What changed, and why.

2026-09-01 · Site release 0.13.0 / second-critique response: honest headlines, source concentration, coverage audits

• Verified a second external critique point by point against frozen data and published the results as RCI 057; most factual points held.
• Inverted every evidence headline on the homepage, evidence page and data page to lead with the strict numbers: 33 of 120 models with direct external component measurements and 14 of 1,205 claims independently supported. The broader 90-model bounded-coverage figure is now stated second, as context.
• Replaced the completed self-set 90-model milestone with an open one: direct component measurements for all 120 catalog models (33 done, 87 remaining, 27.5%). The evidence-status JSON milestone metric changed accordingly.
• Published per-model source concentration: every component page now shows how many distinct manufacturer URLs stand behind its manufacturer-published claims, with an explicit note when the answer is one. Site-wide: 84 of 120 models are single-URL. Audit in RCI 058.
• Relabeled the RCI024 pre-registration file as a protocol registration, not a dataset; it contains zero observations and is now listed under that heading.
• Added current-composition lines to each catalog layer on the component index and published coverage-gap audits for the interface layer (10 of 11 records are CAN interface devices, RCI 059) and the power layer (zero 48 V system packs, RCI 060).
• Named the published-source inclusion rule as a structural bias on the component index and in RCI 061, and defined documentation-gap records as its counterweight.

Reason: the critique was substantially correct; honest framing that requires no decoding is a prerequisite for the evidence work still outstanding.

2026-08-26 · Site release 0.12.10 / evidence-language correction

• Kept the 90/120 model-linked physical-evidence coverage milestone while removing language that could be read as a 75% independent-verification rate.
• Separated the 33 direct-component measurement models from system-context, family-level, certification and publisher-hosted evidence.
• Kept claim outcomes unchanged: 14 supported, 214 inconclusive, 977 unassessed and zero RCI physical tests.

Reason: evidence coverage and independent claim verification are different denominators and must not be conflated.

2026-08-26 · Site release 0.12.9 / 90 physical-evidence models

• Expanded bounded physical-evidence coverage from 65 to 90 of the 120 exact component records, adding 25 model-linked external records across actuation, sensing, compute, interfaces and batteries.
• Published 121 external physical-measurement rows and 284 claim mappings covering 228 unique catalog claims; 14 aggregate to supported, 214 to inconclusive and 977 remain unassessed.
• Kept revision and system-context limits explicit. Real-system use and family-level captures remain inconclusive where the exact subrevision, metric or manufacturer test condition was not reproduced.
• RCI-run physical component tests remain zero.

Reason: broaden practical exact-model evidence without converting deployment evidence into false performance verification.

2026-08-26 · Site release 0.12.8 / 65 physical-evidence models and supplier landscapes

• Corrected the earlier 15-model headline: nine responsibility-party BOM crosswalks no longer count as independent physical evidence.
• Completed the declared 65-model physical-evidence milestone by adding 37 bounded records across actuation, sensing, compute, interfaces and batteries.
• Separated direct component measurements from system-context experiments and certification-lab evidence; the latter two do not validate manufacturer performance envelopes.
• Split coverage reporting into 19 models with direct component measurements and nine additional models supported by bounded system-context experiments or embedded external V records, preventing deployed use from being presented as isolated performance characterization.
• Added external-measurement sections to affected component pages, with tested conditions, claim mapping, outcome and source location.
• Published six bounded supplier landscapes across integrated actuators, 3D cameras, 3D LiDAR, edge compute, batteries and industrial networks, covering a defined 111-profile source universe.

Reason: replace record-count optics with exact-model physical-evidence coverage and make same-category supplier discovery useful without claiming a global market census.

2026-08-25 · Site release 0.12.7 / 200 mapped assessment records

• Completed the 200-record mapped-assessment milestone: six catalog V records, 37 peer-reviewed study observations, nine responsibility-party exact-model crosswalks and 148 source-condition audits.
• Expanded unique claim-assessment completion from 35 to 183 of 1,205. Nine aggregate to supported, 174 to inconclusive and 1,022 remain unassessed.
• Distributed the 148 new source-condition assessments across motion actuation, perception sensing, compute control and power energy using six already-published exact-model audit datasets.
• Kept the evidence boundary explicit: source-condition audits add no independent measurement, crosswalks establish identity rather than performance and RCI physical tests remain zero.

Reason: complete the requested 200-record milestone without duplicating evidence rows or relabelling editorial source audits as independent validation.

2026-08-25 · Site release 0.12.6 / 35 unique claim assessments

• Expanded unique claim-assessment completion from 13 to 35 of 1,205: 9 aggregate to supported, 26 to inconclusive and 1,170 remain unassessed.
• Published 52 mapped evidence records across 15 exact models: six catalog V records, 37 peer-reviewed study observations and nine responsibility-party exact-model crosswalks.
• Mapped additional D455, Mid-70, Jetson Orin Nano and RPLIDAR A1M8 study boundaries without promoting unmatched metrics or conditions to support.
• Classified responsibility-party crosswalks as public exact-model identity evidence only; they do not validate product performance and do not count as RCI physical tests.
• Advanced the 200-record support milestone to 52 complete, 148 remaining and 26% complete while preserving the zero-physical-test boundary.

Reason: raise claim-level assessment completion to the requested 35-claim checkpoint while keeping evidence-class, outcome and physical-test accounting explicit.

2026-08-25 · Site release 0.12.5 / 24 external-result claim mappings

• Mapped every one of the 24 external study results to a named catalog claim with a relationship class, assessment outcome and written rationale; the unmapped external-result count is now zero.
• Kept evidence-row and unique-claim accounting separate. Six catalog V records plus 24 external mappings produce 30 assessment records, but repeated D455, Mid-70, Jetson Orin Nano and RPLIDAR A1M8 results cover 13 unique catalog claims rather than 30.
• Aggregated the 13 unique claim outcomes to 7 supported and 6 inconclusive, leaving 1,192 of 1,205 claims unassessed. Metric and condition mismatches remain inconclusive instead of being promoted to support.
• Published each mapping and rationale in the public evidence table, claim ledger and machine-readable JSON; the 200-record support milestone remains 30 complete, 170 remaining and 15% complete.

Reason: turn the 24 literature findings into auditable claim-assessment records without treating repeated experiments as distinct claims or converting non-matching metrics into false verification.

2026-08-24 · Site release 0.12.4 / 200-record independent-support milestone

• Set the current operational milestone at 200 qualified independent support records. The published set remains 30, leaving 170 records and placing milestone progress at 15%.
• Kept the milestone separate from the long-term all-claim program: the frozen catalog still contains 1,205 claims, claim-level assessment completion remains 6, the claim backlog remains 1,199 and RCI physical tests remain zero.
• Published the 200 target, current count, remaining count and progress percentage in the evidence page, homepage, Coverage, Component index, Data, press facts, release manifest and evidence-status JSON.
• Defined milestone admission as qualified records with exact-model identity, traceable source location, stated conditions and an explicit limitation boundary; a target does not authorize synthetic or duplicated evidence.

Reason: establish a measurable intermediate delivery target between the current 30-record evidence set and the longer-term 1,205-claim assessment objective without presenting planned work as completed verification.

2026-08-24 · Site release 0.12.3 / 30 independent support records

• Expanded the independent-support set from 6 to 30 records across 6 exact models by adding 24 condition-bounded measured findings from four peer-reviewed studies.
• Added seven RealSense D455 metrology results, seven Livox Mid-70 characterization and registration results, four Jetson Orin Nano 8GB workload results and six SLAMTEC RPLIDAR A1M8 calibration results, each with source location, apparatus or workload boundary and check date.
• Kept the accounting dimensions separate: the frozen catalog remains 1,205 claims with 6 V records, claim-level completion remains 6 of 1,205, the assessment backlog remains 1,199 and RCI physical tests remain zero.
• Published all 30 records in the human-readable evidence table and machine-readable evidence-status JSON; updated homepage, Coverage, Component index, Data, press facts and machine-readable summaries from the same release object.

Reason: increase useful independent evidence immediately without misrepresenting literature extraction as RCI testing or silently counting several results from one study as completed assessments of unrelated manufacturer claims.

2026-08-24 · Site release 0.12.2 / all-claim independent-assessment target

• Set the public long-term target to independent assessment of every claim in the current frozen catalog: 1,205 of 1,205, with 6 independently supported today and 1,199 remaining.
• Separated original claim provenance from the independent-assessment outcome. Manufacturer wording remains preserved while a second dimension may support, contradict, remain inconclusive or be marked not applicable.
• Published a human-readable evidence-status ledger and machine-readable JSON with one accountability row for every one of the 1,205 catalog claims, including the current six independently supported claims, 1,199-row backlog, five covered exact models, zero RCI physical tests, accounting rules and completion definition.
• Published six zero-record coverage debts: force/torque sensors, tactile sensing, grippers/end effectors, robot cables, standalone connectors and slip rings, each with an admission boundary and first source-review targets.
• Added evidence-status links to the primary navigation, homepage, component index, Coverage, Data and press facts; replaced the remaining footer “facts” wording with claim-specific language.

Reason: make independent assessment of the complete claim ledger—not catalog volume—the site’s measurable north-star objective while preserving contradictory and inconclusive results instead of assuming every manufacturer claim will be confirmed.

2026-08-24 · Site release 0.12.1 / evidence-first field guide and search semantics

• Published RCI 056, a practical verification guide that requires exact identity, metric meaning, operating conditions, hardware boundary, evidence state and source date to align before two robot-component values enter one comparison.
• Added worked examples for the Livox MID‑360 reflectivity-qualified range, Unitree GO‑M8010‑6 maximum-versus-continuous torque boundary and the unresolved maxon 734439 nominal-torque source conflict.
• Added site-wide WebPage identity, organization authorship, explicit snippet controls and Article Open Graph publication metadata derived from each page’s visible structured data.
• Replaced the blanket weekly sitemap timestamp with exact 2026-08-24 last-modified entries only for pages materially changed in this release; unchanged URLs no longer receive a synthetic freshness signal.
• Updated the homepage research introduction, RSS feed, press facts and machine-readable release labels to 56 publications and site release 0.12.1.

Reason: publish a genuinely new, decision-useful article while improving crawl, entity and citation clarity without claiming that technical eligibility guarantees indexing, retrieval, citation or ranking.

