RCI 055 · Editorial transparency note

Most RCI records are sourced, not tested.

The catalog is useful as a traceable map of public claims. It is not yet an independently tested component index. This note states the current evidence composition, explains the concentrated publication burst and sets a slower, verification-led editorial direction.

Published 2026-08-238 min readCatalog 0.4.150 RCI physical tests
1,080manufacturer claims (89.6%)
6independently verified claims (0.5%)
119derived, conflicting or unknown
0RCI physical component tests

The criticism is substantially fair

RCI has done real work: it fixes exact model identities, preserves units and stated conditions, exposes conflicts and publishes source locations in machine-readable form. But the word independent was too prominent relative to the composition of the evidence. A reasonable reader could interpret “Independent component evidence” or “independent robot component database” as a claim that the catalog had been independently measured. It has not.

The current catalog contains 1,205 claim-level records across 120 exact models. Of those, 1,080 are manufacturer claimed, 6 are independently verified, 1 is RCI derived, 6 are conflicting and 112 are unknown. RCI has completed no EV4 physical bench test. The six V records come from qualified external evidence; they are not six RCI hardware runs.

Direct correction

Editorial independence and independent verification are now described separately throughout the site. RCI is independently operated. The catalog is predominantly structured manufacturer disclosure, and its physical-test count is zero.

What the evidence mix supports

StateCountShareWhat it supports
M · Manufacturer claimed1,08089.6%A traceable statement by the manufacturer under the recorded condition. Not independent verification.
V · Independently verified60.5%A qualified external record supports the bounded claim. Not necessarily an RCI physical test.
D · RCI derived10.1%A reversible calculation from cited source values.
C · Conflicting60.5%Credible sources disagree; RCI has not resolved the disagreement.
U · Unknown1129.3%No adequate public evidence was found for the selected field.

This mix supports source discovery, specification-boundary checks, due-diligence questions and the identification of missing evidence. It does not support a broad statement that the products were independently tested, that recorded values are correct in practice, or that one component is safe, compatible or superior.

The MID‑360 record is a useful example. Its six published fields are all M-state claims from the same Livox product record. RCI preserves the 10%/80% reflectivity and 100 klx range conditions, which is better than copying a single headline range. It still remains data organization, not independent LiDAR measurement.

What “independent” means—and no longer means

RCI uses independently operated only for the editorial relationship: vendors cannot purchase inclusion, evidence states or conclusions. The site no longer uses “Independent component evidence” as a global tagline. Independent evidence is now reserved for the V state or for a declared EV5 external replication. Physical work is separately declared as EV4, EV5 or EV6.

This distinction also applies to software and data audits. Matching a release hash, inventorying a repository or recomputing a public dataset can be valuable EV2 or EV3 work. None of those actions demonstrates camera accuracy, actuator torque, battery endurance or complete robot compatibility.

The August publication cadence looked industrial because it was

34 of the site’s 61 current publications carry dates of 10 or 11 August 2026. 28 titles use “Zero” or “0”, and many use the same colon-separated numeric construction. The intention was to make scope and missing evidence visible in search results. The effect was a templated research shelf whose volume could reasonably raise doubts about editorial depth.

The dates are publication dates, not evidence that each article was independently researched from scratch on that day. Several notes were assembled from shared structured datasets and repeatable audit frames. That reuse improves consistency, but publishing the entire batch without a visible cadence explanation made the process look more human-intensive than the output demonstrated.

AI assistance is part of the workflow and is now stated plainly

RCI uses automated and AI-assisted tools for source discovery, candidate-field extraction, translation, deduplication, structure and drafting. These tools do not become a source and do not upgrade a claim’s evidence state. The RCI Editorial Team remains responsible for publication, citations, boundaries and corrections.

That policy reduces hallucination risk but does not eliminate review-capacity risk. A source-linked sentence can still omit context, apply the wrong variant or receive less scrutiny than its length suggests. Readers should use the claim ledger and upstream source for decisions, report errors through the corrections channel, and treat reading-time labels as estimates—not proof of review effort.

What changes now

  • Evidence composition appears on the homepage. M, V, D/C/U and physical-test counts are visible before any coverage totals.
  • Editorial independence is separated from evidence independence. The header, footer, metadata, About page and press language no longer imply that the whole catalog is independently verified.
  • Research output slows down. New long-form audits should answer a distinct decision question, disclose the highest EV work level and avoid the repeated “N / M / Zero” title formula.
  • Updates prioritize verification depth. Re-checks, corrections, cross-source validation, physical-test acquisition and results take precedence over adding another batch of lightly differentiated source audits.
  • AI assistance remains disclosed. AI can assist retrieval and drafting; it cannot be cited as evidence or substitute for accountable editorial review.

RCI will keep the earlier pages and titles visible because erasing the pattern would hide the history that prompted this correction. Future changes will be logged, and old machine-readable releases remain immutable.

Review the methodology and EV0–EV6 work levels, the AI assistance policy, the public change log and the current component catalog JSON. Corrections and evidence can be submitted through the public corrections channel.