Skip to content
Botlero

Corrections

Every value Botlero has stopped publishing is kept, not deleted, with the reason that was recorded when it was withdrawn and a link to the page that published it. Newest first, across every robot.

An entry says a value stopped being published and why. It does not say what replaced it: often nothing did, and where something did, the replacement is on the robot's own page with its source and its date.

367 withdrawn values · 172 recorded reasons · newest 2026-09-27

2026-09-27109 values, 49 reasons · 57 on this page

  1. Withdrawn for the same reason as the summary sentence: it gave 157 TOPS of AI compute where the published figure is 100 TOPS, from three independent sources and consistent with the Arm Cortex-A78AE module Unitree names for G1. The 157 came from one directory cell whose own page also prints "100-157 TOPS", and that statement has been withdrawn. The compute module, the 16 GB of memory and the 2 TB of storage are unchanged and still shown. No replacement text has been written.

    • Unitree G1FaqsWhat are the computing specifications of the Unitree G1?

    1 value · recorded by p0-2026-09-28

  2. Withdrawn because it gave 157 TOPS of AI compute, which is not what this robot's published facts say. The AI compute now shown is 100 TOPS, stated by three independent sources and consistent with the computer Unitree itself names for G1 — an Arm Cortex-A78AE module with 16 GB of memory and 2 TB of storage. The 157 came from one directory's spec cell whose own prose on the same page reads "100-157 TOPS", and that statement has been withdrawn and is listed among this robot's corrected values. The compute module, its 16 GB of memory and its 2 TB of storage are unchanged and still shown. No replacement sentence has been written; nothing else in the summary was changed.

    • Unitree G1DescriptionThe compute unit is an Arm® Cortex®-A78AE with 16 GB memory and 2000 GB storage, providing 157 TOPS AI performance, and…

    1 value · recorded by p0-2026-09-28

  3. Both quotes say in their own words that these two degrees of freedom belong to the EDU configuration and are optional — "The EDU configuration adds even more flexibility with a 3-DOF waist (pitch, roll, yaw) and 2-DOF wrists" and "7 DOF (optional force-control three-fingered hand) + 2 optional wrist DOF", the second on a page that is only about G1 EDU. Both were stored without a version, and a statement with no version is shown on every version, so an optional EDU addition was published as a fixed spec of the base G1. Unitree's own statement of the same fact is written "+2(Optional 2 additional wrist degrees of freedom)" and is correctly attached to the EDU version, where it still stands.

    2 values · recorded by p0-2026-09-28

  4. Both figures are contradicted by the computer Unitree itself names for G1 — an Arm Cortex-A78AE module with 16 GB of memory and 2 TB of storage, i.e. a Jetson Orin NX 16 GB, whose rating is the 100 TOPS that three independent sources state and that this field now shows. "EDU with 29 DOF and 40 TOPS compute" attaches its number to one named configuration, is stored with no version so it showed on the base G1, and names a DoF count that exists nowhere in our data — G1 EDU is 23-43 DoF on twelve statements including Unitree's own. "AI Compute 157 TOPS" is the top of a range the same page states as a range: livium's own prose on that URL reads "100-157 TOPS of AI compute performance", and this field holds a single number, so the cell published the ceiling as the figure.

    2 values · recorded by p0-2026-09-28

  5. Agility publishes no price for Digit on any page we read, and each value withdrawn here is one this field's own rules exclude. "Price RaaS subscription (~$250,000 purchase equivalent)" is a subscription, is written with a tilde, and a "purchase equivalent" is not a price the seller asks — and its byte-identical twin from the same page was already withdrawn on 2026-09-23 and was re-added by a later run, because a statement is recognised by field + page + value and the withdrawn copy is not consulted. "Enterprise pricing, typically custom-quoted; likely >$100,000 per unit" hedges twice in words this field excludes by name and then says outright that the price is quoted case by case. The bare "45 000 USD" cell is from humanoid.guide's pre-Digit-5 page, while the SAME directory's Digit 5 page prints "150 000 USD" — one source publishing two prices on two pages is what settles which page is about which machine.

    3 values · recorded by p0-2026-09-28

  6. This is the previous Digit's delivery plan, from Agility's 2019 introduction post: "Agility will announce Digit pricing mid-year, with deliveries beginning Q1 2020." This page is about Digit 5, and Agility's own Digit 5 release states "Early access to Digit 5 is expected to begin in the first half of 2027, with general availability … expected by the end of 2027", which four further sources restate. Six statements on this field have already been withdrawn for the same reason. Until now this one was also the value the public JSON served, while the page itself already showed 2027.

    1 value · recorded by p0-2026-09-28

  7. The whole robot's payload, relabelled as one arm's. Figure states "Payload: 20KG" on its own page and thirteen further sources restate 20 kg as the robot's payload. originofbots prints the identical figure as "20 kg per arm", which would make the robot's payload 40 kg — a number no source states, that page included. This field is the payload of ONE arm and its rules exclude the total for both arms or for the whole robot.

