Identity mismatch
Quarantine the record and mechanism until configuration and calibration binding are resolved.
Level-seven field evidence tool · 01
SO-101 calibration and bounded-motion evidence record
A printable and CSV-ready field record for binding one physical arm pair to its software identity, calibration files, motion evidence and retest decision.
Evidence statusReusable engineering record · blank template · no certification claimField definitions reviewed 4 August 2026
How to use
Freeze the object and boundary before recording any result.
Use one record per exact leader/follower pair and calibration identity. The record should make a later reviewer able to reconstruct what moved, under which limits, with which files, and why the result was accepted, rejected or scheduled for retest.
- 01Assign a record owner and a separate motion-stop operator.
- 02Create an immutable folder for configuration, calibration, raw signals, photographs and trial outcomes.
- 03Write the task, workspace, object set, speed and duration boundaries before powered motion.
- 04Record failed and interrupted trials with the same discipline as successful trials.
Field dictionary
Every conclusion should resolve to a field, raw evidence and a rejection condition.
| Code | Field | Input format | Evidence to preserve | Reject condition |
|---|---|---|---|---|
| ARM-001 | Record identity | Unique record ID, date/time, owner and reviewer | Signed header and immutable evidence-folder link | Duplicate ID, missing owner or mutable destination |
| ARM-002 | Physical build | Leader/follower serial, SO-101 revision, servo models/IDs, gripper, material and substitutions | As-built table and labelled photographs | Unidentified mechanism, motor or substitution |
| ARM-003 | Software baseline | LeRobot commit/release, dependency lock, OS, ports and persistent robot identifiers | Lockfile, configuration export and environment hash | Environment cannot be reconstructed cleanly |
| ARM-004 | Calibration identity | File names/checksums, mechanism binding, joint ranges, offsets and creation session | Original calibration files plus verified recovery copy | File belongs to another mechanism or hash is missing |
| ARM-005 | Motion envelope | Maximum speed, relative target limit, workspace, payload/object set and test duration | Approved pre-test boundary sheet | Boundary written after the run or left qualitative |
| ARM-006 | Motion trace | Command/state timestamps, joint error, current, temperature, stop events and video clock | Raw export with schema and time-base note | Unsynchronised, edited-only or summary-only evidence |
| ARM-007 | Trial outcome | Declared task, attempt number, outcome, intervention, failure code and evidence pointer | Complete trial matrix including aborted attempts | Missing failures or changed success definition |
| ARM-008 | Disposition and retest | Accept/reject/rework, unresolved deviations, configuration hash, retest trigger and signatures | Reviewer-signed release record | Acceptance has no bounded claim or named reviewer |
Run matrix
Record success, failure, aborts and human intervention in one matrix.
| Attempt ID | Configuration hash | Calibration hash | Task & boundary | Command/state evidence | Outcome | Intervention/failure | Disposition |
|---|---|---|---|---|---|---|---|
This blank structure is not evidence of a passed test, certification, compliance opinion or permission for unsupervised operation.
Failure taxonomy
Name the failure so the next run can be materially different.
Unexpected motion or divergence
Stop power, preserve raw trace, inspect mapping and repeat only from the previous passed gate.
Thermal/current excursion
Stop, record cooling state, inspect load and limits; do not erase the event by restarting.
Evidence or time-base failure
Reject the affected trial even if the visible task appeared successful.
Release questions
Discuss a bounded release only when every question points to evidence.
- 01Can another operator identify the exact pair and reload the same calibration without oral knowledge?
- 02Do raw traces, video and trial outcomes share a documented time base?
- 03Are every failure, intervention and excluded result visible in the trial matrix?
- 04Is the accepted claim limited to the exact build, object set, workspace and speed?
- 05Is every change that requires recalibration or retest explicitly named?
FUURAA analysis
A record should reconstruct the decision—not decorate the result.
This record turns calibration from a hidden setup step into a versioned engineering object. The decisive link is mechanism identity → calibration hash → configuration hash → raw trial evidence → bounded disposition.
Primary sources