Nine files. Three version DOIs. Zero raw test traces.

The direct answer

Is LT-PAM an open fabrication package? Yes, within the deposit RCI reviewed. The current Zenodo record contains three STL meshes, three editable Autodesk Inventor files, an eight-row BOM, a README and an assembly video. All nine files match Zenodo’s published byte counts and MD5 values. The six CAD/STL design files are byte-identical across all three public version records.

Does that make the reported force, contraction, cost and cycle life independently reproducible? No. The deposit contains no raw force–length trace, pressure trace, cycle trace, specimen table, calibration record, failure log or analysis code. RCI did not build or pressure-test an actuator. Performance values remain author-reported source claims.

Can the approximately USD 5 statement be reconstructed exactly? No. For a 150 mm nominal actuator, the video’s cut-length instructions and BOM yield ¥583.27 in known consumed parts before cement, printed parts and tools, or ¥861.27 if the entire cement container is charged. Printing cost, cement allocation, waste, shipping, tax and a dated exchange rate are not published.

Three records, but only one explicit version label

The February record is labelled 1.0. The July and August records have separate version DOIs but no explicit version label in the Zenodo metadata. From the first to the second record, the BOM was added and the README changed. From the second to the current record, only the video changed, reducing the package by 176,569,771 bytes.

The design files did not drift: all three IPT files and all three STL files preserve the same byte count and MD5 in every record. That is useful evidence of design-core stability, but it does not establish that a physical build, material lot or test specimen stayed unchanged.

The current nine-file package matches its registry

RCI downloaded 65,953,142 bytes and matched every file to the current Zenodo record. The deterministic manifest SHA-256 is e18840be5844824547dd56cad35322a4bb987553891c99f9597c048151748018.

File integrity means the local bytes match the deposited registry. It does not prove that the CAD represents an as-built specimen, that supplier parts are unchanged or that the assembly video documents the specimen used for the reported tests.

Three closed meshes—with a units warning

Each binary STL has one connected triangle component, zero degenerate triangles, zero boundary edges and zero edges shared by more than two triangles. RCI therefore marks all three as watertight two-manifolds by edge count.

What the BOM can—and cannot—price

The BOM has 8 rows: seven materials/components and one reusable tool. Two printed-part rows have blank totals. Summing disclosed material procurement quantities gives ¥1,624.2, including a full optional cement container; excluding that optional container gives ¥1,346.2. The reusable crimping tool is ¥2,290 and is correctly separated from per-unit material cost.

The current video says the silicone tube should be 20 mm longer than the desired actuator and the sleeve about 10 mm longer than the tube. Applying those rules to a 150 mm actuator gives a 170 mm tube and 180 mm sleeve. The known consumed sleeve, tube, clamp and fitting subtotal is ¥583.27.

That calculation is not a complete factory or laboratory cost. It excludes print material, printer time, failed prints, allocated cement use, scrap, tax, shipping and a dated JPY/USD conversion. The source’s approximately USD 5 statement should remain a cost claim with those assumptions attached.

The assembly video changed more than its compression ratio

The current file is 17.473 seconds shorter, 74.21% smaller and changes from nominal 60 fps to 30 fps while retaining 1920×1080 dimensions. Its hash is different.

Visual review also found a changed caption and claim set. The initial Japanese ending reports 200 N tension at 0.6 MPa. The current English ending instead reports approximately 25% no-load contraction, repeated lifting of about 20 kg and stable stroke after more than 9,000 cycles. These are source-video statements, not measurements made or reproduced by RCI.

Performance claims stay source-reported

The initial README contained additional numeric statements about force, contraction, operating pressure and cycle count that the current README no longer repeats. The current Zenodo description still presents rupture, recommended-load and durability values. RCI records that source-level drift rather than choosing one version as the verified specification.

The associated DOI is presently a preprint, and the current README says the HardwareX manuscript is under review. RCI could inspect the public abstract and metadata, but SSRN served a security-verification page instead of the full text. No paper table, calibration method or reported result was treated as independently reviewed.