2026-08-23 · Site release 0.12.0 / evidence and coverage transparency

• Replaced the global “Independent component evidence” tagline with “Source-linked component claims” and separated editorial independence from independent verification across the homepage, footer, About page, structured metadata and press boilerplate.
• Placed the current catalog evidence composition on the homepage: 1,080 manufacturer claims, 6 independently verified claims, 1 RCI-derived record, 6 conflicts, 112 unknowns and zero RCI physical component tests.
• Published RCI 055, an editorial transparency note addressing the 10–11 August publication burst, repeated numeric/“Zero” headline pattern, AI-assisted workflow and review-capacity risk.
• Added category and manufacturer concentration disclosure to Coverage, including the current 43/28/24/14/11 category split and the 55.8% share of motion-actuation records contributed by the four most represented manufacturers.
• Confirmed by direct production requests that current homepage, component index, About, Methodology and Research pages return one 0.11.0/0.4.15/2026-08-21 footer with Cloudflare DYNAMIC status; added response validators, explicit release headers, sitemap last-modified metadata and stronger distribution gates to reduce stale crawler snapshots after this release.

Reason: make the site’s evidence depth, editorial process, sample bias and deployed-version identity visible before readers infer more verification or coverage than the public record supports.

2026-08-21 · Site release 0.11.0 / procurement evidence wave 4

• Expanded procurement evidence from eight to fifteen exact catalog records and from thirty-six to fifty-six manufacturer-published fields.
• Added RealSense D435i and D457, RRC2040-2 and RRC2057, Kvaser Leaf v3 and PCIEcan HS v2, and Raspberry Pi Compute Module 5.
• Preserved packaging and family boundaries: D435i peripheral and boxed identifiers remain separate, Kvaser connector variants are excluded, and the CM5 price is explicitly a family starting price rather than an exact-variant quote.
• Kept RRC records at official quote-route level where document identifiers were not captured, instead of upgrading linked compliance downloads into unverified certificate claims.

Reason: reach the fifteen-model procurement milestone with source-specific boundaries intact before production visual QA and deployment.

2026-08-21 · Site release 0.10.0 / procurement overview

• Added a public procurement overview covering all eight evidence-backed exact components in one sortable release-derived view.
• Each row exposes the manufacturer and exact model, evidence-field count, thirteen-requirement procurement coverage, earliest review date and every unresolved requirement.
• Linked every overview row directly to its component evidence ledger and kept the percentage explicitly bounded as record completeness rather than a product or supplier score.
• Added a build gate that verifies representative records and the first high-volatility review date render on the public data page.

Reason: make the procurement layer discoverable and actionable without requiring buyers to inspect 120 component pages one by one.

2026-08-21 · Site release 0.9.9 / procurement refresh policy

• Advanced procurement evidence to schema 2 and release 0.3.1 without changing its eight-record, thirty-six-field evidence boundary.
• Assigned every field a deterministic review interval and due date: fourteen days for high-volatility price and channel evidence, ninety days for medium-volatility identifiers, lifecycle and warranty evidence, and 365 days for low-volatility declarations and change-control policies.
• Published each review schedule on component pages and in the JSON release. Passing a due date creates a re-check obligation; it does not imply that the underlying value changed.
• Added build gates for review metadata and volatility-to-interval consistency.

Reason: turn dated procurement observations into an explicit maintenance queue so commercial facts cannot silently age into apparent permanent specifications.

2026-08-21 · Site release 0.9.8 / procurement evidence wave 3

• Expanded procurement evidence from six to eight exact records and from twenty-six to thirty-six manufacturer-published fields.
• Added Jetson Orin Nano 8GB order identity, quantity-qualified price, lifecycle horizon, official sales route, warranty and change-control policy, explicitly separated from the developer kit.
• Added Kvaser Ethercan HS exact EAN, displayed price, manufacturer purchase route and warranty.
• Re-checked NVIDIA's live FAQ and corrected the high-volatility 1KU+ prices for Jetson AGX Orin 64GB and Jetson Orin NX 16GB to the values displayed on 2026-08-21.

Reason: extend procurement coverage and demonstrate that volatile commercial facts are dated observations requiring re-checks, not permanent catalog specifications.

2026-08-21 · Site release 0.9.7 / procurement evidence wave 2

• Expanded the procurement evidence layer from four to six exact records and from sixteen to twenty-six manufacturer-published fields.
• Added Jetson Orin NX 16GB production-module part numbers, quantity-qualified suggested price, lifecycle horizon, official channel route, warranty and change-control policy.
• Added Kvaser U100P exact EAN, displayed direct price, manufacturer purchase routes and warranty while keeping stock, lead time, regional fulfillment and quote validity unresolved.
• Replaced hand-written procurement release labels and counts on the data page and component pages with values derived from the release object.

Reason: increase buyer-facing depth using exact manufacturer evidence while preventing the procurement layer itself from developing the version and count drift it was created to expose.

2026-08-21 · Site release 0.9.6 / procurement evidence wave 1

• Published sixteen manufacturer-supported procurement fields across four exact records: RealSense D455, NVIDIA Jetson AGX Orin 64GB, RRC2054-2 and Kvaser USBcan Pro 2xHS v2.
• Captured only bounded facts such as exact SKU or part-number pattern, quantity-qualified price, official sales route, lifecycle horizon, warranty, declaration or test-report identifier and PCN/EOL policy. Every field carries source location, check date, volatility and limitations.
• Connected the new evidence layer to decision-readiness coverage without rewriting frozen catalog 0.4.15. Component pages now show the commercial evidence separately from the technical claim ledger.
• Observed that the catalog's HEBI actuator landing URL currently returns 404; no procurement evidence was inferred from the failed route.

Reason: begin closing the procurement-data gap with auditable manufacturer facts while preserving the distinction between a public commercial statement, a live quote and actual stock.

2026-08-21 · Site release 0.9.5 / decision-readiness coverage

• Published component-readiness v0.1.0 for all 120 catalog records, separating category-specific engineering fields from thirteen common procurement fields.
• Every selected field is labelled published, explicitly unknown or missing. Unknown and missing fields receive no completeness credit, and the resulting percentage measures record coverage rather than product quality.
• Added engineering and procurement coverage panels to every component page, with field-level gaps and a visible reference-record, engineering-shortlist or procurement-evidence-complete stage.
• Added aggregate readiness coverage to the Coverage page and a complete machine-readable JSON export. No compatibility, certification, availability or purchasing recommendation is inferred from the result.

Reason: expose the gap between claim-level reference evidence and actual engineering or procurement decision support instead of allowing record count alone to imply decision readiness.

2026-08-20 · Site release 0.9.4 / release-manifest consistency repair

• Published one machine-readable release manifest for the site, component catalog, manufacturer intelligence directory and source directory; public version labels and headline counts now derive from those release objects.
• Corrected the Coverage page from its stale catalog 0.4.5 and 2026-08-10 boundary to the current catalog 0.4.15 release metadata, and removed historical evidence counts that no longer described the current catalog.
• Separated the two manufacturer scopes on the homepage: 55 manufacturers represented in the exact-model catalog versus 111 directory profiles, of which 55 are exact-model linked and 56 are source-screened.
• Aligned package metadata and the lockfile at site release 0.9.4, updated the site review date, and added build gates that reject inconsistent footers, stale Coverage boundaries and manufacturer-label scope drift.

Reason: ensure that RCI applies its own version-drift discipline to every public release label and never presents a catalog-linked subset as the complete manufacturer directory.

2026-08-19 · Site release 0.9.3 / actuation depth wave 1 and version single-sourcing

• Released component catalog v0.4.15 with 120 exact records, 55 catalog-linked manufacturers and 1,205 claims. The first actuation-depth wave adds 17 motion-actuation records (26→43 in that category) across three clusters: quasi-direct-drive joint modules and bus servos (Unitree GO-M8010-6, mjbots qdd100 beta 3, MYACTUATOR RMD-X8-P9-25-C-N, FEETECH ST-3215-C001 and ST-3215-C018, Damiao DM-J4310-2EC V1.2 24 V), strain-wave and cycloidal reducers (Harmonic Drive SHG-20-100-2UH and CSG-17-100-2UH, Nabtesco RV-25N ratio 81, Leaderdrive LCS size 20 ratio 100, Laifual FSS size 20 ratio 100), and frameless torque motors and servo drives (Kollmorgen TBM2G-05013-A-NNAA-00, TQ-RoboDrive ILM-E85x13, maxon EC frameless DT 50 S 734439, FAULHABER 3268 G024 BX4, ODrive Pro, Elmo G-TWI 25/100).
• Ten manufacturers enter the catalog-linked index: seven new to the catalog — Unitree Robotics, mjbots, FEETECH, Damiao, ODrive Robotics, Leaderdrive and Laifual Drive, including five Chinese suppliers — plus TQ-RoboDrive, FAULHABER and Elmo Motion Control, whose existing directory profiles gain their first exact records; the manufacturer intelligence directory advances to v1.6.0 with 111 profiles, 55 exact-model-linked and 56 source-screened. Linkage remains coverage, not partnership, availability, compatibility or physical validation.
• Every added value is a conditioned manufacturer claim (225 M) or an explicit unknown (35 U); independently verified claims stay at 6 and RCI physical tests stay at zero. Rated, average, repeated-peak and momentary reducer torques are carried as four separately conditioned values; backlash, lost motion and hysteresis keep the manufacturers’ exact terms; the FEETECH C001/C018 records carry their exact electrical boundaries alongside the still-open SO-101 power-chain conflict.
• Recorded, without merging, that Leaderdrive LCS-20-100 and Laifual FSS-20-100 publish an identical 40/82/49/147 N·m torque set on their own official pages under different rated-speed definitions and different backlash classes (≤10 versus ≤20 arcsec); no equivalence or substitution claim is made for any manufacturer pair.
• Published RCI 054 / Data Brief 004: catalog release delta v0.4.14 → v0.4.15 — 17 added records, 260 added claims, 10 added catalog-linked manufacturers, 0 removed IDs, V unchanged at 6; all 945 prior claim IDs remain byte-identical and the integrated-joint comparison table membership is unchanged at 21 pending a condition-preserving extension.
• Advanced the citation registry to v0.39.0 with the v0.4.15 catalog asset; 34 versioned assets remain and all earlier frozen releases stay immutable.
• Single-sourced the public version labels: the components index page and the site footer now derive the catalog and source-directory versions from the release data instead of hand-written strings, fixing a stale “Catalog v0.4.2” label that had survived on the components page since that release. Build gates now fail on any stale hardcoded catalog version label on public pages and extend the retired-experiment sweep to every public HTML page.

Reason: move the deepest first-stage category from symbolic one-model coverage toward selection-grade depth — including Chinese suppliers — without converting documentation growth into verification claims, and make version labels structurally unable to drift again.

