[icesixgod]

Summary

Add an explicit Calculate task timing action to the native task panel and MCP viewer. It partitions the selected task's complete active history into tool execution, recorded and estimated model time, and unattributed time, excluding gaps between turns and counting overlapping tools once. Distinct coverage counters preserve missing, invalid, zero, and sub-millisecond timing without changing public trajectory schemas.

Load timing with the MCP dispatcher and show restart guidance for stale runtimes and internal timing failures. Keep the task-panel and MCP viewer entries, remove the global left-toolbar registration, and synchronize the English and Chinese documentation.

Privacy impact

Timing reads the selected task's safe summary and reuses its validated source/index when available. Cached sources are revalidated against the current authorized active/archive roots; cold exact-ID lookups prefer matching rollout filenames, with existing bounded metadata resolution retained for legacy names and prefixes. The new temporary-index fields contain timing quality and boundary flags. Task logs, redaction, full-detail consent, and public summary defaults remain unchanged. Tests and smoke fixtures use synthetic logs.

Validation

• uv run ruff format --check .
• uv run ruff check .
• uv run mypy
• uv run mypy --no-incremental --platform win32
• uv run pytest --cov --cov-report=term-missing — 623 passed; total coverage 92.01%; all focused coverage floors passed.
• RUN_UI_TESTS=1 uv run pytest -m ui -q — 152 passed, covering English/Chinese, narrow/wide panels, native/MCP surfaces, retries, task switches, stale runtimes, and escaped metadata.
• Python 3.14.6 backend suite with resource warnings treated as errors — 623 passed; packaged MCP smoke passed.
• Release metadata validation and packaged MCP stdio smoke.
• Reproducible Windows launcher source/binary parity.
• ZIP and tar.gz archive safety/parity checks — 99 byte-identical files; extracted MCP smoke passed for each format.
• English and Chinese documentation are synchronized.
• Reviewed changed paths and added content for private files; no common credential-pattern matches.

Validation ran locally on macOS with Python 3.10.18 and 3.14.6. The Linux/Windows runtime matrix and native Windows smoke remain for PR CI.