Calibrate token accounting and Agent activity

[Backtthefuture]

Summary

• normalize cached/reasoning token buckets and move Codex history to revision 8 cumulative accounting
• add the source-aware Agent activity dashboard with Today / 7-day views, hourly activity, and cache-coverage-aware metrics
• protect recalibration migrations and add deterministic fixtures, visual renders, CI gates, and a frozen-history audit

Why

Lifecycle token events, compaction, child-Agent replay, and provider-specific cache semantics could inflate totals or present unknown cache coverage as a real zero. Cross-source proxy matching could also remove independent requests when only their timing and usage looked similar. The previous Agent card mixed unverifiable score/rank/work-time semantics.

Impact

Historical Codex totals may decrease after recalibration, but source logs are preserved. Cache hit rate appears only when coverage is complete. Cross-source records are deduplicated only with a shared request/response ID, or a shared session plus matching time/model/usage. Experimental Hermes/ZCode sources remain local-first and opt-in.

Validation

• clean staged-snapshot full app and helper build
• Codex cumulative, CC Switch, migration, Shanghai-calendar, dedupe, and Agent fixtures
• frozen 608-file history audit: PASS at 4,316,026,634 calibrated Token
• Swift/Python audit parity including revision, strategy, cache-creation, and token breakdown
• Simplified Chinese, English, Traditional Chinese, and language-refresh checks
• Agent card and recalibration notice render checks
• local SwiftPM XCTest runner is unavailable because the active CommandLineTools has a PackageDescription linker mismatch; GitHub macOS/Xcode CI is enabled