMDD. (Sources/CodexBarCore/Vendored/CostUsage/CostUsagePricing.swift:419, 91364f2009ac)
• Vertex route can supply @ model names: The Claude scanner explicitly treats claude-* model names containing @ as Vertex AI format, so those model IDs can reach the same cost-pricing path. (Sources/CodexBarCore/Vendored/CostUsage/CostUsageScanner+Claude.swift:405, 91364f2009ac)
• Prior review context: The existing review discussion already flagged the Vertex-formatted Fable alias gap at Fix Claude Fable 5 cost totals #1372 (comment), and the latest head still leaves that path uncovered. (Sources/CodexBarCore/Vendored/CostUsage/CostUsagePricing.swift:216, 91364f2009ac)
• Repository policy read: AGENTS.md was read fully; its provider/parser guidance pushed this review toward focused source/test inspection without live provider probes. (AGENTS.md:19, 6f6cb097dc58)

Likely related people:

• steipete: Current-main blame points the Claude pricing table and Vertex-detection paths to Peter Steinberger, and the PR head modifies the same pricing/scanner surface. (role: recent area contributor; confidence: high; commits: 920997c6a365, 91364f2009ac; files: Sources/CodexBarCore/Vendored/CostUsage/CostUsagePricing.swift, Sources/CodexBarCore/Vendored/CostUsage/CostUsageScanner+Claude.swift)

What the crustacean ranks mean

• 🦀 challenger crab: rare, exceptional readiness with strong proof, clean implementation, and convincing validation.
• 🦞 diamond lobster: very strong readiness with only minor maintainer review expected.
• 🐚 platinum hermit: good normal PR, likely mergeable with ordinary maintainer review.
• 🦐 gold shrimp: useful signal, but proof or patch confidence is still limited.
• 🦪 silver shellfish: thin signal; proof, validation, or implementation needs work.
• 🧂 unranked krab: not merge-ready because proof is missing/unusable or there are serious correctness or safety concerns.
• 🌊 off-meta tidepool: rating does not apply to this item.

Shiny media proof means a screenshot, video, or linked artifact directly shows the changed behavior. Runtime, network, CSP, and security claims still need visible diagnostics.

How this review workflow works

• ClawSweeper keeps one durable marker-backed review comment per issue or PR.
• Re-runs edit this comment so the latest verdict, findings, and automation markers stay together instead of adding duplicate bot comments.
• A fresh review can be triggered by eligible @clawsweeper re-review comments, exact-item GitHub events, scheduled/background review runs, or manual workflow dispatch.
• PR/issue authors and users with repository write access can comment @clawsweeper re-review or @clawsweeper re-run on an open PR or issue to request a fresh review only.
• Maintainers can also comment @clawsweeper review to request a fresh review only.
• Fresh-review commands do not start repair, autofix, rebase, CI repair, or automerge.
• Maintainer-only repair and merge flows require explicit commands such as @clawsweeper autofix, @clawsweeper automerge, @clawsweeper fix ci, or @clawsweeper address review.
• Maintainers can comment @clawsweeper explain to ask for more context, or @clawsweeper stop to stop active automation.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 4413eb4133

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Price Vertex-formatted Fable aliases in the fallback

When Fable is logged in Vertex AI format (for example claude-fable-5@20260609 or @default), this new bundled fallback still won't be used. The models.dev lookup can normalize @ aliases, but with a stale/missing models.dev cache the code falls back to normalizeClaudeModel, which only strips -YYYYMMDD dates and leaves the @... suffix, so the dictionary lookup misses and the scanner records a zero/unpriced cost. The repo already treats @ as Claude Vertex format in CostUsageScanner+Claude, so this fallback should cover those aliases too.

Useful? React with 👍 / 👎.