The direct answer
Does a changed specification prove that the component changed? No. It first proves that two source surfaces publish different statements. A product change is supported only when the responsibility party identifies the affected revision, ordering code, serial/date boundary or change notice.
RCI found nine selected numeric field differences across seven products. Two—the Ouster OS1 point-rate and frame-rate fields—sit behind an explicit Rev7-to-Rev8 generation boundary. The other seven remain current documentation conflicts across six products. They must not be silently merged into one “best” value.
A dated PDF can be stronger than an undated page, but “newer” does not establish which shipped unit it describes. RCI preserves both values until the responsible manufacturer publishes the missing boundary.
Three evidence states prevent false version histories
| Class | RCI test | What may be concluded | Selected count |
|---|---|---|---|
| Revision-bounded change | The responsibility party publicly names the before/after generation or revision. | The published product-family specification changed at that boundary; exact-SKU compatibility still requires proof. | 2 |
| Current official-source conflict | Current responsibility-party pages, manuals or datasheets differ without mapping the affected hardware. | Only that the documentation bundle is unresolved. | 6 |
| Same-page conflict | One current responsibility-party page contradicts itself. | Only that the page cannot support a single unqualified value. | 1 |
RCI excludes editorial extraction changes from manufacturer-drift counts. If RCI later parses an existing table more accurately, that is an RCI correction—not evidence that the manufacturer changed the component.
Nine differences across every hardware layer
| Product / field | Published value A | Published value B | Evidence state | Design consequence |
|---|---|---|---|---|
| Ouster OS1 Perception & sensing · Maximum point rate | 5.2 million points/s | up to 10,485,760 points/s | Revision-bounded change Approximately 2× higher headline point rate in the Rev8 surface. | Point-cloud bandwidth, host processing, storage and network sizing can change materially between generations sold under the OS1 family name. |
| Ouster OS1 Perception & sensing · Maximum frame rate | 20 Hz maximum | 5, 10, 15, 20, 30 or 40 Hz | Revision-bounded change Maximum listed frame rate increases from 20 Hz to 40 Hz. | Control-loop timing, motion distortion, data rate and integration tests depend on the selected generation and operating mode. |
| Orbbec Gemini 335 / G40155-170 Perception & sensing · Ideal depth-range lower bound | 0.26–3 m optimal range | 0.3–3 m ideal range | Current official-source conflict Lower bound differs by 0.04 m; rounding or a test-boundary difference is not publicly resolved. | Near-field placement and acceptance criteria should not assume that 0.26 m and 0.30 m are interchangeable test boundaries. |
| CubeMars AK70-10 KV100 Motion & actuation · Actuator mass | 621 g | 521 g | Same-page conflict 100 g absolute difference; approximately 19.2% relative to 521 g. | A 100 g joint-level difference affects link inertia, payload allocation and whole-robot mass models. |
| Hesai QT128 / QT128C2X Perception & sensing · Single-return point rate | 1,152,000 points/s | 864,000 valid points/s | Current official-source conflict 288,000 points/s absolute difference; the page value is 33.3% above the manual value. | Network, logging and perception throughput can be undersized or overstated if emitted, global-resolution and valid-point definitions are mixed. |
| Hesai QT128 / QT128C2X Perception & sensing · Sensor mass | 700 g | 900 ±15 g | Current official-source conflict Nominal values differ by 200 g; 900 g is approximately 28.6% above 700 g. | A 200 g roof- or mast-mounted difference affects structural load, vibration, center of gravity and vehicle dynamics. |
| Advantech MIC-713-OX Compute & control · Headline AI performance | up to 100 TOPS | up to 157 TOPS with Super mode and JetPack 6.2+ | Current official-source conflict 57 TOPS headline difference; the higher value is software-mode and ordering-configuration dependent. | Capacity plans can be wrong when module memory, installed JetPack and Super-mode support are omitted from a family-level TOPS number. |
| RRC power solutions RRC2054-2 / P/N 1100640 Power & energy · Battery-pack mass | 440 g | 450 g typical | Current official-source conflict 10 g absolute difference; approximately 2.3% relative to 440 g. | The difference is small for one pack but matters in multi-pack mass budgets and signals that the live page and controlled document are not synchronized. |
| Kvaser USBcan Pro 2xHS v2 / EAN 73-30130-00752-9 Interfaces & integration · Minimum classic CAN bit rate | 20 kbit/s minimum | 50 kbit/s minimum | Current official-source conflict 30 kbit/s difference at the lower bound; the datasheet excludes 20–49 kbit/s modes listed by the page and guide. | A buyer targeting 20 or 33.3 kbit/s legacy networks cannot establish support from the current document bundle without clarification. |
The sample covers motion, perception, compute, power and interfaces. It is not a prevalence estimate: RCI selected material numeric differences already exposed by the v0.4.5 evidence catalog and then re-opened the responsibility-party sources on 2026-08-10.
