Skip to content

Lineage, audit, telemetry and analytics

Proposed product controls, not deployed general telemetry functionality. Scientific history and diagnostics have different purposes and retention requirements.

Stream Purpose Contract
Scientific lineage Explain exact inputs, operations and results Durable, unsampled customer project records; authorized reading and export
Security audit Record access, changes, exports and authorization Restricted stream with governed access and retention
Operational telemetry Diagnose slow jobs, failed connections and stale data Customer-controlled diagnostic destination; bounded retention and sampling
Product analytics Understand workflow completion, failure and reuse Customer dashboards first; separate opt-in for minimized vendor sharing

The proposed OpenTelemetry instrumentation supplies traces, metrics and logs across capture, publication, OSDU requests, workers and consumer updates. A customer-controlled collector exports to a chosen OpenTelemetry Protocol (OTLP) endpoint. This diagnostic destination is independent of scientific data destinations.

Useful measurements include publication latency, stale-asset age, queue delay, operator duration, reconciliation recovery, failure classes and resource cost. Bound metric labels: do not use unlimited asset or user identifiers as metric dimensions. Redact sensitive content and limit retention.

Trace identifiers can connect diagnostics to a run, but sampled traces cannot be the authority for scientific lineage.

History / Derived from explains inputs, operations and outputs. Data and diagnostics shows the purpose, fields, destination, retention and sharing state of each stream. Local inspection/download, optional diagnostic controls and a customer endpoint make those decisions visible.

Optional sharing with us requires explicit administrator opt-in. Turning it off must produce no vendor-bound analytics traffic, proven by executed tests. Essential provenance remains part of saving scientific work even when analytics is disabled.

Default vendor events exclude payloads, well names, coordinates, paths, free text, prompts, secrets and raw identifiers. Use allowlisted, versioned event schemas and bounded aggregation windows. Hashing an identifier alone does not make it anonymous.

Customers inspect and redact a support bundle before choosing to send it. Training permission is separate from diagnostic or analytics permission. Covert session capture and employee productivity ranking are outside this direction.

M1 defines durable lineage and independent reproduction. M3 adds History/Derived from and manual-edit provenance. M4 qualifies diagnostics, redaction, customer destinations, access-controlled audit and sharing-off behavior. Impact analysis and advanced customer workflow dashboards follow later evidence and demand.

Useful paid tools may include impact graphs, approved workflows, audit integrations, cost analysis and managed operation. Underlying customer history stays readable and exportable; baseline security and essential lineage remain baseline requirements. Packaging and pricing are not decided.

Customer-facing overview · Ownership commitment.