[spalen0]

Checks organization-wide Tenderly TU usage in the existing hourly profile. A Telegram alert fires when a settled hour consumes at least 3x the busiest hour in the preceding 24 hours and at least 100,000 additional TUs. Complete hourly buckets and the prior peak tolerate normal hourly and daily cron bursts.

The monitor checkpoints its first pending hour before fetching data and processes missed hours chronologically. Normal hours and successfully delivered spikes advance the checkpoint; API failures, missing data, and skipped or failed delivery retain it for a later run. Whole-valued float timestamps are normalized without rounding, while malformed timestamps and values are rejected.

Messages go to Yearn maintenance with the existing Yearn fallback and are stored under yearn-internal. --dry-run evaluates the latest settled hour without reading or writing monitor state or sending messages. The hourly schedule, monitoring metadata, and Tenderly README are included.

Uses the existing Tenderly credential and the dashboard's undocumented usage endpoint. Catch-up uses one weekly history response; if required history is unavailable, the monitor reports an error and retains the checkpoint for investigation. Small leaks can be hidden by normal organization traffic, and sustained high usage enters subsequent baselines.

Validation:

• Full suite: 2236 passed, 4 skipped, 117 subtests passed. Local Safe credentials were isolated and a short temporary path was used for macOS socket tests.
• prek run --all-files passed; full Ruff lint/format and type checks passed.
• Dependency audit found no known vulnerabilities or adverse project statuses.
• Live dry-run against the configured Yearn account succeeded without alerting.
• Regression tests failed before the fixes and now cover hour rollover, multi-hour catch-up, failed/skipped delivery, API/data failures, and timestamp normalization.
• Codex review found no actionable regressions; Gemini was unavailable locally.