01.4 / SYNTHETIC WALKTHROUGH
From a failed test to a clearer record.
Follow an invented manipulation test. Inspect what changed, what is missing and what an engineer would investigate next.
Try the comparison
All records, authors and evidence identifiers in this example are invented. No customer outcome or time saving is demonstrated.
Open the working demo ↗- Open Robot 01 and select failed run-043. Choose successful run-042 as the baseline.
- Inspect six recorded differences: gripper, Git metadata, firmware, policy model, dataset version and calibration. Grip force remains unknown.
- Expand Selected record to inspect the exact revision, source and observation time. Hardware reports remain manual; synthetic evidence references are unavailable.
- Choose run-044 as baseline to see warnings about time, procedure and conditions. A successful result alone does not make a suitable baseline.
- Read Working setup reference. It preserves the earlier revision for an engineer to investigate. Missing values and artifacts prevent a claim that the setup can be restored.
Differences are facts to investigate, not automatic root-cause findings.
Supported inputs
The local tools and private beta accept bounded metadata. This public demo does not accept uploads.
Manual records and JSON
Robot identities, hardware reports, immutable snapshots, test outcomes and evidence references. Generic tests can use their actual measurements without invented pick counts.
Selected JSON and CSV fields
Review discovered fields, labels, types and units, then reuse approved mappings. Unapproved new fields are not silently captured. CAD identities and revisions are references; geometry is not parsed.
Developer-tool status
Git metadata capture and the Python collector are implemented locally. The ROS 2 adapter awaits live hardware validation. The Onshape connector is implemented with simulated-provider checks; activation and real-provider acceptance remain pending. These are connection methods, not technology partnerships.
Explicit collector sources
Git commit and dirty status, model and calibration hashes, selected settings, timestamps and an operator hardware manifest. The base collector needs no ROS installation. The optional ROS 2 adapter is limited to one approved grip-force parameter.
Raw source, model weights, datasets, recordings, CAD files and credentials are outside ingestion. Current workspaces are limited to 2 MiB and 1,000 entries per record collection.
Continue the investigation
The separate local workspace and private beta support notes, hypotheses, unresolved cases and engineer-confirmed resolutions with an evidence note or retest reference. Reports retain exact run and snapshot references.
Review failures together
The local/private workspace reviews up to 1,000 loaded runs for one robot, suggests comparable earlier successes, groups complete recorded transitions and retrieves prior engineer-confirmed findings. Engineers select the baseline and retain control of conclusions. Export a JSON or text handoff with exact references; unknown evidence cannot establish a repeated match.
Know where work is saved
Local drafts require JSON export before closing. In the private beta, Save to organization persists approved records; changes are not automatically saved. Hosted persistence, organization isolation and pilot measurements have been checked with synthetic data.
Customer access still awaits operational recovery and production checks. No billing, autonomous diagnosis or robot control is provided.
Read the data-handling status ↗Discuss a real regression
The proposed four-week pilot covers one workflow and up to five robots. Pricing is tailored to your team after a demo and discussion.
Agree a baseline, an independent eligible-run count, support effort and success criteria. Actual customer adoption and willingness to pay remain unverified.
Read the proposed pilot ↗Contact the founders with a sanitized description. Do not send private robot records or secrets.