2026-08-18 · Site release 0.9.2 / research backfill and source re-verification

• Published RCI 046, a launch audit of the Robot System Index terminal data layer (v0.8.21): recomputed all headline counts from the released dataset, tabulated the ten platforms carrying the 20 exact row-level component relations, documented the four rejection classes and restated the EV3 ceiling with zero physical validation and zero rebuild packages.
• Published RCI 047 / Data Brief 003: catalog release delta v0.4.7 → v0.4.14 — 27 added components, 346 added claims, 19 added manufacturers, 0 removed IDs, independently verified claims unchanged at 6. All 346 added claims sit on the 27 new records; the 599 prior claims are byte-identical.
• Published RCI 048, a manufacturer directory progress audit from v1.0.0 to v1.5.2: 104 profiles, 45 exact-model-linked, 59 source-screened and 15 sector dossiers each linking one exact record, with all figures recomputed from the live directory release and zero new validation claims.
• Published RCI 049, a power-chain conflict field note documenting three incompatible public supply-voltage statements — the responsibility-party 5 V supply path, the FEETECH 6–7.4 V servo range and the Waveshare 9–12.6 V controller input — for the SO-101 follower servo bus. The conflict is held open with zero RCI physical tests, and corrections are invited through the public evidence channel.
• Published RCI 050, a TurtleBot3 Waffle Pi field note mapping the 46 official shared-table rows per configuration column, the XM430-W210-T ×2 and OpenCR1.0 ×1 exact-model crosswalks and the preserved Pi 3B→3B+→4B and LDS-01→LDS-02 boundaries, with exact Raspberry Pi, camera and LDS revisions left unresolved.
• Published RCI 051, an AlohaMini2 field note documenting the ten-node, nine-relation Host/Client system map lifted from 42 pinned BOM rows, while holding open the unexplained four-cable/two-controller count, battery-branch allocation and camera protocol/hub/synchronization unknowns.
• Published RCI 052, a Berkeley Humanoid Lite field note on the nine-node, ten-relation system map, restating that live GitBook/Sheet documentation is an observation layer distinct from release v1.1.0 and that the two IMU paths are alternatives, not simultaneous.
• Published RCI 053, a ToddlerBot field note covering the five pinned configurations — full-body variants at 30 active positions, gripper variants at 32 and the teleoperation leader at 14 with software IDs from 16 — derived only from pinned YAML filtered against fixed XML at one commit, with actuator counts kept at family level and compute kept at recommendation level.
• Re-checked all 18 unique external primary-source links across the eight evergreen guides: 17 confirmed reachable and one retained as automation-restricted behind an interactive verification gate, with zero sources removed. Each guide now displays a sources-re-checked date alongside its unchanged version and publication date.
• Refreshed the machine-discovery summary to the current public releases — robot system index 0.8.21, manufacturer directory 1.5.2, component catalog 0.4.14, citation registry 0.38.0 — and added explicit robot-system and manufacturer-directory evidence boundaries to it.
• Corrected a stale count boundary shipped in the robot dataset: the selected-entry boundary text still referenced the historical 207-row five-platform audit while the release carries 1,730 normalized records. The boundary rule is unchanged; the sentence is now count-agnostic so it cannot contradict the release it ships with.

Reason: restore the rule that every material data release ships with a citable, dated research entry, and keep evergreen reference content visibly re-verified, without converting any documentation growth into a claim of physical validation.

2026-08-13 · Site release 0.9.1 / Robot System Index terminal data layer

• Advanced Robot System Index to v0.8.21 by lifting five pinned ToddlerBot configurations from file inventories into bounded system maps. Forty-five selected rows now separate fixed-XML-filtered actuator-family assignments, recommended J4012 / Jetson Orin NX compute, an eight-channel communication-board boundary, dual BNO08X-family sensing, stereo fisheye paths, Distro Board v8 power and the external teleoperation leader boundary.
• Added twenty-eight ToddlerBot topology nodes and twenty-seven relations across four full-body variants and one teleoperation-leader configuration. The full-body maps expose CH9344-backed serial fan-out, Dynamixel Protocol 2.0 control, I²C addresses 0x4A/0x4B, two V4L2 paths and separate compute/actuator power branches; the leader map preserves its fourteen active positions and software ID offset from 16.
• Derived the family-level actuator counts only after deep-merging pinned default and per-configuration YAML and filtering against each fixed XML. Counts range from thirty active positions in the non-gripper variants to thirty-two in gripper variants; they are configuration assignments, not exact order codes, serialised inventory or an as-built inspection.
• Kept the J4012 / Jetson Orin NX 16GB record at recommendation-and-flashing-path level. No exact carrier SKU, module serial, storage image, mutable procurement-sheet snapshot, physical unit, connector map, powered result or complete rebuild package is claimed.
• Advanced Robot System Index to v0.8.20 by lifting reBot B601 DM v1.1, RS v1.0 and the separate DM supply assembly into three bounded system maps. Sixteen selected responsibility-party motor, CAN-interface, power-separation, harness and optional-supply rows join the fixed production files; DM and RS retain distinct 24 V versus 48 V, Damiao versus RobStride and optional-supply boundaries.
• Added seventeen reBot topology nodes and fifteen relations covering external host, generic CAN-to-USB boards, DM4310(V4), DM4340P(V4), RS00, RS06, XT30 2+2 harness/power separation and the mains-to-arm LRS-350-24 path. Optional camera mount filenames remain mounting evidence only and do not establish an installed D435i, Gemini 2 or any sensor.
• Advanced Robot System Index to v0.8.19 with a nine-node, ten-relation Berkeley Humanoid Lite system map. It separates on-board NUC, two generic USB hubs, four USB-CAN adapters, M6C12 and 5010 actuator banks, original BNO085/Arduino and recommended IM10A USB alternatives, protected 6S power distribution and low-voltage conversion while keeping the live documentation distinct from release v1.1.0.
• Advanced Robot System Index to v0.8.18 with an eight-node, seven-relation Sesame Option A system map. It separates programming host, alternative USB-C and battery sources, switched/buck 5 V rails, S2 Mini control, eight MG90S servos, OLED and frame/harness assembly while explicitly prohibiting the two power alternatives from being read as a simultaneous dual-source design.
• Advanced Robot System Index to v0.8.17 by converting the SO-101 follower from a seventeen-row parts list into a five-node, four-relation system map: external LeRobot host, generic MotorBus controller, six C001 follower servos, pinned printed assembly and external power boundary. The USB path, three-pin daisy chain and controller-fed power surface remain responsibility-party evidence with no RCI enumeration, bus capture, powered motion or load test.
• Released component catalog v0.4.14 with 103 exact records, 45 catalog-linked manufacturers and 945 claims. The new ROBOTIS XM430-W210-T record preserves the TTL-only identity, manufacturer stall-reference values, missing continuous-torque envelope and zero RCI physical tests.
• Advanced Robot System Index to v0.8.16 by adding a ten-node, nine-relation AlohaMini2 Host/Client topology. It separates Pi 5 compute, mobile-base control, follower-arm controllers, five camera paths, Client PC/leader arms, 12 V Host power and 5 V Client power while retaining the unexplained four-cable/two-controller count, battery-branch allocation and all zero-test boundaries.
• Released component catalog v0.4.13 with 102 exact records, 45 catalog-linked manufacturers and 930 claims. Four model- or full-configuration-bounded additions cover Nabtesco RV-20E ratio 57, RoboSense M2, Moxa EDS-2008-EL and Kollmorgen TBM2G-07608-A-NNAA-00, with eight new explicit unknowns and zero new physical tests.
• Advanced the manufacturer directory to v1.5.2: all fifteen sector intelligence dossiers now link to at least one exact-model evidence record. Exact-model linkage is not partnership, availability, compatibility or physical validation.
• Released component catalog v0.4.12 with 98 exact records, 41 catalog-linked manufacturers and 880 claims. Six order-code-bounded additions cover maxon 688690, Zivid ZVD2P-M60, AMD SM-K26-XCL2GC, Inventus Power 57658-101, PEAK-System IPEH-004022 and Beckhoff CX5130-0100, with twelve new explicit unknowns and zero new physical tests.
• Advanced the manufacturer directory to v1.5.1: 41 of 104 profiles now link to exact-model evidence and 63 remain source-screened. Exact-model linkage is not partnership, availability, compatibility or physical validation.
• Replaced the 107-word Robot/BOM framework with 31 versioned robot and configuration profiles spanning desktop arms, humanoids, mobile bases, quadrupeds and mobile manipulators.
• Published 1,730 normalized responsibility-party records from pinned BOM rows, hardware requirements and commit- or release-addressable engineering file identities. Shared TurtleBot3 source-table rows are counted once per independently bounded Burger or Waffle Pi profile, not as unique component identities.
• Added 21 explicit lifecycle, replacement, end-of-life and mutually exclusive configuration relations without silently merging versions or variants; SO-100 → SO-101 is recorded as a responsibility-party version transition, not a drop-in migration guarantee.
• Published platform JSON/CSV and line-level BOM CSV exports while keeping file-record quantities as not published.
• Published 20 row-level exact-model crosswalks. The latest Waffle Pi profile links the official two-unit XM430-W210-T row and one-unit OpenCR1.0 row without inferring exact Raspberry Pi, camera or LDS revisions.
• Advanced Robot System Index to v0.8.21 with twenty exact-component relations and one hundred and fifty-nine source-supported assembly/interface relations across twenty-eight configurations. ToddlerBot, Berkeley Humanoid Lite and reBot B601 expose bounded compute, serial/CAN, sensing, actuator and power paths without treating recommendations, mutable documentation, optional hardware or published tables as as-built configurations.
• Published a dedicated interface-topology CSV and embedded one hundred and sixty-eight topology nodes in the robot JSON. All one hundred and fifty-nine relations retain unresolved connector, wiring, firmware, voltage-domain, protection or load details and a physical-validation state of none.
• Released component catalog v0.4.11 with 92 exact records, 35 catalog-linked manufacturers and 805 claims. CSF-14-50-2UH, BFS-PGE-161S7C, Atom x7425E order code FJ8071504822030, DCM3623T50M31C2T00 and VN1630A add one order-code-bounded record in each hardware layer while preserving twelve explicit unknowns and zero RCI hardware tests.
• Advanced the manufacturer directory to v1.5.0. Fifteen sector intelligence dossiers remain available; Harmonic Drive, Teledyne FLIR, Intel, Vicor and Vector Informatik now combine dossier-level selection context with an exact-model evidence record. A dossier or exact record is not partnership, availability, compatibility or physical validation.
• Continued to reject the Solo12 bearing and reBot mount-filename substring false positives, and kept STS3215 voltage/ratio variants split pending exact identities.
• Preserved optional-installation, board-revision and unresolved physical-identity boundaries, together with zero complete independent rebuild packages and zero physical evidence paths until serialised as-built or laboratory evidence exists.

Reason: turn Robot/BOM into a source-addressable engineering index while preventing repository file counts or public parts lists from being misrepresented as an as-built manifest, successful rebuild or hardware validation.