The two defensible version changes are both Ouster OS1 fields
The Rev7-generation OS1 surface published 5.2 million points per second and a 20 Hz maximum frame rate. The Rev8 OS1 table publishes up to 10,485,760 points per second and frame-rate options through 40 Hz.
The key evidence is not simply that the second page has larger numbers. Ouster’s 2026-05-04 Rev8 launch explicitly names a new generation, calls the OS family upgraded and compares it with the previous generation. That public boundary supports a family-level change classification.
It still does not make “OS1” a complete procurement identity. A buyer must record Rev7 or Rev8, exact ordering code, channels, horizontal resolution, returns, Ethernet option and firmware. RCI does not infer that a Rev8 mode can be applied to a Rev7 unit or that the generations are mechanically, electrically or software compatible.
The other seven differences remain open
Motion: CubeMars AK70-10 mass
The same current product page publishes 621 g in the product-specific table and 521 g in its core/comparison surface. The 100 g difference is large enough to affect link inertia, but the page does not say whether one figure includes a driver, cable, fastener or another revision.
Perception: Orbbec Gemini 335 range
The Gemini 335 page gives an optimal range of 0.26–3 m. The linked V1.6 datasheet dated 2025-04-10 gives 0.3–3 m. The 40 mm lower-bound difference may be rounding or a test/mode boundary; its change log does not resolve that question.
Perception: Hesai QT128 point rate and mass
The current QT128 product page publishes 1,152,000 points/s for single return and describes a 700 g sensor. The Q03 manual dated 2026-04-10 publishes 864,000 valid points/s for single return and 900 ±15 g.
The point-rate labels may use different counting semantics; RCI therefore does not call either one wrong. The mass difference is clearer, but the page’s general note that batches, hardware versions and customizations may differ does not map 700 g or 900 ±15 g to an order code.
Compute: Advantech MIC-713-OX TOPS
The current product page says up to 100 TOPS. The datasheet dated 2026-07-30 says up to 157 TOPS and conditions Super mode on JetPack 6.2 or later; its ordering table shows that installed JetPack and Super-mode support vary by order code. Neither headline alone is an adequate system-performance identity.
Power: RRC2054-2 mass
The current product page publishes 440 g. Its linked revision-J datasheet publishes 450 g typical. The PDF exposes neither a date nor a change history, so the 10 g difference cannot be turned into a product-revision claim.
Interfaces: Kvaser minimum CAN bit rate
The USBcan Pro 2xHS v2 page and current revision-4.5 user guide publish 20 kbit/s as the lower classic-CAN limit. The datasheet linked from the same current download surface publishes 50–1000 kbit/s. The guide’s revision history does not identify a minimum-bit-rate change, so support below 50 kbit/s remains a validation question.
A procurement-safe workflow
- Freeze identity first: manufacturer, exact model, ordering code, hardware revision, serial/date range and options.
- Archive every cited surface: document ID, revision, publication date, retrieved date, URL and cryptographic hash where rights allow local retention.
- Keep values source-bound: retain units, conditions, precision, firmware, operating mode and source location. Never average conflicting numbers.
- Classify the difference: revision-bounded change, current cross-source conflict, same-page conflict or RCI editorial correction.
- Escalate by consequence: ask the responsibility party to resolve fields that affect safety, fit, load, timing, bandwidth, power or compatibility.
- Validate the incoming configuration: inspection or testing closes the as-built boundary; a web page cannot do that.
Even the Ouster transition must keep its Rev7/Rev8 label. The remaining seven require clarification or configuration-level validation before one value can replace the other.
What a manufacturer change record should publish
- Stable document identity: document number, revision, publication date and an immutable old-revision archive.
- Affected product identity: model, ordering codes, hardware revision, serial/lot or manufacturing-date range.
