Technical note

ROS bags, logs, traces, and the next layer of workcell incident evidence.

Robotics teams already use recorded sensor data, ROS bags, logs, traces, dashboards, and simulation replays when investigating incidents. Metriplane does not try to replace those artifacts. It explores what can happen after raw evidence has enough structure to be replayed and reviewed.

Raw evidence is necessary, but often workflow-specific.

A ROS bag or log archive can capture important details: topics, timestamps, state messages, commands, warnings, and sensor streams. That raw evidence is valuable because it preserves the context needed for later debugging. The challenge is that raw evidence usually needs local knowledge before another reviewer can answer simple questions: what happened, which step was delayed, which asset was missing, and what should become a repeatable regression check.

Metriplane formalizes a structured evidence layer.

In the v0.2.0 assembly-cell evidence demo, replayed workcell state becomes an event timeline. A missing torque driver delays a process step by 35.0 seconds. Metriplane then generates incident INC-0001, an Incident Report (formal artifact name: Cell Truth Report), an evidence bundle, a bundle verification result, and a repeatable regression check.

raw/replayed state
-> event timeline
-> incident context
-> evidence bundle
-> local verification
-> repeatable regression check

Why bundles and repeatable regression checks matter.

Incident evidence is more useful when it can travel with a manifest, checksums, a human-readable report, and a command that lets another reviewer verify the artifact. It becomes more useful again when the incident can be turned into a repeatable regression check that protects against losing the behavior during future development.

Package delivery and evidence remain separate.

Metriplane v0.4.1 is the current installable PyPI package. The workflow and quantitative evidence described in this note remain tied to the frozen v0.2.0 SoftwareX artifact. The newer package improves distribution and verification; it is not retroactive evidence for the older release.

What this does not prove.

The v0.2.0 release artifact demonstrates this workflow on a deterministic, camera-free assembly-cell replay. It does not claim production-factory validation, robot control, safety certification, full 3D reconstruction, marker-free tracking, or quality-release approval.

How to inspect it.

Start with the 3-minute demo, then run the exact camera-free v0.2.0 reproduction path and compare the output against the evidence table. For ordinary package use, install the current v0.4.1 release from PyPI. If something is unclear or weak, leave a comment on Issue #6.