The mechanism
Every RCI record must meet a claim-level bar: exact model identity, per-claim source URL and location, preserved conditions, explicit unknowns. That bar is the site’s integrity mechanism—and it is also a filter. A manufacturer with per-SKU datasheets, stable URLs and revision histories can clear it cheaply. A manufacturer that publishes a marketing page with three numbers and no conditions cannot, so it never enters the catalog at all. The result is selection on transparency: the index grows fastest where independent scrutiny adds the least, and stays silent where it would add the most. Silence is the worst possible output for those vendors, because absence from an index reads as neutrality, not as a finding.
Where it shaped the catalog
- Interfaces: ten of eleven records are CAN interface devices, because Kvaser, HMS, Vector and PEAK-System publish per-SKU documentation of unusual quality. The audit is RCI 059.
- Power: eight of fourteen records are portable smart packs from vendors whose whole business is publishing exact electrical and certification documents, while proprietary 48 V robot system packs—mostly undocumented in public—have zero records. The audit is RCI 060.
- Actuation: coverage clusters on vendors with open datasheets and open-source robot ecosystems; integrated joints from humanoid manufacturers that publish nothing per-SKU are absent, not evaluated.
The tension with data discipline
The obvious fix—lower the bar and index opaque vendors from resellers, forum posts and teardown rumors—is worse than the bias. Distributor pages are not specification evidence; secondhand numbers without conditions are exactly the flattened data RCI exists to replace. The bar stays. What changes is the assumption that a vendor who fails the bar produces no record at all.
The counterweight: documentation-gap records
RCI is introducing a record type whose subject is the absence itself. A documentation-gap record names an exact vendor and product family, lists the specification fields a buyer would need (electrical envelope, duty boundaries, certification basis, lifecycle policy), and states—with a check date—which of them the vendor publishes, where, and which it does not. No values are invented; the record’s claims are about the documentation, which is itself verifiable against public reality. The evidence-state system extends naturally: an unpublished continuous-torque rating is a U with the vendor named, not silence.
Three boundaries keep this honest. First, a gap record is not an accusation of bad data—unpublished is a fact, wrong is a separate finding requiring separate evidence. Second, gap records cannot be paid away; a vendor closes one only by publishing, and the record then converts to a normal catalog entry with a visible history. Third, gap coverage starts where selection pressure is highest—48 V system packs and humanoid joint modules—rather than where it is easiest.
This is the one structural answer RCI has to the reviewer’s hardest point. An index of the transparent is a convenience; an index that also documents opacity, vendor by vendor and field by field, is doing something the manufacturer’s own website will never do.
Corrections
No corrections yet. Material changes will be logged with timestamps on the Changes page.