Fabrication reproducibility is not performance reproducibility

The package is actionable for fabrication: it identifies off-the-shelf parts, provides printable meshes and proprietary editable CAD, describes print settings and shows assembly. Its three editable sources are Autodesk Inventor IPT files; there is no STEP, STP, IGES or IGS neutral exchange file.

A performance reproduction package would additionally need specimen IDs and dimensions, material lot and print records, pressure and temperature traces, calibrated force/displacement instrumentation, raw quasi-static and cycle data, failure definitions, rejected-run logs, analysis code and uncertainty treatment. None of those raw mechanical-test surfaces appears among the nine files.

RCI’s conclusion is therefore intentionally asymmetric: the deposit supports independent integrity and geometry checks and offers a practical fabrication starting point; it does not allow the reported force, rupture, contraction or fatigue values to be recomputed from source traces.

Source, attribution and rights boundary

The current source is “Low-Cost and Tough Pneumatic Artificial Muscle (LT-PAM): Design Files and Documentation” by Keisuke Naniwa, Yasuhiro Sugimoto, Daisuke Nakanishi, Yoichi Masuda, version DOI 10.5281/zenodo.21862439. The Zenodo record-level metadata is marked CC BY 4.0.

The current README separately states CERN-OHL-P-2.0 for design files and CC BY 4.0 for documentation/video. No standalone LICENSE or NOTICE file is deposited, and the split does not explicitly assign the editable BOM. RCI does not resolve that ambiguity by assumption and does not mirror any source file.

The associated preprint is DOI 10.2139/ssrn.6480279. The Crossref metadata reviewed by RCI did not provide an item license. The Zenodo documentation declaration must not be silently extended to the preprint.

What RCI independently did

1. Resolved the concept DOI and three public version records: 18665923, 21132366, 21862439.
2. Downloaded all nine current files, matched every Zenodo byte count and MD5, computed SHA-256 for each file and generated one deterministic manifest hash.
3. Compared filename, size and MD5 histories across all three records to identify additions and modifications without relying on narrative version notes.
4. Parsed the BOM as CSV, separated material/component and reusable-tool rows, and applied the video’s length rules to a stated 150 mm case.
5. Parsed each binary STL triangle, counted vertices, connected components, degenerate triangles and edge incidence, and recorded coordinate bounds without assigning a unit.
6. Probed initial and current videos for duration, bytes, bitrate, codec, frame rate and dimensions; visually reviewed their caption changes.
7. Counted raw-test, standalone-license, editable-CAD and neutral-CAD files and kept every physical-performance conclusion outside the independent audit boundary.

The deterministic audit is maintained as scripts/analyze-lt-pam-open-hardware.mjs. It can verify the public release when the three Zenodo metadata records and the source files are available locally.

Limits that stay attached

• RCI did not fabricate an LT-PAM, inspect a physical specimen, verify material lots, pressurize the actuator or measure force, contraction, leakage, rupture or fatigue.
• The full SSRN preprint was not reviewed because SSRN served a security-verification page; only public DOI metadata and abstract-level information were available.
• STL coordinate units are not embedded in the binary files, so RCI reports coordinate-unit extents rather than asserting millimetres from the files alone.
• Zero boundary or non-manifold edges by edge count does not establish sliceability, dimensional accuracy, airtightness, pressure rating or mechanical strength.
• The approximately USD 5 statement depends on unpublished printing, consumable-allocation and currency-conversion assumptions.
• The latest two Zenodo records omit explicit version labels, and file-level license scope remains ambiguous for the BOM.
• Author-reported performance values are preserved as source claims and must not be represented as RCI verification or product certification.

Download the audit

Download the immutable JSON release and file-level CSV. Stable aliases are current JSON and current CSV.

Suggested citation: Robot Component Index. LT-PAM Open-Hardware Audit: Nine Files, Three Versions, Zero Raw Test Traces. RCI 027, research dataset v0.1.0, 2026-08-10. https://robotcomponentindex.com/research/lt-pam-open-hardware-reproducibility-audit/