    1 value · recorded by p0-2026-09-28

  8. This field holds a charge capacity in Ah or mAh, and its own rules exclude an energy figure in Wh — which is what this number is. 1X publishes "Capacity 842 Wh", restated by nine directories and stored under battery energy. livium copied that same 842 into a column headed mAh, so the quote reads "Battery Capacity 842000 mAh" and the value became 842 Ah — roughly 42 kWh, about fifty times the pack 1X publishes, in a 30 kg robot. No source states any charge capacity in Ah or mAh for NEO.

    1 value · recorded by p0-2026-09-28

  9. This was a RATE, not a duration. 1X's own words are "Quick charge 6min per hour runtime": six minutes of charging buys one hour of running, so the 6 is the numerator of a ratio whose denominator is the runtime. This field is the time to charge the robot fully, and livium states the same vendor fact completely — "can fast-charge to full in about 24 minutes (6 minutes per hour of runtime)" — which is the value four independent sources give and the one this key now shows.

    1 value · recorded by p0-2026-09-28

  10. Unitree states this as an ARM load, not the robot's payload: www.unitree.com/h1 prints the row "Arm normal load" / "Peak: About 21Kg; Rated: About 7Kg" (Chinese: 手臂常规负载 / 峰值：约21Kg；额定：约7Kg), and both figures are already stored correctly under loadPerArmKg from that same vendor page. Each value withdrawn here republishes one of them as the whole robot's rated payload: roboatlas writes "Rated 7 kg / peak 21 kg per arm" on the same page as its own "21 kg General Payload"; biorobot.ai calls its 21 kg a "rated arm payload"; robotoverflow calls its 7 kg "a nominal payload of 7 kg per arm" and adds that "the base H1 has no rated arm payload"; humanoid.guide prints a bare "Strength [kg] 21" cell that names no subject at all. This field is the rated payload of the WHOLE robot, and its registry row excludes both a peak load and a per-arm value.

    4 values · recorded by cleanup4-2026-09-27

  11. humanoid.guide's RE-HOSTED copy of Agility's mark, which registry.txt's logoUrl row rejects BY PATH. Its only backing statement was closed earlier today (validTo 2026-09-27T10:42:47Z) and deriveRobot cleared Robot.logoUrl, but an Organization column deliberately never clears, so the stale value survived. It is not merely stored: /api/robots/agility-digit serves it live as manufacturer.logoUrl, so the public JSON published a directory's copy as the maker's own logo. It is correctly absent from Organization.logo in JSON-LD (the same-domain test fails) and is rendered as no image anywhere.

    • Agility Robotics DigitManufacturer · Logo URLhttps://humanoid.guide/wp-content/uploads/2026/09/Agility-Robotics-logo-humanoid-guide.webp

    1 value · recorded by cleanup3-2026-09-27

  12. description is an ARRAY of Cited, one per sentence, joined with " " — so the gate drops exactly this one item. check() returns number-not-in-facts:30 degrees on it. No other sentence is touched, nothing is rewritten, nothing is renumbered.

    • Figure 03DescriptionFigure 03 is 172.72 cm tall and weighs 61 kg. It has 30 degrees of freedom, with 16 degrees of freedom per hand, and 5…

    1 value · recorded by cleanup3-2026-09-27

  13. check() returns number-not-in-facts:30 degrees. quickAnswer is ONE Cited object covering 1-2 sentences (editorial.ts SCHEMA), so the gate drops both sentences together even though only the first states 30 — the unit is the field, not the sentence.

    • Figure 03Quick AnswerFigure 03 is a bipedal robot with 30 degrees of freedom and a rated payload of 20 kg. It can walk at 1.2 m/s, climb sta…

    1 value · recorded by cleanup3-2026-09-27

  14. check() returns number-not-in-facts:30 degrees on the WHOLE field — tagline is one Cited object, so the gate drops it entirely. jointCountTotal is now 44 (5 sources, two with the decomposition "44 (32 in hands)"); 30 has 4 and figure.ai publishes no total DoF at all.

    • Figure 03TaglineFigure 03 is a bipedal robot with 30 degrees of freedom, designed in the US.

    1 value · recorded by cleanup3-2026-09-27

  15. The quote is about TELEOPERATION, which this row's own REJECT names: "on-site or remote said of something else (on-site training, remote teleoperation, a remote operator) — the page must say the SERVICE or repair is on-site / depot / remote". Re-gated with the production gate() against this same cached page, the statement now FAILS enum:not-in-quote: not one of the REMOTE option's six declared aliases (remote support, remote service, remote diagnosis, online support, 远程支持, 远程协助) is in the quote. It only ever passed because a long option CODE counted as its own spelling, the hole closed on 2026-09-25 whose three measured removals were exactly these. Control: "remote diagnosis and remote service are included for every NEO owner" still PASSES as REMOTE, so the rule is not refusing the concept.

    3 values · recorded by cleanup3-2026-09-27

  16. "app integration" proves no companion phone app: on originofbots the phrase is one item of a DEPLOYMENT-MODES cell — "AI autonomous control with voice command, remote monitoring, app integration" — which is equally satisfied by a web dashboard, the thing this row's REJECT excludes. It is the key's only statement, and figure.ai publishes no app, so the honest render is no field at all.