- Field-level delta: exact before and after values, units, tolerances, conditions and definitions.
- Reason and consequence: correction, test-method change, software mode, component substitution or hardware redesign—plus compatibility impact.
- Lifecycle links: product-change notice, end-of-life notice, approved alternatives and qualification evidence.
- Machine-readable change log: stable IDs that let catalogs and AI systems cite the right revision without guessing from a mutable page.
Method and limitations
RCI compared responsibility-party sources already attached to catalog v0.4.5, inspected historical frozen RCI releases for candidate drift, and then re-opened current official pages and documents. Only numeric differences with exact source locations were retained. The audit uses 15 unique official source surfaces across 7 manufacturers.
This is a documentary audit, not a teardown, calibration or incoming inspection. RCI did not ask manufacturers to adjudicate the selected differences before publication and did not test the seven products. Search-engine snapshots and RCI’s own frozen catalogs can prove what RCI observed, but they are not substitutes for a manufacturer-controlled archive. The sample is intentionally bounded and cannot estimate an industry-wide conflict rate.
Responsibility-party sources
The source list below publishes exact locations and revision labels without redistributing the documents. Repeated source IDs are shown once; linked owners retain their respective rights.
Ouster OS1 product page — Rev7 generation
Rev7/L3 OS1 headline specifications · 5.2M points/s; 20 Hz maximum frame rate · Rev7 generation · page observed before the Rev8 launch
Ouster OS sensor overview — Rev8
Rev8 OS1 specification table · Up to 10,485,760 points/s; 5, 10, 15, 20, 30 or 40 Hz · Rev8 / OS1-080-xxx
Ouster Releases the Rev8 OS Family
Launch date and explicit previous-generation upgrade boundary · Rev8 is an upgraded OS family and provides up to double the range and resolution of the previous generation · 2026-05-04
Orbbec Gemini 335 product page
Gemini 335 specifications and range footnotes · 0.10–20 m+; optimal range 0.26–3 m · Current page verified 2026-08-10
Orbbec Gemini 330 Series Datasheet for USB Devices
Product Specifications, Working Range and Ideal Range · 0.1–20 m+; ideal range 0.3–3 m · V1.6 / 2025-04-10
CubeMars AK70-10 KV100 product page
Product-specific table: 621 g; core specifications / AK-series comparison surface: 521 g · 621 g and 521 g on the same page · Current page verified 2026-08-10
Hesai QT128 product page
Product benefit text: “only 700 g” · 700 g · Current page verified 2026-08-10
Hesai QT128C2X User Manual
Section 1.5 Specifications: Weight · 900 ±15 g · Q03 / 2026-04-10
Advantech MIC-713-OX product page
Main Features · Embedded NVIDIA Jetson Orin NX up to 100 TOPS · Current page verified 2026-08-10
Advantech MIC-713-OX Datasheet
Features and Ordering Information · Up to 157 TOPS; Super mode with JetPack 6.2+; support varies by order code · 2026-07-30
RRC2054-2 current product page
Battery Dimensions and Safety Features · 440 g · Current page verified 2026-08-10
RRC2054-2 Product Data Sheet
Battery Dimensions · 450 g typical · DS_RRC2054-2_J · public date not stated
Kvaser USBcan Pro 2xHS v2 product page
Tech Specs: Bit Rate · 20–1000 kbit/s · Current page verified 2026-08-10
Kvaser USBcan Pro 2xHS v2 User’s Guide
Table 6: Technical data · 20 kbit/s to 1 Mbit/s · Revision 4.5 / 2025-01-13 · current file generated 2026-02-19
Kvaser USBcan Pro 2xHS v2 Datasheet
Technical Data: Bitrate · 50–1000 kbit/s · Ref. No. 98308-3 v2.0 · public date not stated
Download the audit
Download the immutable v0.1.0 release as JSON or CSV. Stable current aliases are /data/research/component-datasheet-version-drift.json and /data/research/component-datasheet-version-drift.csv.
The dataset includes classifications, both values, normalized differences, revision-boundary flags, procurement consequences, required actions and full source trails. It does not redistribute third-party document bodies.
Suggested citation: Robot Component Index, “Robot Component Specification Drift: 9 Field Differences Across 7 Products,” RCI 022, research dataset v0.1.0, 2026-08-10.