Skip to content
MECHDOJO

Type to search.

    Corrections

    A site whose whole claim is on the record and that never corrects itself is saying one of two things: that it never gets anything wrong, or that it does not tell you. Both are worse than the list.

    Every entry keeps the previous value. A correction that erases what it replaced is not a correction, it is a rewrite.

    Ours 8

    1. We invented a manufacturer. There is no such thing as a “Unitree T1”

      Platform: Unitree T1 Platform: Booster T1, and its specifications are a declared gap

      Our source said only “T1”. We added “Unitree” because the other machine in that bout was a Unitree G1, and that was a guess dressed as a fact. Unitree’s catalogue contains no T1 at all — the URL 404s in English and in Chinese — and Booster Robotics documents the T1 in its own newsroom as an official RoboCup partner machine. Filling a gap quietly is the one thing this site is not allowed to do, and we did it.

      See it on the page →

    2. We called the World Humanoid Robot Games an autonomous league. Its own rules say otherwise

      Control: Autonomous Control: manual remote, whole-body teleoperation or full autonomy

      The organiser’s 187-page rulebook allows all three, in the same weight class and on the same mat. Autonomy is an entry option there, not the regime of the competition — which is also why its fair-competition review only polices teams that entered as fully autonomous. We had taken the label from press coverage instead of the rulebook, and the rulebook was published on the same site we were already reading.

      See it on the page →

    3. We said the sport had five bouts with a named winner. It has forty-eight

      “There is no primary record to aggregate” Forty-three bouts from the organiser’s own brackets, plus five named champions at REK

      It was the headline on our front page and it was wrong the day we published it. The World Humanoid Robot Games had posted three official result PDFs on 26 August 2026 — full brackets, score and method for every bout — and REK publishes twelve events with five champions. We missed both for one reason: we read the HTML and not the resource. rek.com is a React app that serves the same 7,171-byte shell at every URL and keeps its archive in a JavaScript bundle; whrgoc.com serves its bulletins from an API. Reading a shell and calling it an absence is our error, not theirs.

      See it on the page →

    4. Matador’s page said it tore the head off its opponent. It was the other way round

      “Kept fighting after tearing the head off its opponent” “Kept fighting after having its own head torn loose”

      White Eagle landed the high kick and it was Matador’s head that came loose — which is the whole reason the bout travelled. Our own White Eagle page and our own fight page had it right, so the error was ours alone and it inverted the single most quoted fact in the sport.

      See it on the page →

    5. REK was listed with one bout, an approximate date and an unnamed event

      One bout, February 2026 (approx.), “REK’s second bout” Twelve events with date and venue, five champions, and the bout dated 7 February 2026 at Kezar Pavilion

      The league had published all of it. We had recorded “rek.com publishes no results” after reading a page that renders entirely in JavaScript.

      See it on the page →

    6. Height and weight were shown as “pending” on every machine page

      Marked as located-but-unread 3,990 weights and 4,025 heights, published

      The figures were in the entry lists all along. On a site whose whole claim is that gaps are shown, marking something as a gap when you do have it lies exactly as much as filling one you do not.

    7. Fifty-three fighters linked to pages that did not exist

      Podium winners with no page Every machine that reached a podium has a page

      They are fighters that stood on a podium and have no line in the results record, so they were never built. A link checker now runs against the live site and fails the build if one breaks.

    8. Japanese and secondary labels were unreadable on the black band

      Contrast 2.47:1 Contrast 12.89:1

      The grey is calibrated for text on paper, where it gives 5.90:1. On the dark band it gives 2.47:1, under half the AA threshold. Caught by an automated audit against the live site, not by looking.

    In what others publish 2

    These are not mistakes of ours. They are figures in circulation that do not match what the primary source says, and we say so on the page where the figure appears.

    1. EngineAI T800 specifications

      A single spec sheet: 173 cm, 75 kg, 43 DoF, $25,000 Four editions: 25, 25, 43 and 46 DoF, from $40,500 to $80,800

      Read at the manufacturer’s own product page. The price in circulation is 62 % below their cheapest edition and matches none of the four. A robot with four editions does not have a spec sheet.

      See it on the page →

    2. The Chinese name of White Eagle

      白鹰 白色雄鹰

      The Chinese sources we read say 白色雄鹰, “imposing white eagle”. Our own census had carried the shorter form. Corrected against the sources, not against a preference.

      See it on the page →

    Standing discrepancies we do not resolve

    Separate from corrections, and deliberately left open: in 23 editions the entry count the organiser declares does not match the entry list they published. We show both figures and pick neither. It is their discrepancy, and resolving it by preference would be inventing a record.

    Everything that is missing →