The direct answer
Does RCI’s first stage cover more than joints? Yes. This matrix joins six immutable audits across vision, LiDAR, main compute and battery management. It makes those four non-joint layers explicit in one machine-readable dataset.
What public evidence exists? The six source releases expose 1,992 test-surface path observations under project-specific definitions, 32 public check runs with success conclusions, and 12 downloaded files whose publisher digest records matched RCI’s copies.
Did RCI test the devices? No. RCI ran zero cameras, LiDARs, Ryzen AI devices, BMS hardware or robot systems for this matrix. None of the six audited release surfaces currently provides one retained validation bundle joining exact hardware, firmware/runtime, commands and structured physical results.
All six rows have fixed release identities and inspectable public evidence. Three expose successful public checks, and three include downloaded publisher-digest matches. These facts support release traceability and bounded software-evidence statements—not device performance, safety, compatibility or production readiness.
One evidence frame across four non-joint layers
| Layer | Release | Test-surface paths | Successful public checks | Downloaded digest matches | Retained validation bundles | RCI device test |
|---|---|---|---|---|---|---|
| Vision | RealSense SDK 2.0 2.58.3 RCI 036 | 380 files under explicit unit-test paths | 28 | 8 | 0 | no |
| Vision | DepthAI Core 3.8.0 RCI 031 | 311 test-related source-archive paths | 0 | 3 | 0 | no |
| 3D LiDAR | Ouster SDK 1.0.0 RCI 032 | 371 test-related files in the fixed source tree | 3 | 1 | 0 | no |
| 3D LiDAR | Livox SDK2 1.3.1 RCI 035 | 0 explicit test paths in the fixed tree | 0 | 0 | 0 | no |
| Main compute | AMD Ryzen AI Software 1.8.0 RCI 033 | 25 test-related fixed-tree paths | 1 | 0 | 0 | no |
| Battery management | foxBMS 2 1.11.0 RCI 034 | 905 files under the fixed tests tree | 0 | 0 | 0 | no |
The row labels are intentionally narrower than product categories. “Vision” here means two software release surfaces, not every depth camera; “battery management” means one foxBMS release, not a complete robot pack or safety certification.
1,992 path observations are not 1,992 tests
The total is the arithmetic sum of six different source-tree observations: 380 RealSense unit-test-path files, 311 DepthAI test-related paths, 371 Ouster test-related files, zero explicit Livox test paths, 25 Ryzen AI test-related paths and 905 foxBMS files under tests/.
Those definitions are not interchangeable. A path may be source, fixture, helper, configuration or documentation; one test may span several files, and one file may define several tests. The total describes public inspectability only. It is not a pass count, coverage percentage, maturity score or cross-project ranking.
Thirty-two successful checks are real—but heterogeneous
RealSense contributes 28 successful GitHub checks, Ouster three and Ryzen AI one. The Ryzen AI check was an update-pip-graph check; Ouster’s three were not publicly named build/test checks in its audit; RealSense’s checks span three successful workflows but do not disclose one complete physical-camera matrix.
DepthAI, Livox and foxBMS expose zero checks in the public GitHub surfaces queried by their fixed audits. That is a public-disclosure result—not proof that their maintainers performed no private CI, external validation or hardware testing.
Twelve downloaded digest matches identify bytes, not behavior
| Release | Downloaded publisher-digest matches | Archive comparison depth | Boundary |
|---|---|---|---|
| RealSense SDK 2.0 2.58.3 | 8 | two generated archives matched every fixed-checkout path, regular-file byte and symlink target | The release name says beta while the API says prerelease=false. RCI did not execute assets, build the SDK, update firmware, connect a camera or validate robot integration. |
| DepthAI Core 3.8.0 | 3 | source ZIP and TAR normalized path sets matched; three GitHub assets matched RCI-computed SHA-256 records | Source paths and issuer integration statements are not a passed device matrix, retained execution result or robot workload benchmark. |
| Ouster SDK 1.0.0 | 1 | two generated archive path sets matched; one PyPI sdist matched its publisher SHA-256 | The three successful public checks were not named build/test checks; no retained sensor benchmark bundle or RCI LiDAR test was found. |
| Livox SDK2 1.3.1 | 0 | all 221 files in both generated archives matched fixed Git blobs | Six sample targets, configurations and protocol links are not six tests, firmware compatibility proof, LiDAR performance evidence or robot integration. |
| AMD Ryzen AI Software 1.8.0 | 0 | two generated archive normalized path sets matched; runtime and LFS payloads were outside the downloaded source surface | The public repository declares itself a subset of the full release; RCI did not download EULA-gated runtime/model payloads or run a Ryzen AI or robot workload. |
| foxBMS 2 1.11.0 | 0 | two generated archives matched normalized paths, file counts and expanded bytes | Public test source is not a release pass; HIL files, coverage outputs and exact hardware result bundles were not published or executed by RCI. |
RealSense accounts for eight matches, DepthAI three and the Ouster PyPI sdist one. The other archive checks use RCI audit-date hashes, normalized paths, expanded bytes or fixed Git blobs rather than publisher-provided distribution digests. The matrix preserves those different proof depths instead of inventing one “verified release” score.