2026-08-13 · Site release 0.9.0 / 100-manufacturer intelligence directory

• Expanded the public manufacturer intelligence directory from 26 to 100 profiles through four explicit source-screening batches of 19, 19, 18 and 18 additions.
• Reached a five-layer distribution of 30 motion-and-actuation, 25 perception-and-sensing, 20 compute-and-control, 15 power-and-energy and 10 interface-and-integration profiles.
• Kept the immutable component evidence catalog at 76 exact models from 26 manufacturers and 599 claims; the other 74 manufacturer profiles are marked source-screened with exact-model audits queued.
• Published machine-readable JSON and CSV directory endpoints, category-specific audit checklists and a public engagement register.
• Disclosed that RCI established direct technical fact-check contact with the ROBOTIS Technical Team and completed one bounded XH540-W270-R documentation exchange, while explicitly rejecting any inference of formal partnership, endorsement, sponsorship or independent validation.

Reason: broaden industry coverage without converting a company name, official website or private response into audited product evidence or a partnership claim.

2026-08-12 · Site release 0.8.14 / public-surface retirement

• Retired a self-referential visibility experiment from the public website while retaining its records in the internal operations archive.
• Removed its landing page, homepage and footer promotion, press fact, data downloads, machine-discovery entry and current citation record.
• Configured every retired dedicated URL to return 410 Gone with a no-index response header.

Reason: keep public attention on robot-component evidence, independent validation and correction-ready industry data.

2026-08-11 · Site release 0.8.13 / release-label integrity gate

• Moved the public site, catalog and source-directory release labels into one shared metadata source.
• Made the homepage, global footer and change-log header consume the package release automatically instead of repeating hand-maintained literals.
• Added a distribution check that rejects a build when the rendered public release labels drift from package.json.
• Left all frozen data releases and their integrity hashes unchanged.

Reason: prevent a deployable build from silently presenting a stale public site version.

2026-08-11 · Site release 0.8.12 / public release-label correction

• Corrected the homepage and global footer from the stale visible label 0.8.8 to the current public release.
• Updated the homepage evidence focus while retaining the existing public release identity.
• Restored direct change-log navigation for releases 0.7.3 through 0.8.12 without rewriting any historical dataset.

Reason: keep the human-visible release identity aligned with the deployed assets and machine-readable change history.

2026-08-11 · Site release 0.8.9 / RCI 045

• Published a reproducible delta between immutable source-directory v0.1.2 and v0.1.5 instead of rewriting the fixed RCI 038 baseline.
• Identified five added manufacturer-maintained release channels: Orbbec SDK v2, Stereolabs ZED SDK, Hesai LiDAR SDK 2.0, ROBOTIS DynamixelSDK and NVIDIA Isaac ROS Common.
• Recomputed the current boundary as 101 channels, 74 primary channels and 21 direct monitors across all five first-stage hardware layers.
• Recorded that all five additions remain ITEM_LICENSE sources and grant zero blanket right to reproduce release prose, manuals, binaries, firmware, models, containers, media or third-party files.
• Joined the 152-endpoint network observation to its five-record manual review while preserving zero source removals and zero rights-class changes caused by transport behavior.

Reason: keep the public industry-source map current, machine-readable and historically honest while separating responsibility, monitoring, network access, independent validation and republication rights.

2026-08-11 · Site release 0.8.8 / RCI 044

• Published a claim-provenance audit of AMD Kria AI SOM and Robotics Developer Platform launch surfaces.
• Traced four prominent metrics to a modeled agent sweep, a Bosch Rexroth controller result and a GMKtec proxy system configured to reflect X199 specifications.
• Recorded fourteen observations and twelve disclosure checks while preserving zero exact orderable models fixed on the page and zero RCI hardware runs.
• Released immutable JSON/CSV and citation-registry v0.29.0 without converting family maxima or proxy benchmarks into exact-module performance.

Reason: make launch-footnote provenance directly citable before broad product-family claims are repeated as measurements of an orderable robot main-compute module.

2026-08-11 · Site release 0.8.7 / RCI 043

• Published a fixed-paper review of an actuator-level CDPR policy across one simulated four-cable training setup, selected eight-cable transfers and one author-reported physical eight-motor trajectory.
• Separated the 100 Hz configuration and reported <0.04 m Figure 9 position-error norm from measured real-time behavior, repeatability, arbitrary-configuration validity and component recommendations.
• Recorded fourteen bounded observations, fourteen disclosure checks, two fixed paper digests and zero RCI simulation, controller, actuator or robot runs.
• Published immutable JSON/CSV and citation-registry v0.28.0 while preserving v0.27.0 and all earlier snapshots.

Reason: add an actuation-control research signal without converting one reported transfer trajectory into independent hardware validation or a universal control claim.

2026-08-11 · Site release 0.8.6 / RCI 042

• Published a fixed-filing review of STMicroelectronics' Q2 2026 6-K across power, sensing, embedded processing and silicon-carbide manufacturing signals.
• Separated P&D's 19.2% sequential revenue increase and two reported 200 mm SiC manufacturing starts from robot revenue, exact product demand, fab capacity, qualified yield and availability.
• Recorded twelve bounded observations, ten disclosure checks, one fixed SEC accession and zero RCI hardware or manufacturing tests.
• Published immutable JSON/CSV and citation-registry v0.27.0 while preserving v0.26.0 and all earlier snapshots.

Reason: convert a broad upstream semiconductor signal into bounded, machine-readable evidence without presenting segment movement as robot-component demand or facility ramp-up as validated output.

2026-08-11 · Site release 0.8.5 / first testing-partner outreach

• Sent three separately authorized, one-to-one enquiries to the official public routes for A*STAR ARTC / ROS-Industrial APAC, NTU Robotics Research Centre and A*STAR National Metrology Centre.
• Observed each message in Zoho Sent with a delivered transport indicator, while recording zero human replies, agreements, specimens, runs or results.
• Changed the first three public candidate states to contacted-awaiting-reply; the remaining two candidates stay uncontacted.
• Preserved the boundary that delivery is not partnership, endorsement, capability confirmation, pricing, facility access, accreditation evidence or independent replication.

Reason: keep the public program synchronized with actual outreach without converting mail transport evidence into a substantive external commitment.

2026-08-11 · Site release 0.8.4 / physical testing partner program

• Published four separate collaboration tracks for protocol review, hosted EV4 testing, externally operated EV5 replication and applicable EV6 accredited-laboratory work.
• Published eight independence rules covering loans, paid laboratory time, failed-run retention, conflicts, editorial control and named responsibility.
• Identified five candidate organizations in Singapore and the United States while explicitly recording zero contacts, agreements, specimens, runs or results.
• Connected the public program to a machine-readable JSON endpoint, the EV0–EV6 methodology, corrections channel and a bounded internal outreach ledger.

Reason: convert the hardware-testing gap into an executable collaboration contract without misrepresenting researched candidates as partners or planned tests as completed evidence.

2026-08-11 · Site release 0.8.3 / RCI 041

• Published an Ouster Q2 2026 SEC filing brief with twelve bounded observations, ten disclosure checks and explicit separation of a >17,000 combined shipment threshold from an exact lidar count or model-level market claim.
• Released immutable JSON/CSV and citation-registry v0.26.0 while preserving v0.25.0 and every earlier snapshot.
• Updated the public methodology to v1.1 with EV0–EV6 work levels that distinguish source checking, artifact integrity and public-data recomputation from RCI bench tests, external replication and accredited-laboratory work.
• Relabeled two prominent historical pages as EV2 artifact-integrity and EV3 public-data work, explicitly stating that RCI did not test the hardware.

2026-08-11 · Site release 0.8.2 / RCI 040

• Published a fixed-paper evidence brief for EsaacSim across five reported event-camera resolutions, three interpolation settings and two output modes.
• Preserved the paper's backend-only benchmark boundary: scene loading, rendering initialization, ROS 2 communication, logging and file I/O are excluded.
• Kept the 6.98–29.16 ms timing range and 51.4–399.7 MB allocated-memory range author-reported; RCI ran zero simulations, GPUs, physical cameras and robot workloads.
• Recorded absent real-camera comparison, unimplemented physics-grounded backend, unevaluated robot tasks and unresolved public-code availability.
• Published immutable JSON/CSV and citation-registry v0.25.0 while preserving v0.24.0 and all earlier snapshots.

Reason: turn one high-value perception signal from the internal source radar into bounded, machine-readable evidence without converting simulation settings into real-sensor or full-system performance claims.

2026-08-11 · Site release 0.7.9 / source directory 0.1.5

• Promoted the official ROBOTIS DynamixelSDK and NVIDIA Isaac ROS Common release feeds from the internal staging ledger into public source-directory v0.1.5.
• Expanded the directory from 99 to 101 channels, from 72 to 74 primary channels and from 19 to 21 direct feeds/APIs while preserving all nine source groups and five hardware layers.
• Rechecked both Atom endpoints and both repository-license endpoints at HTTP 200 and observed ten entries in each feed.
• Kept both sources at ITEM_LICENSE: Apache-2.0 repository code licensing does not grant blanket rights to firmware, manuals, binaries, examples, containers, models, documentation or third-party dependencies.
• Published an immutable promotion audit, source-directory JSON/CSV and citation-registry v0.24.0 without changing v0.1.4, v0.23.0 or any earlier snapshot.

Reason: add first-party actuation, perception and compute integration release signals without turning software-release metadata into exact-hardware compatibility or independent-performance evidence.

2026-08-11 · Site release 0.7.8 / public license boundary

• Licensed original RCI editorial text and original RCI database/data material under CC BY 4.0.
• Licensed repository code authored by RCI under MIT while preserving dependency and third-party code licenses.
• Explicitly excluded third-party documents, source text, figures, tables, photographs, logos, trademarks, standards material, datasets, code, CAD and attachments from RCI's grants.
• Published a human-readable license page and machine-readable license-policy JSON, and updated crawler policy without conflating crawl access with reuse rights.
• Advanced the citation registry to v0.23.0 while preserving v0.22.0 and all older immutable snapshots unchanged.

Reason: make legitimate reuse easier while keeping ownership, attribution and third-party rights unambiguous.

2026-08-11 · Site release 0.7.7 / source-network evidence refresh

• Ran the current source-directory v0.1.4 validator over 99 channels and 146 source, rights and monitor endpoints.
• Observed 126 reachable responses, 15 automation-restricted responses, zero explicit 404/410 failures and five other transport errors during the bounded run.
• Manually rechecked all five other errors: three endpoints returned HTTP 200, one retained an expired-certificate failure and one retained an access or automation failure.
• Did not bypass TLS verification, remove a source solely because an automated request failed, change a rights class or infer reliability, quality, uptime or reuse permission from reachability.
• Published the complete machine network ledger and a five-endpoint manual follow-up from the source directory and data pages.

