[currentbits]

Closes #1961.

After a turn completes and the Claude CLI process is replaced (app restart, idle reap, model/effort change), the resumed process reports the session's restored cumulative total_cost_usd and modelUsage. Solenta subtracted from a zero baseline, so it re-billed history: the #1960 fixture recorded $0.05 / 26 inputs for a session that cost $0.04 / 16.

What changed

• Checkpoint (electron/claude-usage.js): a versioned per-session record on the thread with the subtraction baseline plus the raw provider-reported totals, scoped by provider/account ref, remote/WSL target and CLAUDE_CONFIG_DIR. Normalized on load (store-migrate.js); old stores get none.
• Proof before subtraction: the first cumulative result of a fresh --resume process is classified by exact per-model counter identities, never magnitude:
  • restored: some checkpoint model reads exactly checkpoint + this turn's observations, others monotone → subtract the checkpoint;
  • reset: every model reads exactly this turn's observations, or CLI < 2.1.277 → zero baseline (unchanged);
  • unknown (no/mismatched checkpoint, fork, ambiguous) → that turn's cost is marked unavailable (not zero), observed tokens still count, and warm deltas are exact again afterwards.
• Runner (runner-provider-claude.js): hydrate only when session ID + scope match and it isn't a fork; write the checkpoint in the same tick/save as the ledger rows, and only while this handle still owns the session.

Deviation from the issue: there is no no-generation handshake in Solenta, so exact counter identity on the first result is the proof used instead.

Tests

• New electron/test/claude-resume-accounting.test.js (16): restore/reset/unknown, legacy reset above and below the old total, partial pre-result usage, malformed checkpoints, replay, captured 2.1.283 frames (classify as reset), and real runner + Store reload across three app lifetimes, model respawn, old store and account-scope mismatch. Resume now records $0.03/14, then $0.04/16, in thread usage, daily spend and per-model cells.
• claude.test.js / budget-spend.test.js fakes now declare claude_code_version: 2.1.276: a version-less, cost-only resumed result is no longer assumed to be a reset.

🤖 Generated with Claude Code

[chatgpt-codex-connector]

Codex Review Summary: ✅ Completed

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-11-11T10:26:48.903815Z	6887c98	PR opened

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" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.