RunVouch in numbers
These are not benchmarks. They are the last 30 days of the 41 scheduled agents and cron jobs that run the RunVouch and DataSignals Lab businesses, monitored by RunVouch itself. Rebuilt every week; this build is from 2026-08-28.
runs in 30 days
364 in the last 7 days, across 41 agents.
ended in failure
12 runs exited non-zero or reported fail. Each one produced an alert with the stderr excerpt.
green runs without evidence
0.6% of successful runs said ok while the required file, URL or assertion was missing. A ping monitor calls these successes.
median time to alert
From the end of a failed or evidence-less run to the alert being written: it happens in the same database transaction that ends the run. Delivery to Telegram, Slack or e-mail follows within seconds. Missed runs alert at cadence plus grace, by definition.
proof days sealed
291 run leaves in the public Merkle chain since 2026-08-25; 3 day roots anchored in Bitcoin via OpenTimestamps. See the chain.
Alerts by kind, last 30 days
| Kind | Count |
|---|---|
| FAILED | 12 |
| DRIFT | 5 |
| NO_EVIDENCE | 2 |
What the kinds mean is on the alerts page. TEST alerts are excluded. Numbers come straight from the production database at build time; nothing is edited by hand.
How to read this
The failure rate is what you would expect from a fleet of scrapers and LLM jobs hitting external sources: most failures are an upstream 5xx or a rate limit, and the interesting number is not the rate but the time it took to know. The evidence figure is the one a heartbeat monitor cannot produce at all: those runs would have been green.
If you publish your own numbers from RunVouch and want them linked here, say so.