Live robot operations evidence

Commissioning evidence is gathered in a planned deployment window with known test scenarios and an acceptance target. Live operations evidence is gathered under production pressure, across shifts, often by people whose primary job is throughput rather than documentation. That difference changes capture design.

Live evidence must be lightweight enough to collect during an event and structured enough to survive shift handover. It must also tolerate incomplete knowledge: operators should record what they observed and what they tried, including uncertainty, rather than delaying capture until a perfect explanation exists. Deployment Intelligence can refine hypotheses later; it cannot invent a timeline that nobody preserved.

For any event that might escalate, capture robot and site identity, time bounds for the incident window, mission state snapshots, alert identities and acknowledgement records, command history, operator narrative, and links to available media or telemetry exports. If facility conditions were visibly relevant, note them in plain language: congestion, lighting change, charging queue length, nearby equipment downtime, or network symptoms observed by the floor.

If a recovery procedure was used, name the procedure and the outcome. This minimum set lets an investigator start without reconstructing the first hour from memory. Additional telemetry depth can follow once the case exists.

Live evidence becomes investigation evidence when it is attached to a Deployment Intelligence case with stable identifiers and timestamps. Keep original artifacts. Summaries are useful, but summaries without source links decay into opinion.

When multiple events escalate into one recurring-failure case, preserve the individual operational windows rather than merging them into a single vague narrative. Cross-link related cases when the same root cause is suspected. The goal is a chain a reviewer can follow from production symptom to operational record to investigation conclusion to corrective action and retest.

Not every live event belongs in the acceptance package. Live evidence affects acceptance when the event shows that an accepted condition is no longer true, when a workaround from deployment is failing in production, or when a customer requires post-acceptance operational performance evidence under contract. In those cases, the Deployment Intelligence case should reference the acceptance condition explicitly and state whether retesting or acceptance amendment is required.

Do not paste Command acknowledgement streams into acceptance reports. Translate live evidence into the acceptance-relevant claim it supports or challenges, and keep the raw operational record available for inspection.

Decide retention before the first serious incident. Operations evidence often contains facility details and actor identities that require access control. Grant investigators and authorized customer reviewers the access your contracts require, and keep broader operational streams restricted to roles that need them for supervision.

Manufacturer and integrator partners may need event packages without receiving the entire supervision history of the site. Design export packages accordingly. Retention should cover warranty and audit obligations at minimum; longer retention supports fleet learning when policy allows.

Live evidence quality is a habit, not a tool setting. Train new operators on the minimum evidence set during onboarding. Include package completeness in shift-lead handover checks.

Audit a sample of escalations each month for missing identities, missing command history, or narratives that omit procedural steps already tried. Celebrate complete packages in operational reviews the same way teams celebrate recovered uptime. Over multiple quarters, sites with strong live evidence habits investigate faster, argue less about what happened, and produce clearer acceptance or warranty updates when production events require them.

• Define the minimum live evidence set for any event that might escalate
• Capture identities, time bounds, mission state, alerts, acknowledgements, and commands during the event window
• Attach original artifacts to the Deployment Intelligence case, not only summaries
• Preserve separate operational windows when multiple events feed one recurring case
• Reference acceptance conditions only when live evidence actually affects them
• Set retention and access rules before a serious incident forces the decision
• Train operators on uncertainty notes rather than delayed perfect explanations
• Audit escalation packages monthly for completeness across shifts