Reason: keep the industry information-source directory operationally auditable while separating transport health, editorial value and reuse rights.

2026-08-11 · Site release 0.7.6 / AI crawler access policy

• Confirmed an allow-all crawler policy for search indexing, user-triggered AI retrieval, AI search and citation, and training-oriented crawling.
• Kept the existing wildcard robots rule, which already permits every user-agent, and added a machine-readable /ai-crawler-policy.json record.
• Separated technical crawl access from copyright and database licensing: access is not a reuse license, endorsement, warranty or permission to bypass controls.
• Preserved the existing item-level third-party rights boundary and left the RCI-owned dataset license undeclared until a separate public decision is approved.

Reason: make RCI easy for search and AI systems to retrieve while preventing crawler access from being misread as a blanket content or data grant.

2026-08-11 · Site release 0.7.5 / source directory 0.1.4

• Promoted the official Orbbec SDK v2, Stereolabs ZED SDK and Hesai LiDAR SDK 2.0 GitHub release feeds from the internal staging ledger into public source-directory v0.1.4.
• Expanded the public directory from 96 to 99 channels, from 69 to 72 primary channels and from 16 to 19 direct feeds/APIs while preserving all nine source groups and five first-stage hardware layers.
• Verified all three Atom endpoints and license pages at HTTP 200; observed ten, ten and five feed entries respectively during the bounded review.
• Classified all three as item-license sources: repository code licenses do not provide blanket rights to release prose, binaries, models, firmware, installers, calibration files, documentation, media or third-party components.
• Published a machine-readable promotion audit and froze source-directory v0.1.4 JSON/CSV; v0.1.3, v0.1.2 and every earlier source snapshot remain unchanged.
• Advanced the citation registry to v0.22.0 with the same 29 citable assets, updating only the current source-directory record and preserving v0.21.0 and all older registries.

Reason: bring three first-party perception and integration engineering channels into the public monitoring map without conflating source discovery, code licensing, artifact rights, reliability or independent validation.

2026-08-11 · Site release 0.7.4 / Orbbec SDK 2.9.3 evidence audit

• Published RCI 039, an independent release-integrity, provenance, public-CI, device-scope and rights-boundary audit of Orbbec SDK v2.9.3.
• Resolved the annotated unsigned tag, unsigned target commit and complete 3,369-entry recursive tree; separated the July publication timestamp from the August release edit timestamp.
• Downloaded twelve uploaded assets totaling 649,731,276 bytes and matched 12/12 publisher sizes and SHA-256 digests without executing, installing or redistributing them.
• Compared both generated source archives with the fixed checkout: each exposes 2,995 file/symlink paths, identical included bytes and the same two export-ignored libusb dotfiles.
• Recorded a 252,582,062-byte model.sm4, a Jetson/Linux ARM64 and device-license requirement, and zero RCI model, camera, LiDAR, firmware or robot tests.
• Separated thirteen first-party test files and seventeen benchmark-tool files from zero public check runs, workflow runs at the target commit or legacy statuses.
• Exposed conflicting fixed rights wording: the extension README says MIT while the root license, extension license and EULA impose separate proprietary-use restrictions; RCI grants no blanket reuse conclusion.
• Published twenty bounded observations, twenty disclosure checks and immutable JSON/CSV; advanced the citation registry to v0.21.0 with 29 assets while preserving v0.20.0 and all earlier releases.

Reason: deepen first-stage vision and LiDAR coverage with an auditable responsibility-source release while keeping byte integrity, build provenance, rights, public CI, device support and physical performance as distinct evidence layers.

2026-08-10 · Site release 0.7.3 / public press facts

• Published an English press and media kit with approved one-line, short and extended descriptions for consistent reporting and AI retrieval.
• Added a machine-readable public facts file whose current counts are generated from the catalog, source directory, research library and citation registry.
• Published the five first-stage hardware layers, evidence-state counts, corporate authorship, public editorial identity and RCI-domain media contacts in one canonical record.
• Made seven non-claims explicit: RCI is not a certification body, complete market census, paid ranking, endorsement, universal compatibility proof or blanket republication license.

Reason: give media, manufacturers, researchers and AI systems a canonical description of RCI without overstating the database's coverage, verification or authority.

2026-08-10 · Site release 0.7.2 / structured evidence submissions

• Published a seven-route evidence-submission protocol for corrections, new components, source recommendations, lifecycle changes, independent tests, compatibility evidence and rights clarifications.
• Added a machine-readable empty JSON template spanning all five first-stage layers, exact identity, claim location, provenance, test reproducibility, contributor conflicts, rights and contact consent.
• Made the review boundary explicit: submissions are never auto-published, vendor material does not become independent evidence, and payment or access cannot purchase ranking or editorial approval.
• Recorded the user-approved public identity RCI Editorial Team and fact-check From address [email protected]; authenticated outbound sending remains unconfigured and external sends remain zero.

Reason: let manufacturers, researchers and integrators contribute precise evidence without weakening exact-model identity, rights controls, independent review or historical traceability.

2026-08-10 · Site release 0.7.1 / source directory 0.1.3

• Published source-directory maintenance release v0.1.3 with the same 96-channel, 69-primary-channel and 16-direct-monitor boundary.
• Corrected the official IROS, Humanoids, CNIPA search and ST battery-management entry points without rewriting immutable v0.1.2 or RCI 038.
• Classified HTTP 412 as an automation restriction, retained EPO Espacenet as manual-only with OPS as the automation path, and recorded the degraded GGII HTTPS/HTTP endpoint for weekly recheck instead of deleting the lead source.
• Added a machine-readable manual-review ledger and improved the network auditor with detailed error cause codes and optional JSON output.
• Advanced the citation registry to immutable v0.20.0 with the same 28 assets, now pointing the source-directory record to v0.1.3 while preserving v0.19.0 and every earlier release.

Reason: keep the source directory operationally current while preserving immutable evidence boundaries and separating endpoint health from editorial rights, quality and source responsibility.

2026-08-10 · Site release 0.7.0 / catalog 0.4.7

• Published RCI 038, a quantitative robot-component industry source and reuse atlas derived from immutable source-directory v0.1.2.
• Mapped 96 channels into nine source groups, 69 primary channels, 16 direct feeds/APIs, five overlapping hardware layers and five conservative reuse classes.
• Made the republication boundary explicit: 8 open/public-record defaults, 21 item-license channels, 40 facts-plus-commentary channels, 13 permission-required channels and 14 lead-only channels.
• Recorded a bounded 137-endpoint network audit with 112 reachable, 13 automation-restricted, zero explicitly broken and 12 other-error observations; reachability is not a rights or quality conclusion.
• Published immutable JSON/CSV and advanced the citation registry to immutable v0.19.0 with 28 assets while preserving v0.18.0 and all earlier releases unchanged.

Reason: answer where frontier robot-component information originates and how it may be used, while keeping discovery, responsibility, evidence strength and republication rights as separate editorial decisions.

2026-08-10 · Site release 0.6.9 / catalog 0.4.7

• Published RCI 037, a cross-release evidence matrix spanning vision, LiDAR, main compute and battery management.
• Joined six previously published immutable upstream release audits without replacing their project-specific definitions or boundaries.
• Recorded 1,992 heterogeneous test-surface path observations, 32 successful public check conclusions and 12 downloaded publisher-digest matches; the path total is not a test count, pass count, coverage measure or ranking.
• Found zero releases with a current retained complete physical-validation bundle; RCI performed zero physical-device tests and zero robot-system tests.
• Published immutable JSON/CSV and advanced the citation registry to immutable v0.18.0 with 27 assets while preserving v0.17.0 and all earlier releases unchanged.

Reason: make first-stage non-joint coverage explicit and queryable while preserving the distinction between source structure, public CI conclusions, byte identity, retained device results and robot validation.

2026-08-10 · Site release 0.6.8 / catalog 0.4.7

• Published RCI 036, an independent release-integrity, uploaded-asset, public-CI, compatibility, firmware and rights audit of RealSense SDK 2.0 v2.58.3.
• Preserved the human-facing beta release name beside GitHub's prerelease=false field instead of inventing a readiness interpretation.
• Resolved the lightweight tag, GitHub-verified merge commit and complete 4,542-entry recursive tree containing 3,840 blobs, 702 directories and two symlinks.
• Downloaded both GitHub-generated source archives, computed audit-date SHA-256 identifiers and confirmed zero path, regular-file-byte or symlink-target differences versus the fixed checkout.
• Downloaded all eight uploaded Windows executables—229,877,008 bytes total—and matched 8/8 publisher sizes and 8/8 publisher SHA-256 digests without executing, installing or unpacking them.
• Recorded 28 successful public check runs across three successful workflow runs; two static-analysis artifact records are expired and the build workflows expose no artifacts through the current API.
• Separated 380 unit-test-path files, benchmark source, five platform families and six camera-matrix rows from missing exact-device, firmware-byte, host, protocol and raw-result evidence.
• Preserved the release's compilable-but-not-validated and L515-specific wording; RCI ran zero camera, firmware or robot tests.
• Published twenty bounded observations, twenty disclosure checks and immutable JSON/CSV; advanced the citation registry to immutable v0.17.0 with 26 assets while preserving v0.16.0 and all earlier releases unchanged.

Reason: deepen first-stage vision coverage with a byte-identifiable responsibility-source release while keeping CI success, compatibility tables, feature claims, firmware identity, physical camera results and robot validation as distinct evidence layers.

2026-08-10 · Site release 0.6.7 / catalog 0.4.7

• Published RCI 035, an independent release-integrity, Mid-360S support-surface, sample, test, firmware and rights audit of Livox SDK2 v1.3.1.
• Resolved the official non-prerelease release, lightweight tag, unsigned target commit and complete 258-entry recursive tree.
• Downloaded both GitHub-generated source archives, computed audit-date SHA-256 identifiers and confirmed 221 of 221 files in each archive against the fixed Git blob identities.
• Recorded the API-header version 1.3.1 alongside an unexplained core CMake 0.0.2 variable without assigning undocumented ABI or release meaning.
• Separated six CMake sample targets, 24 JSON configurations and six Mid-360S configurations from zero explicit test, benchmark, result or public CI paths.
• Preserved the fixed README's three named samples alongside the CMake tree's six targets as a documentation-scope question rather than silently reconciling them.
• Recorded explicit Mid-360S device handling, ports, ESC mode and PPS synchronization calls without converting source presence into exact firmware compatibility, timing performance or physical validation.
• Published twenty bounded observations, twenty disclosure checks and immutable JSON/CSV; RCI compiled no software, downloaded no firmware and accessed no LiDAR or robot.
• Advanced the citation registry to immutable v0.16.0 with 25 assets while preserving v0.15.0 and all earlier releases unchanged.