The missing object is a retained physical-validation bundle
A strong device-level release package would join exact hardware revision or serial, firmware/runtime/model bytes, host/toolchain, commands, test selection, fixtures and scenes, calibration, raw outputs, structured results, failures and uncertainty. None of the six audited release surfaces currently exposes that complete object.
This does not negate their source, asset or CI evidence. It limits what downstream engineers, procurement teams and AI systems should infer from it.
Twelve cross-release disclosure checks
| Field | Status | Public evidence | Why it matters |
|---|---|---|---|
| fixed release identity | 6-of-6-disclosed | Each row links an immutable upstream RCI release audit with tag and fixed commit. | Cross-release synthesis stays traceable to exact source records rather than current project homepages. |
| four non joint system layers | covered | Vision, LiDAR, main compute and battery management each have at least one fixed release row. | The first-stage site scope is broader than joint actuation. |
| test surface normalization | bounded-not-comparable | The matrix stores each project's path definition beside its count. | 1,992 is a disclosure sum, not a test count, pass count, coverage measure or ranking. |
| public check conclusions | 3-of-6-have-public-checks | RealSense exposes 28 successful checks, Ouster three and Ryzen AI one; the other three rows expose none in their audited public GitHub surfaces. | Zero public checks does not prove no private or external validation occurred. |
| publisher digest matches | 12-downloaded-files-matched | Eight RealSense assets, three DepthAI assets and one Ouster PyPI sdist matched publisher digest records. | Digest agreement identifies downloaded bytes; it does not establish execution, safety or reproducible builds. |
| archive comparison depth | heterogeneous-and-explicit | Each row states whether RCI compared path sets, expanded bytes, fixed Git blobs or a complete fixed checkout. | Archive checks with different depth must not be collapsed into one binary provenance score. |
| retained release validation bundle | 0-of-6-currently-available-in-audited-surfaces | No row exposes a current retained bundle joining exact hardware, firmware/runtime, commands and structured physical results. | Source and CI visibility alone cannot reproduce device behavior. |
| physical device testing by rci | 0-of-6 | The six upstream audits all state that RCI did not run the relevant camera, LiDAR, Ryzen AI device or BMS hardware. | The matrix is a release-evidence synthesis, not an independent hardware benchmark. |
| robot system testing by rci | 0-of-6 | No upstream audit reports an RCI robot workload or full-system integration test. | A component software release cannot be generalized to robot compatibility or autonomy performance. |
| firmware runtime and model identity | not-uniformly-fixed | RealSense firmware bytes, Livox firmware pairs and Ryzen AI runtime/model payloads remain outside a complete common release-result package. | Exact hardware behavior depends on more than a repository tag. |
| rights boundary | row-specific | Upstream audits preserve root licenses, third-party notices, binary terms, linked-hardware terms and trademark boundaries separately. | The matrix republishes RCI-derived metadata and analysis, not upstream code, binaries or hardware packages. |
| ranking boundary | no-performance-ranking | No row contains a common device, firmware/runtime, host, workload, protocol and result metric. | The matrix supports evidence-quality questions, not product speed, accuracy, safety or procurement rankings. |
Method and limits
RCI joined six previously published immutable release audits by their fixed RCI dataset versions. For each row, it selected the nearest available test-source path count, public GitHub check-run count and conclusion, downloaded distribution whose publisher digest was matched by RCI, current retained validation-result bundle, and RCI physical-device/robot test boundary. The six upstream audits used different repository structures and path heuristics, so test-surface counts are summed only as disclosure observations and are not comparable test counts, pass counts, coverage values or performance scores.
- The 1,992 path total combines different source-tree heuristics and must not be used as a test-count or maturity ranking.
- The 32 successful check conclusions include different workflows and do not form a common device-validation protocol.
- Publisher digest matches identify downloaded bytes only; RCI did not execute or safety-scan distributions for this matrix.
- The matrix covers one selected fixed release per project, not the entire support history or latest-current state after the audit date.
- No conclusion establishes exact hardware/firmware/runtime compatibility, physical performance, functional safety or robot-system readiness.
Every project row links its immutable upstream RCI JSON audit. Readers should use those source audits for release-specific archive hashes, commit identities, compatibility wording, license boundaries and missing-evidence details.
Download the matrix
Download the immutable JSON release and record-level CSV. Stable aliases are current JSON and current CSV.
Suggested citation: Robot Component Index. “Six Robot Software Releases: 1,992 Test-Surface Paths, 32 Successful Public Checks, Zero RCI Device Tests.” RCI 037, version 0.1.0, 2026-08-10. https://robotcomponentindex.com/research/non-joint-robot-release-evidence-matrix/
For version-aware BibTeX and machine citation records, use Cite RCI. For evidence labels and correction rules, use the methodology.