fix(usage): Claude usage no longer lists a <synthetic> model

[tris203]

Claude Code writes local <synthetic> placeholders into transcripts. Counting them as usage adds a bogus model row, including after a restart when the row is already cached.

The shared line/projected parser now skips that exact marker. Cache loading applies the same filter to Claude regular and tail records, preserving real models, real zero-token records, other providers, parse positions, and saved usage for deleted transcripts. The cache version stays unchanged; no cold rescan is required.

Rebuilt from @tris203's #14672, preserving the contributor commits and merging main. Fixes #14670.

Verification

• Two focused suites: baseline 19 passed; revised regressions against main exposed 2 expected failures (18 passed); after implementation 20 passed. The final review also ran the existing UsageService, transcript-reader, streaming, and pricing suites: 98 passed across six suites overall. The cache round-trip test was extended instead of adding another case/file.
• Scoped server typecheck and changed-file lint/format checks passed.
• Two independent adversarial rounds with Claude Opus 5.5 and Codex gpt-6-sol. Round 1 found no correctness issues; its test-simplification suggestion was applied without adding cases. Round 2 found no remaining findings across the complete diff against main.
• Live headless Linux web app with isolated transcripts and an existing version 5 cache: main displays the bogus model; head removes it while retaining both real models, 3.75K tokens, and 2 sessions, including real cached history whose transcript no longer exists. Appending another placeholder and refreshing retains the same rows and totals. The persisted cache offset advanced to the appended file size.
• Both captures use the same 1280×800 viewport and Past 24h token view. Chromium system fonts measured zero width, so the preview tab loaded a local DejaVu font for both captures; no product font changes.
• Not verified: a live Claude account, macOS, Windows, iOS, Android, or Electron desktop. The server aggregation path is shared by clients; no client layout or contract changes.

Before — main parser/cache:

After — head parser/cache, same real usage:

Implementation: Codex gpt-6.1-sol; tests: Codex gpt-6-astra; reviews: Claude claude-opus-5-5 and Codex gpt-6-sol, through T3 Code delegated tasks. Maintainer integration: Codex gpt-6.1-sol in T3 Code.

[macroscopeapp]

Approvability

Verdict: Approved at 3d7b402

Macroscope's review found this PR approvable — This is a small, self-contained usage bug fix that filters only Claude’s known local <synthetic> placeholder in both fresh parsing and persisted cache loading. Tests cover parser paths, cache cleanup, real zero-token records, and preservation of other providers.

^{You can add or adjust custom eligibility rules. Learn more.}

[coderabbitai]

Important

Review skipped

We couldn't safely recover the incremental review. No full review was started, and the last reviewed checkpoint was preserved. Retry later, or explicitly request a full review by commenting @coderabbitai full review.

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

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

• Configuration used: Path: .coderabbit.config.ts
• Review profile: CHILL
• Plan: Advanced
• Run ID: 4912b3a3-2f28-4920-8cfb-e4883f2f0926

📥 Commits

Reviewing files that changed from the base of the PR and between 05826ad and 172b99f.

📒 Files selected for processing: 4

• apps/server/src/provider/Drivers/claudeUsage.test.ts
• apps/server/src/provider/Drivers/claudeUsage.ts
• apps/server/src/usage/usageScanCache.test.ts
• apps/server/src/usage/usageScanCache.ts

Included review availability: This review used your included allowance. Your plan provides up to 10 included reviews per hour; 8 remain after this review.

📝 Walkthrough

The Claude transcript parser rejects records with the <synthetic> model marker. The usage scan cache decoder also skips cached Claude records with that marker. Tests cover both filtering paths and verify that ordinary model records remain.

Changes: Claude usage transcript filtering

Layer / File(s)	Summary
Reject synthetic transcript records apps/server/src/provider/Drivers/claudeUsage.ts, apps/server/src/provider/Drivers/claudeUsage.test.ts	The parser rejects records with an empty model or the <synthetic> marker. Tests verify that synthetic records are ignored and a real model record is retained.
Skip synthetic cached records apps/server/src/usage/usageScanCache.ts, apps/server/src/usage/usageScanCache.test.ts	The cache decoder skips Claude records with the <synthetic> marker. A test verifies that an ordinary record remains after decoding.

Priority: ⬇️ Low

Estimated code review effort: 2 (Simple) | ~10 minutes

Change: Bug fix · Severity of issue fixed: Low

Suggested reviewers: juliusmarminge

Merge Risk: ⚪ Minimal · up to 172b9

The change removes synthetic entries from Claude usage while preserving ordinary model records. No actionable merge-blocking risk is evident.

Pre-merge checks: 4

Check name	Status	Explanation
Linked Issues check		Issue #14670 requires the Usage model breakdown to list only real models or explain other entries. parseClaudeRecord now rejects CLAUDE_SYNTHETIC_MODEL, and the shared parser covers line and strea…
Out of Scope Changes check		The parser filter directly fixes issue #14670. The cache-load filter applies the same fix to existing cached transcripts and leaves other providers and real Claude models unchanged. The added tests su…
Title check		The title is concise, specific, and accurately describes the primary change: excluding Claude usage records.
Description check		The description clearly explains the problem, implementation, scope, linked issue, and verification results. It does not use the template's explicit Problem, Change, and Scope and approval headings, b…

Full details: Linked Issues check

Explanation

Issue #14670 requires the Usage model breakdown to list only real models or explain other entries. parseClaudeRecord now rejects CLAUDE_SYNTHETIC_MODEL, and the shared parser covers line and streaming readers. decodeScanCache also removes existing cached Claude records with that model. Tests cover dropping the placeholder and retaining a real-model zero-token record. These changes satisfy the issue requirement without removing billed records.

Full details: Out of Scope Changes check

Explanation

The parser filter directly fixes issue #14670. The cache-load filter applies the same fix to existing cached transcripts and leaves other providers and real Claude models unchanged. The added tests support both changes. The reviewed diff shows no unrelated change.

Full details: Description check

Explanation

The description clearly explains the problem, implementation, scope, linked issue, and verification results. It does not use the template's explicit Problem, Change, and Scope and approval headings, but the required information is present and the description is mostly complete.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create a new PR

• Autopilot · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Autopilot is currently an internal CodeRabbit preview.

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

Dismissing prior approval to re-evaluate 46897b7

[tris203]

Note

Opus 5.5 responding on behalf of @tris203

@coderabbitai review

[coderabbitai]

⚠️ Action not completed

Already reviewed the last commit. Use @coderabbitai full review to rerun a review of the entire changeset.

Note: CodeRabbit is an incremental review system and does not re-review already reviewed commits. This command is applicable only when automatic reviews are paused.

Dismissing prior approval to re-evaluate d9e0047

[juliusmarminge]

Taking over this one for batch claude-01. I’ll rebuild the synthetic-model filtering against current main, cover both newly parsed and cached records with minimal tests, and refresh the live proof before publishing.

[juliusmarminge]

@coderabbitai resume

Dismissing prior approval to re-evaluate 3d7b402

[coderabbitai]

⚠️ Action not completed

Incremental review skipped.