Reason: extend first-stage LiDAR coverage with a fixed responsibility-source software asset while keeping source support, examples, firmware, public tests, sensor results and robot integration as distinct evidence layers.

2026-08-10 · Site release 0.6.6 / catalog 0.4.7

• Published RCI 034, an independent release-integrity, test-surface, hardware-link and rights audit of foxBMS 2 v1.11.0.
• Resolved the official non-prerelease release, annotated tag, target commit and complete 3,573-entry recursive tree; GitHub reports both tag and commit unsigned.
• Downloaded both GitHub-generated source archives, computed audit-date SHA-256 identifiers and confirmed identical normalized sets of 2,822 files and 751 directories.
• Mechanically counted 93 top-level changelog bullets—34 Added, 38 Changed, zero Deprecated, four Removed, fifteen Fixed and two dependency headings—without treating issuer statements as test outcomes.
• Separated 905 public test-tree files from release execution evidence: the fixed tree contains no published HIL path or matched coverage/JUnit result output, and the tagged commit exposes no public GitHub check run.
• Recorded ten version-labelled hardware links and the OSHWA DE000128 open-hardware record without converting link availability or disclosure certification into an exact software-hardware, safety or robot validation matrix.
• Published twenty bounded observations, twenty disclosure checks and immutable JSON/CSV; RCI executed no software and accessed no BMS, battery or robot system.
• Advanced the citation registry to immutable v0.15.0 with 24 assets while preserving v0.14.0 and all earlier releases unchanged.

Reason: extend first-stage robot power and BMS coverage with an inspectable responsibility-source release asset while keeping public code, test structure, hardware links and open-hardware certification separate from physical validation.

2026-08-10 · Site release 0.6.5 / catalog 0.4.7

• Published RCI 033, an independent release-integrity and disclosure audit of AMD Ryzen AI Software v1.8.0.
• Resolved the lightweight tag, GitHub-verified target commit, 669-entry recursive source tree, current AMD 1.8 documentation and fixed root MIT license.
• Downloaded both GitHub-generated source archives, computed audit-date SHA-256 identifiers and confirmed identical normalized 669-entry path sets; neither file is an AMD runtime installer or a publisher-hash match.
• Recorded eight Git LFS pointer blobs declaring 404,905,134 bytes of payload outside the generated archives; RCI did not fetch or hash those objects.
• Separated three EULA-gated runtime package names, processor-family and model-class support tables, eight LLM/VLM additions and one Whisper RTF statement from missing exact-host protocols, raw outputs and robot workloads.
• Published twenty bounded observations, twenty disclosure checks and immutable JSON/CSV; RCI accepted no download EULA, executed no software and accessed no Ryzen AI device.
• Advanced the citation registry to immutable v0.14.0 with 23 assets while preserving v0.13.0 and all earlier releases unchanged.

Reason: extend RCI's compute-and-control coverage with a release-bound software evidence asset without converting compatibility tables, source code or an under-specified issuer performance statement into robot-computer validation.

2026-08-10 · Site release 0.6.4 / catalog 0.4.7

• Published RCI 032, an independent release-integrity and disclosure audit of Ouster SDK v1.0.0.
• Resolved the annotated tag, verified target commit, complete recursive tree, versioned changelog, root and binary-license notices and PyPI package matrix.
• Downloaded both GitHub-generated source archives and confirmed identical normalized 1,735-entry path sets; their RCI hashes remain audit-date identifiers because GitHub publishes no digest for generated downloads.
• Recorded 35 wheels plus one PyPI sdist across CPython 3.8–3.14 and five platform targets; independently matched the 37,563,121-byte sdist SHA-256 while downloading and running zero wheels.
• Counted 138 v1.0.0 changelog bullets, including 54 explicit breaking, 24 future-breaking and 15 bugfix records, without treating those labels as test outcomes.
• Published twenty observations and twenty disclosure checks separating public test source and one bag fixture from zero separately released sensor benchmark result assets.
• Advanced the citation registry to immutable v0.13.0 with twenty-two assets while preserving v0.12.0 and all earlier releases unchanged.

Reason: turn a responsibility-source LiDAR software signal into a citable evidence asset without converting a package matrix, changelog or test source into sensor accuracy, compatibility or performance proof.

2026-08-10 · Site release 0.6.3 / catalog 0.4.7

• Published RCI 031, an independent release-integrity and disclosure audit of Luxonis DepthAI Core v3.8.0.
• Downloaded all three separately uploaded GitHub release assets and independently matched every publisher byte count and SHA-256 digest.
• Confirmed the two source archives contain the same 2,488 normalized paths, the Windows archive contains 641 entries and the source tree declares version 3.8.0 under the MIT License.
• Recorded five publisher-indexed PyPI wheels across macOS, manylinux and Windows, while explicitly noting that RCI did not download, import or execute those wheels.
• Published twenty condition-bounded observations and twenty disclosure checks separating new depth, ToF, health, timestamp, calibration and GPU-backend statements from missing device identities, protocols, raw logs and numerical results.
• Kept the issuer-stated Luxonis OS compatibility matrix separate from exact-device integration evidence; RCI performed zero physical-device tests and zero software-test executions.
• Advanced the citation registry to immutable v0.12.0 with twenty-one assets while preserving v0.11.0 and all earlier releases unchanged.

Reason: turn a responsibility-source software-release signal into a citable non-joint evidence asset without converting package integrity or feature notes into camera-performance or robot-integration proof.

2026-08-10 · Site release 0.6.2 / catalog 0.4.7

• Expanded the source directory to immutable v0.1.2 with 96 channels and 69 first-party channels.
• Added bounded official release feeds for Ouster SDK, Livox SDK2, Luxonis DepthAI Core, AMD Ryzen AI Software and foxBMS 2, covering LiDAR, vision, edge compute and battery management.
• Expanded the internal metadata-only radar from seven to twelve enabled feeds/APIs while preserving three watchlist-gated adapters as disabled.
• Verified all five new Atom endpoints with HTTP 200 and kept version-only release titles through source-scoped relevance rules; no feed body, release prose, binary or media enters the radar ledger.
• Advanced the citation registry to immutable v0.11.0 with the same twenty asset classes, now pointing the source-directory record to v0.1.2; v0.10.0 and all earlier releases remain unchanged.
• Corrected llms.txt from its stale v0.6.0／RCI 029 summary to the current release and RCI 030 evidence boundary.

Reason: make first-stage non-joint coverage operational at the responsibility-source release level instead of depending mainly on broad research indexes and media leads.

2026-08-10 · Site release 0.6.1 / catalog 0.4.7

• Published RCI 030, a source-level and reproducibility audit of Research Square preprint rs-10623292 v1 and its S-HASEL expansion actuator.
• Verified the item-level CC BY 4.0 statement on the official record and covered PDF, then hashed and visually checked all sixteen PDF pages.
• Published thirty condition-bounded observations, twenty-five disclosure checks and seven application-stage boundaries spanning geometry, materials, high-voltage drive, self-sensing and the proposed biopsy use.
• Preserved BOPP with 1.2 mL fill as a within-study selection across two films and three fill volumes—not a universal HASEL optimum.
• Preserved four NRMSE values as author-reported single-actuator sensing results; none is relabelled as absolute, closed-loop, MRI or biopsy-targeting accuracy.
• Recorded zero supplementary, raw-data, code, CAD, BOM and schematic files on the official public item.
• Separated tested actuator evidence from future closed-loop control, an untested three-stack robot, untested needle puncture and zero MRI-environment tests.
• Advanced the citation registry to immutable v0.10.0 with twenty assets while preserving v0.9.0 and all earlier releases unchanged.

Reason: turn a timely soft-actuation lead into a citable engineering evidence asset without converting an application concept, preliminary sensing metric or open license into proof of a working clinical robot.

2026-08-10 · Site release 0.6.0 / catalog 0.4.7

• Published RCI 029, a public-package and reproducibility audit of arXiv:2608.06488 v1 and its six-unit RAVEN II parallel motor-cable force-actuation system.
• Hashed and visually checked the complete three-page PDF, then enumerated all ten official TeX source-bundle files: five figures and zero dataset, code, CAD or BOM files.
• Recorded fifteen condition-bounded paper observations and eighteen disclosure checks covering hardware identity, control, calibration, timing, force reference, trial definition, metrics and released artifacts.
• Preserved the author-reported 0.80 N, 0.79 N and 0.44 N X/Y/Z mean absolute errors as three axis-wise results; they are not relabelled as sample-wise, vector or independently reproduced errors.
• Preserved 320 Hz as a load-cell-driver measurement-rate statement instead of converting it into closed-loop bandwidth, synchronized six-channel rate or end-to-end latency.
• Recorded five Figure 3 electronics labels as lead-level facts without creating thin component records or claiming a complete BOM.
• Advanced the citation registry to immutable v0.9.0 with nineteen assets while preserving v0.8.0 and all earlier releases unchanged.

Reason: turn a fresh actuation-system lead into an inspectable frontier evidence asset while keeping preliminary author results, public-package completeness and independent verification as separate questions.

2026-08-10 · Site release 0.5.9 / catalog 0.4.7

• Published RCI 028, an exact-model public-evidence audit of the STMicroelectronics L99BM114 battery-monitoring and balancing IC.
• Hashed and reviewed the 159-page DS14722 Rev 1 datasheet, publishing fourteen conditioned specification observations without mirroring its tables or figures.
• Audited two ST-authored transient plots—250 A discharge and approximately −135 A charge, each 450 seconds—and recorded zero public raw traces, instrument models, calibration records, sample rates, filter configurations or numeric transient-error metrics.
• Preserved the datasheet's conversion, filtering and synchronization timing boundary instead of repeating an unqualified zero-latency statement.
• Added L99BM114-TR to catalog v0.4.7, bringing the live catalog to 76 products, 26 manufacturers and 599 claims; its sixteen specification claims are M and three missing public-evidence fields are U.
• Expanded the source directory to immutable v0.1.1 with 91 channels and 64 first-party channels, adding ST battery-management engineering, Weston Robot documentation/news and Avnet Silica robotics.
• Advanced the citation registry to immutable v0.8.0 with eighteen assets while preserving v0.7.0 and all earlier releases unchanged.

Reason: add a non-joint power-system component and a current humanoid-robot evidence lead without turning manufacturer plots, a rights-ambiguous mirror or incomplete bench disclosure into independent validation.

