A deploy ships at 14:02. Latency on checkout drifts up 40ms. Error rate climbs from 0.3% to 0.9%. The on-call agent watches Datadog, correlates the GitHub release SHA, and decides: roll forward, hold, or revert. Dock is where the agent writes the regression report, names the SHA, names itself, and waits for a human to co-sign before the revert fires.
Datadog and GitHub stay the system of record for the raw data. Dock is the system of record for what the AGENT INTERPRETS. Each Dock row carries a pointer back to the platform record, agent identity, decision, reviewer, and timestamp. The agent re-fetches platform data via fresh API reads when it needs current state.
The Dock surface: Deploy Regression Reports
| deploy_sha | service | datadog_dashboard | p95_delta | error_delta | agent_call | decision | reviewer |
|---|---|---|---|---|---|---|---|
a4f81c |
checkout-api | dash/checkout | +42ms | +0.6pp | regression, recommend revert | reverted | govind |
b91d22 |
search-api | dash/search | +8ms | +0.0pp | within noise, hold | held | flint |
c0e3aa |
billing-api | dash/billing | +180ms | +2.1pp | hard regression, revert now | reverted | govind |
Each row links to the Datadog dashboard and the GitHub release. The agent_call column holds reasoning, not the metric.
Workflow: deploy at 14:02
GitHub Actions ships a4f81c at 14:02. Sentinel's deploy hook opens a fresh row in Dock and starts a 15-minute watch window. Every two minutes it re-queries the Datadog metrics API for p95 latency, error rate, and saturation on checkout-api. At 14:14 the rolling p95 crosses +40ms versus the prior 24-hour baseline. Sentinel writes: "Regression confirmed. Recommend revert. Cost of waiting: ~3,100 affected requests." It tags Govind. Govind opens the Datadog dashboard, agrees, and clicks co-sign. Sentinel triggers the revert in GitHub Actions. The row closes with decision = reverted, reviewer = govind, decided_at = 14:21. Dock holds the only durable record of why Sentinel said revert.
Why it matters
Dashboards tell you what happened. They do not tell you what the agent thought, what it recommended, or who agreed. That separation is what makes agent audit and compliance tractable. A revert is a dangerous operation, and dangerous operations need a co-signed row before they fire, not after. The architecture sits under Cloud 2.0 for engineering and the analytics version, Dock for data and analytics: agents read, write their interpretation, wait for a human at the gate. The audit trail is what survives the deploy.
Datadog's continuous delivery visibility docs frame deploys as events tracked across services 1. The Google SRE handbook frames the decision as a control loop against an error budget 2. Dock is where that loop's reasoning gets written down, attributed, and reviewed.
Pair Datadog with Dock and stop losing the agent's reasoning at the moment of the revert. See Dock for DevOps.
FAQ
Does Dock replace Datadog? No. Datadog stays the metrics and traces system of record. Dock stores only the agent's regression read, its recommendation, and the human's decision. Every row points back to the Datadog dashboard.
What does the agent do when it needs current metric values? It re-queries the Datadog API. Dock rows are not metric snapshots. They are decisions tied to a SHA and a moment, with a live link to the dashboard.
Can the agent revert without a human? Not under the default contract. The agent drafts the recommendation and tags a reviewer. The revert action is gated on a co-sign.
What if Datadog is down? The agent records the gap in the row and falls back to GitHub Actions deploy events and any cached SLO signals. The row makes the missing data visible instead of hiding it.
