See the science.
Understand the workflow.
Customer visibility includes what produced a result, who accessed it and why a connection or job failed. Each record has its own purpose and controls.
| Record | What the customer learns | Control and retention |
|---|---|---|
| Scientific lineage | What produced this result, from which exact inputs and recipe? | Durable customer project data. Authorized reading and portable export; never sampled away. |
| Security audit | Who read, changed, exported or authorized what? | Restricted customer audit records, with policy-defined access and retention. |
| Operational telemetry | Why is a job slow, an asset stale or synchronization failing? | Customer diagnostic destination; bounded retention, redaction and sampling. |
| Product analytics | Which workflows complete, fail or need improvement? | Customer reports first. Sharing with us is optional, minimized and explicitly opted into. |
Understand delays.
Find the failing step.
Planned OpenTelemetry instrumentation follows connector capture, publication, OSDU requests, worker execution and consumer updates. Traces, metrics and logs can reveal stale assets, queue delays, failed transfers, reconciliation recoveries and resource cost.
Export diagnostics through a customer-controlled collector and chosen endpoint. Diagnostic traces can link to scientific lineage, but their sampling and shorter retention make them unsuitable as the only derivation record.
Explore the durable scientific history →Make every stream
inspectable and controlled.
The proposed Data and diagnostics experience shows each stream’s purpose, fields, destination, retention and sharing state. Customers should be able to inspect or download diagnostic records and control optional collection.
Vendor analytics requires separate administrator opt-in. Turning sharing off must stop vendor-bound analytics traffic. Essential provenance remains part of saving scientific work regardless of analytics preferences.
Support bundles are inspected and redacted before a customer chooses to send them. Lineage relationships require access checks too: a restricted input name can reveal information even when its samples are hidden.
Improve workflows.
Respect customer boundaries.
Customer reports can show reuse, stale outputs, bottlenecks, compute cost and reproducibility coverage. Optional aggregate feedback can help us prioritize compatibility fixes and product improvements.
The intended vendor event schema excludes scientific payloads, coordinates, well names, file paths, free text, prompts, secrets and raw identifiers. Hashed identifiers alone do not establish anonymity. Covert session capture and employee productivity ranking are outside this direction.
Diagnostic sharing and AI training require separate decisions. Neither is a condition for retaining your scientific history.