2026-08-10 · Site release 0.5.8 / catalog 0.4.6

• Published RCI 027, an independent integrity, version-history, BOM and mesh-topology audit of the LT-PAM open-hardware deposit.
• Matched all nine current Zenodo files to their published byte counts and MD5 values, recorded file SHA-256 hashes and produced a deterministic manifest hash.
• Resolved three public version records and found that all six CAD/STL design files are byte-identical across them; the middle record adds a BOM and the current record replaces the assembly video.
• Parsed three binary STL files and found one component, zero degenerate triangles, zero boundary edges and zero non-manifold edges in each, while retaining the unresolved coordinate-unit and physical-performance boundaries.
• Recomputed a bounded 150 mm BOM subtotal of ¥583.27 before cement, prints and tools, without turning the source's approximately USD 5 statement into an exact verified cost.
• Recorded zero raw mechanical-test files, zero standalone license files and zero neutral CAD exchange files; all force, contraction, rupture, cost and durability values remain source-reported.
• Advanced the citation registry to immutable v0.7.0 with seventeen assets while preserving v0.6.0 and all earlier releases unchanged.

Reason: make an open actuator design citable without converting file availability, mesh closure or author-reported performance into independent pressure-safety or durability verification.

2026-08-10 · Site release 0.5.7 / catalog 0.4.6

• Published RCI 026, an independent structural audit of the open KUKA KR3 R540 motor-current payload dataset.
• Matched the versioned Zenodo archive MD5, recorded a local SHA-256 and parsed all 128 CSV files without redistributing the source ZIP or raw rows.
• Independently counted 685,400 numeric rows with zero blank, non-finite or malformed rows and reproduced all eight D1–D8 totals printed in the paper’s Table 2.
• Separated the paper’s reported same-unit, mixed-unit and direct R1-to-R2 MLP results from RCI’s own structural checks; RCI did not claim to reproduce the model metrics.
• Published an unresolved timing conflict: the paper and README state 12 ms / approximately 83.3 Hz, while every deposited adjacent Sample interval advances by four units and the README calls the field milliseconds.
• Advanced the citation registry to immutable v0.6.0 with sixteen assets while preserving v0.5.0 and all earlier releases unchanged.

Reason: turn an open research lead into an independently inspectable system-evidence asset without converting source-reported accuracy into RCI verification or silently correcting an unresolved time-base conflict.

2026-08-10 · Site release 0.5.6 / catalog 0.4.6

• Added a source-linked D455 software-support claim from the official RealSense SDK 2.0 beta v2.58.3 release table: firmware 5.17.3.10 or later and librealsense v2.58.3 or later.
• Preserved the official source difference that the living D400 firmware table separately associates firmware 5.17.3.10 with SDK 2.58.1.
• Expanded RCI 024 to immutable dataset v0.2.0 with eleven official source surfaces and an explicit JetPack 5.0.2 / 6.0 / 6.1 / 6.2 / 7.0 platform boundary.
• Kept the integration conclusion unchanged: no selected source validates an exact D455 × J4012 carrier image, kernel, stream workload, timing result or long-run stability result.
• Advanced the catalog to immutable v0.4.6 with 580 claims and the citation registry to immutable v0.5.0 with fifteen assets; all earlier releases remain unchanged.

Reason: make a current software-compatibility floor citable without converting platform-level support tables into an exact robot-stack validation.

2026-08-10 · Site release 0.5.5 / catalog 0.4.5

• Published RCI 025, an independent recomputation of two public 4.3-hour unloaded robot battery and joint-motion traces.
• Verified the 185,123,569-byte RoboBATT 10 Hz archive, its upstream SHA-256 and all 138 internally listed file hashes; reran the deposited validator separately from the RCI analysis.
• Recomputed 310,697 clean BMS rows on strict 100 ms grids, observed 0.1 A current and 0.5 V voltage reporting increments, and kept 10 Hz controller cadence separate from internal BMS sensing bandwidth.
• Found that 97.28% and 97.39% of adjacent clean rows retained the same two stored pack-current values; this is reported as value repetition, not proof of stalled BMS communication.
• Found a condition-B high-minus-low activity current difference of 1.786 A (28.0%) and a within-minute median difference of 1.365 A; condition A did not preserve the same within-minute relationship.
• Published immutable aggregate JSON and CSV with source attribution, CC BY 4.0 notice, formulas, threshold sensitivity and explicit unloaded / uncalibrated / non-causal boundaries.
• Advanced the citation registry to immutable v0.4.0 with fifteen assets while preserving v0.3.0, v0.2.0 and v0.1.0 unchanged.

Reason: turn a high-value open telemetry lead into a citable original analysis without converting dense rows into false measurement resolution or a robot battery performance claim.

2026-08-10 · Site release 0.5.4 / catalog 0.4.5

• Published RCI 024, an exact-model interface audit spanning the RealSense D455, Livox MID-360, Seeed reComputer Industrial J4012 and RRC2054-2 smart battery.
• Separated five integration paths into thirty connector, power, data, timing, software and system-validation observations.
• Recorded two documented data-interface matches, eight conditional findings, five required adapters, two non-direct paths, nine unresolved checks and four not-applicable checks.
• Kept the documentary boundary explicit: zero observations and zero paths were independently tested, and zero complete compatibility packages were claimed.
• Published immutable v0.1.0 JSON and CSV with exact source locations, required actions and a six-step validation package for turning the stack hypothesis into test evidence.
• Advanced the citation registry to immutable v0.3.0 with fourteen assets while preserving v0.2.0 and v0.1.0 unchanged.

Reason: make visual, LiDAR, main-compute and battery integration a first-phase evidence product, not four unrelated specification pages or a falsely confident compatibility claim.

2026-08-10 · Site release 0.5.3 / catalog 0.4.5

• Published RCI 023 / Data Brief 002, a deterministic claim-level delta between immutable catalog v0.4.0 and v0.4.5.
• Recorded 35 added claim IDs, 43 modified existing records, zero removed IDs and 25 touched products while the cohort remained fixed at 75 products and 25 manufacturers.
• Separated five new independent records from three manufacturer-to-conflict transitions and one unknown-to-manufacturer transition; no manufacturer claim moved directly to independent verification.
• Published immutable JSON and CSV with all 78 change rows, release-step counts, evidence transitions, category depth and explicit interpretation boundaries.
• Kept publication, outbound validation and search indexing as separate funnels; accepted submissions are not reported as indexed pages.
• Advanced the citation registry to immutable v0.2.0 with thirteen assets while preserving the twelve-record v0.1.0 files unchanged.

Reason: make RCI’s own evidence evolution auditable and citable without treating deeper records as market growth, external validation, search discovery or AI adoption.

2026-08-10 · Site release 0.5.2 / catalog 0.4.5

• Published a central citation page covering twelve versioned RCI data, research, protocol and benchmark assets.
• Added stable and immutable v0.1.0 citation registries in JSON and BibTeX, with organization authorship, canonical landing pages and exact release-file URLs.
• Declared the current identifier boundary: canonical HTTPS URLs and immutable versioned downloads are used; no DOI has been assigned.
• Declared the current rights boundary without inventing an open-data license: citation is encouraged, no blanket reuse license is granted and third-party rights remain separate.
• Added citation guidance to llms.txt, the data index and RCI-BLM-1.
• Kept catalog v0.4.5 and all prior catalog and research releases unchanged.

Reason: give researchers, procurement teams and AI systems one version-aware route for citing RCI without losing evidence boundaries or implying rights and identifiers that have not been issued.

2026-08-10 · Site release 0.5.1 / catalog 0.4.5

• Published RCI 022, a cross-layer audit of nine numeric specification differences across seven exact robot-component products and all five RCI hardware layers.
• Separated two Ouster OS1 Rev7-to-Rev8 field changes from six current cross-source conflicts and one same-page conflict.
• Marked only the two Ouster fields as product-generation changes because Ouster publishes an explicit Rev8 boundary; the other seven remain open documentation conflicts.
• Preserved source-bound values for Orbbec Gemini 335, CubeMars AK70-10, Hesai QT128, Advantech MIC-713-OX, RRC2054-2 and Kvaser USBcan Pro 2xHS v2 without silently choosing a winner.
• Added immutable JSON and record-level CSV with revision flags, normalized differences, procurement consequences, required actions and fifteen exact responsibility-party source surfaces.
• Kept catalog v0.4.5 and all prior catalog and research releases unchanged.

Reason: make version drift citable while preserving the distinction between a documented product transition, an unresolved current source conflict and an RCI editorial correction.

2026-08-10 · Site release 0.5.0 / catalog 0.4.5

• Published RCI 021, a twelve-field audit of public BOM evidence for SO-101, Berkeley Humanoid Lite, TurtleBot3 Burger, ODRI Solo12 and Poppy Humanoid.
• Counted 207 selected rows or list items while explicitly excluding that total from unique-part, cost, completeness and complexity comparisons.
• Found four commit-addressable selected BOM surfaces, one native spreadsheet and zero complete revision-to-as-built independent rebuild packages.
• Preserved Berkeley’s live-sheet/release separation, TurtleBot3’s Pi/LDS generation boundary, Solo12’s nested v1/v1.1 structure and SO-101’s current-main versus release-tag boundary.
• Recorded the Poppy responsibility-source difference between the README’s 25 required DYNAMIXEL actuators and the selected BOM quantities totaling 20 for head option A.
• Added immutable JSON and record-level CSV containing only RCI audit metadata, counts, source mapping and findings—not copied third-party BOM rows.

Reason: make open robot hardware more citable without converting a public parts list into unsupported claims about completeness, current availability or a verified physical build.

2026-08-10 · Site release 0.4.9 / catalog 0.4.5

• Published RCI 020 and the public RCI-BLM-1 v0.1.0 torque–angle hysteresis protocol for reducers and integrated robot joint actuators.
• Separated reducer-only mechanical lock, integrated unpowered mechanical lock and integrated servo-hold into three non-interchangeable test boundaries.
• Defined a baseline −10% to +10% rated-output-torque triangle, three preconditioning cycles, five recorded cycles, three output positions, 20 ±2°C baseline and at least 100 Hz synchronized acquisition.
• Published explicit H0 and LM4 branch-gap equations plus cycle and position variation metrics; no vendor value is treated as equivalent without an equivalent method.
• Added a 40-column raw-sample template, 30-column summary template, uncertainty threshold, unit-versus-model claim rule and immutable machine-readable protocol JSON.
• Labeled the method public-draft and lab-validation-required; RCI has not yet published physical actuator results or claimed ISO conformity.

Reason: close the exact method gap identified by RCI 016 while preserving the boundary between a reproducible specification, a validated inter-laboratory method and an actual independent measurement.

