要解决的问题The problem
使用模型中转服务时，本地 token 消耗与服务商扣费之间缺少可核对的统一账本。With a model relay service, there is no single ledger to check local token consumption against the provider’s actual billing.
实现思路Approach
增量解析本地会话记录，采集中转站余额快照，并反推有效倍率和异常差异。Incrementally parses local session logs, snapshots relay balances and derives the effective multiplier and anomalous gaps.
公开证据Public evidence
- 实现 token 同步、余额快照、倍率校准、报表导出和阈值告警Implements token sync, balance snapshots, multiplier calibration, report export and threshold alerts
- 使用 SQLite 保存游标、快照与日聚合，支持后台守护进程SQLite stores cursors, snapshots and daily aggregates, with background daemon support
- CI 覆盖 Python 3.9–3.12，每个版本通过 23 项测试并通过 ruff 检查CI covers Python 3.9–3.12, passing 23 tests and the ruff check on each version
当前限制Current limits
- 账单准确度依赖中转站接口和价格配置Billing accuracy depends on the relay’s API and price configuration
- 仓库具备发布工作流，但目前没有 GitHub Release，不能视为已正式发布The repository has a release workflow but no GitHub Release yet, so it is not formally published