2026-08-10 · Site release 0.4.8 / catalog 0.4.5

• Published RCI 019, an eleven-record battery compatibility audit that separates seven smart packs, one non-communicating pack family and three bare 21700 cells.
• Added a ten-field drop-in evidence matrix covering product identity, voltage, continuous and pulse current, connector contacts, communication, charging, temperature, mechanics and transport documents.
• Found that seven of eight packs name SMBus, four publish a versioned communication data boundary, six publish exact connector/contact functions and zero publish a time-qualified pulse-current limit.
• Marked zero complete drop-in evidence packages without converting missing public evidence into a claim that a product is incompatible or unsafe.
• Kept Molicel P42A’s independent characterization at the bare-cell boundary and did not use it to validate a robot pack.
• Added immutable JSON and record-level CSV; catalog v0.4.5 and all earlier releases remain unchanged.

Reason: deepen the first-stage power layer with a citable procurement dataset that distinguishes electrical similarity from evidence-backed interchangeability.

2026-08-10 · Site release 0.4.7 / catalog 0.4.5

• Published RCI 018, a hardware-boundary audit of eight production modules, two development platforms and seven complete robot edge-compute systems.
• Separated nine numeric TOPS statements into sparse INT8, sparse/dense INT8 and condition-underspecified boundaries; only three state both precision and sparsity.
• Separated module power modes, approximate module consumption and input voltage; none of the seven complete systems publishes a selected workload-qualified whole-system power result.
• Audited versioned software prerequisites, exact hardware availability and product-specific software support end dates; five exact products publish hardware dates and zero publish software end dates.
• Preserved Qualcomm’s underlying-SoC longevity and Toradex’s general/revision evidence as partial rather than converting them into exact platform commitments.
• Added immutable JSON and record-level CSV; catalog v0.4.5 and all earlier release files remain unchanged.

Reason: deepen the first-stage main-compute layer with a citable dataset that keeps accelerator, power, software and lifecycle claims on the hardware boundary they actually describe.

2026-08-10 · Site release 0.4.6 / catalog 0.4.5

• Published RCI 017, a condition-preserving audit of 25 range observations and 16 selected error or precision values across eleven robot depth-camera records.
• Added immutable JSON and observation-level CSV with five metric boundaries and thirteen reproducibility-disclosure fields.
• Found zero complete public reproduction packages; only two records publish ambient light, test temperature, warm-up and aggregation beside the selected value.
• Preserved the Gemini 335 current-source 0.26 m / 0.3 m recommended-range difference as a formal C state.
• Added lens-qualified ZED X and ZED X Mini V1.2 range and depth-accuracy claims while keeping metric-definition and host-software limitations visible.
• Reclassified the Gemini 2 and Gemini 2 XL selected rows as plane-fit RMS claims labeled “Depth Accuracy,” rather than silently treating them as absolute Z-error.
• Released immutable component catalog v0.4.5 with 579 claims; all earlier catalog and research files remain unchanged.

Reason: deepen the first-stage visual-system layer with a citable dataset that preserves metric, lens, mode and reproducibility boundaries.

2026-08-10 · Site release 0.4.5 / catalog 0.4.4

• Published RCI 016, an 18-record audit separating numeric backlash, categorical no-backlash wording, qualitative-only language, accuracy, lost motion, transmission error and repeatability.
• Added immutable JSON and record-level CSV with 15 explicit backlash statements, 13 numeric rows, two categorical rows, three qualitative-only rows and seven method-disclosure flags.
• Found zero complete public backlash methods and zero explicit lost-motion, transmission-error or repeatability values in the selected sources; no cross-model ranking was produced.
• Expanded current X8-32 evidence from its embedded technical-parameter image while preserving the earlier RCI015 text-boundary dataset unchanged.
• Marked the current AK70-10 621 g / 521 g same-page conflict as C and added explicit U backlash states for three qualitative CubeMars records.
• Released immutable component catalog v0.4.4 with 577 claims; all earlier catalog and research files remain unchanged.

Reason: make motion-error terminology and missing measurement procedures machine-readable without manufacturing a precision leaderboard.

2026-08-10 · Site release 0.4.4 / catalog 0.4.3

• Published RCI 015, an 18-record audit separating continuous, rated, manufacturer-calculated continuous, stall-only and peak-only torque evidence.
• Added immutable JSON and record-level CSV for the actuator torque duty-boundary research dataset v0.1.0.
• Measured twelve within-model peak-to-continuous-or-rated ratios from 2.00× to 3.00× without converting them into cross-brand rankings.
• Audited rating duration, cooling method and exact rating temperature across thirteen published continuous-or-rated rows; none of the selected rows disclosed those fields.
• Added explicit U-state continuous-torque claims for XH540-W270-R and MYACTUATOR X8-32 instead of leaving ambiguous blank cells.
• Released immutable component catalog v0.4.3 with 563 claims; all earlier catalog and LiDAR research files remain unchanged.

Reason: turn the first deep actuation category into an auditable duty-condition dataset while keeping manufacturer labels, unknowns and thermal limits visible.

2026-08-10 · Site release 0.4.3 / catalog 0.4.2

• Published RCI 014, a condition-preserving audit of 21 target-qualified detection observations across nine revision-bounded 3D LiDAR records.
• Added immutable JSON and observation-level CSV for the LiDAR range-condition research dataset v0.1.0.
• Updated the current Ouster OS0 and OS1 records to Rev8, separated target-qualified detection from maximum representable range, and kept OS1 Max outside the OS1 record.
• Kept Ouster OS2 explicitly at Rev7 and tied its representable-range boundary to firmware and signal-multiplier conditions.
• Added the XT32 80 m / 10% / PoD >90% observation and preserved two current QT128 official-source conflicts: point rate and mass.
• Released immutable component catalog v0.4.2 with 561 claims; all earlier catalog files remain unchanged.

Reason: turn a high-value non-joint category into observation-level research while correcting revision drift instead of ranking incompatible headline distances.

2026-08-10 · Site release 0.4.2 / catalog 0.4.1

• Added five bounded independent-evidence claims for the exact XM540-W270-R, RealSense D455, Livox Mid-70, Jetson Orin Nano 8GB and Molicel P42A records.
• Raised the V-state count from one to six while leaving all 525 manufacturer-supported claims unchanged.
• Published RCI 013, an evidence review that states both the result and non-generalization boundary for every independent record.
• Released immutable component-catalog v0.4.1 JSON and claim-level CSV; v0.4.0 remains available unchanged.

Reason: expand independent evidence across actuation, depth vision, LiDAR, compute and battery records without treating one experiment as a product-wide endorsement.

2026-08-10 · Site release 0.4.1

• Expanded the monitored source directory from 43 to 88 channels across nine groups.
• Added standards bodies, open scholarly indexes, Korean and Hong Kong filings, Japanese programmes, specialist events, component trade media and manufacturer engineering channels.
• Published source-directory schema v2 with language, access, monitoring method, direct feed/API and last-rights-review fields.
• Released versioned source-directory JSON and CSV plus stable current aliases.
• Added a canonical /sitemap.xml compatibility redirect after crawler logs showed repeated requests to that path.
• Recorded the first observed Googlebot, Bingbot and ClaudeBot production crawl baseline.

Reason: turn the source list into an operational intelligence watchlist and remove a crawler-discovered sitemap dead end without changing catalog v0.4.0.

2026-08-10 · Release 0.4.0

• Expanded from 45 to 75 audited component records while retaining 25 stable manufacturer entities.
• Added 246 claim-level records for 544 total claims across eight actuators, eight camera/LiDAR records, six compute platforms, four battery/cell records and four host-to-CAN interfaces.
• Published the first independent V evidence state using a peer-reviewed D455 accuracy study.
• Preserved current-page versus older-manual Hesai QT128 point-rate and two other official-source conflicts as C instead of silently selecting one value.
• Added product-revision, replacement, supply-chain, connector, cell-configuration and explicit unknown fields.
• Froze v0.3.1 JSON and CSV with SHA-256 integrity checks before moving stable aliases to v0.4.0.

Reason: reach the first 75-record coverage milestone while increasing evidence diversity and procurement usefulness rather than merely increasing product count.

2026-08-10 · Release 0.3.1

• Corrected one NVIDIA Jetson official-source URL that returned 404.
• Kept all 45 component records, 25 manufacturer profiles and 298 claim records unchanged.
• Froze v0.3.0 JSON and CSV with SHA-256 integrity checks before moving stable aliases to v0.3.1.

Reason: restore a readable primary-source route without silently rewriting the already-published v0.3.0 snapshot.

2026-08-10 · Release 0.3.0

• Expanded from 25 to 45 audited component records while retaining the same 25 manufacturer entities.
• Added 139 claim-level records for 298 total claims: six actuators, five camera/LiDAR products, four compute platforms, three battery/cell records and two CAN interfaces.
• Added four explicit U evidence states for material fields that current official specifications do not publish.
• Expanded all six comparison tables without introducing composite scores.
• Froze v0.2.0 JSON and CSV with SHA-256 integrity checks before moving stable aliases to v0.3.0.

Reason: increase within-manufacturer and within-category comparability while preserving product-generation, mode, reflectivity, thermal, software and interface conditions.

2026-08-10 · Release 0.2.0

• Expanded to 25 manufacturer profiles, 25 audited component records and 159 claim-level records across all five hardware layers.
• Added MYACTUATOR, Basler, SICK, Advantech, AAEON, Seeed Studio, Toradex, Ultralife, Kvaser and HMS/Ixxat.
• Published the first interface-layer records and a sixth comparison table for USB-to-CAN interfaces.
• Published the first C evidence state: two official Kvaser documents disagree on the minimum classical-CAN bitrate.
• Added multiple-source support for conflict claims and secondary-source columns to the claim CSV.
• Made the v0.0.1 and v0.1.0 release files immutable; stable aliases moved to v0.2.0.

Reason: reach the 25-manufacturer milestone while proving that the same condition-preserving schema works for non-joint systems and formal source conflicts.

2026-08-09 · Release 0.1.0

• Published 15 source-audited component records and 15 manufacturer profiles spanning integrated actuation, depth cameras, 3D LiDAR, embedded compute and batteries.
• Released 89 claim-level records with conditions, source locations, verification dates and explicit evidence states.
• Published five cross-category evidence tables without composite scores.
• Added versioned JSON and flattened claim CSV plus stable current aliases.

Reason: pressure-test the evidence schema with real, unlike product records before expanding record count.

2026-08-09 · Release 0.0.1

• Established the five-layer component taxonomy: motion, perception, compute, power and integration.
• Declared actuation and integrated joints as the first deep evidence file.
• Published the public methodology, editorial policy, corrections process and coverage boundary.
• Released the first 43-channel source directory with monitoring cadence and source-specific reuse guidance.
• Published RCI 001 and RCI 002 as the launch research notes.

Reason: establish the evidence and rights infrastructure before publishing comparative product claims.