    1 value · recorded by cleanup-2026-09-27

  17. A COUNT of speakers stored as watts: every one of these quotes counts loudspeakers — "3 stage speaker in pelvis and chest area", "3 speakers", "3-stage speakers", "3-stage speaker system". The row's unit is W and its ACCEPT is "speaker power in W"; compare unitree-g1's real "5W Speaker". 1X publishes no speaker wattage.

    5 values · recorded by cleanup-2026-09-27

  18. A dimensionless MULTIPLE stored as N·m/kg: guijicankao's "扭矩重量比5.5倍" is "torque-to-weight ratio 5.5 TIMES" (倍 = fold/times) — no unit at all. The row's ACCEPT is "a stated N.m/kg number"; compare unitree-h1's real "Peak Torque Density 189N.m/Kg". This is the same class as CLAUDE.md's RULE UNITLESS.

    1 value · recorded by cleanup-2026-09-27

  19. A DERIVED value, which CLAUDE.md forbids storing as a spec: robozaps's own text is "Degrees of freedom 75 total, DERIVED FROM PUBLISHED JOINT COUNTS" — the page says it computed the number rather than reading it. The other seven statements of 75 are independent and stay, so the key and the shown value are unchanged.

    1 value · recorded by cleanup-2026-09-27

  20. "Proprietary" is a non-disclosure, not an operating system, and this row's REJECT already covers the class: "a cell that says N/A, none or 'not published' — that is not a value". The key keeps three real values (1X OS, 1X OS (Nvidia Jetson Thor), Ubuntu Linux 24.04).

    1 value · recorded by cleanup-2026-09-27

  21. The row's REJECT names this exact shape too: "a 'Bluetooth: Yes / supported' cell — this key is the version number only". The stored quote is aparobot's Connectivity bullet "Bluetooth supported" and the stored value is the word "supported".

    1 value · recorded by cleanup-2026-09-27

  22. The row's REJECT names this exact shape: "the bare word 'Bluetooth' in a connectivity list ('Wifi, Bluetooth, 5G')". The stored quote is 1X's own "Communication Wifi, Bluetooth, 5G" and the stored value is the word "Bluetooth". This key is the version number only.

    1 value · recorded by cleanup-2026-09-27

  23. 腕部 is ARM in Japanese, not wrist (wrist is 手首): humanoidpress.jp's "28 DOF（腕部4 DOF×2）" is 4 DoF per ARM times two arms — and that same quote is already stored correctly as jointCountPerArm = 4, beside Agility's own "Digit's four degree-of-freedom (4-DOF) arms". CLAUDE.md already records this CJK subject-word collision.

    1 value · recorded by cleanup-2026-09-27

  24. A directory's RE-HOSTED copy of the maker's mark, which this row's REJECT names by path: "a logo re-hosted by a directory (humanoid.guide/wp-content/…, a directory's own CDN copy) — it must be on the maker's own domain". The value is literally https://humanoid.guide/wp-content/uploads/2026/09/Agility-Robotics-logo-humanoid-guide.webp. The same defect was fixed on unitree-h1 by today's run and introduced here.

    1 value · recorded by cleanup-2026-09-27

  25. A vertical REACH read as an arm length, which the row rejects by name ("arm span, reach"): aparobot's sentence is "a vertical reach of 0 to 1,670 mm", and that same quote is already stored correctly as reachM and as workspaceHeightRangeM 0–1,670 mm. Agility's own page puts Digit's reach at 2.2 m, so 1,670 mm is not an arm either.

    1 value · recorded by cleanup-2026-09-27

  26. The quote is about a different robot generation than the one it is stored on: "Unitree H1-2 is the second generation of the full-size Unitree humanoid" — H1-2 is the second, so H1 (this robot) is the FIRST. scope=robot, so no variant column can rescue it.

    1 value · recorded by cleanup-2026-09-27

  27. "Remote Control Connection" is a TUTORIAL-VIDEO TITLE, not a statement about what ships with the robot: on www.unitree.com/app/h1 it is one bullet of the "H1/H1-2 Tutorial videos" list (beside "Calibration", "Posture & Movement", "Packing"). The key's ACCEPT is "true when a handheld controller comes with it".

    1 value · recorded by cleanup-2026-09-27

  28. This is the MAXIMUM voltage, which the row's REJECT names word for word ("charger or max voltage"): aparobot prints the bare "67.2 V" in "a 15 Ah (0.864 kWh) battery at 67.2 V" with no "nominal" label, while Unitree's own page prints "Max Voltage 67.2V" / "最大电压67.2V" and the same number is already stored 6× under batteryMaxVoltageV. 67.2 V is also exactly 16S x 4.2 V, i.e. full charge.

    1 value · recorded by cleanup-2026-09-27

  29. An OPTIONAL accessory hand's DoF, not this robot's hand DoF: robotsatlas's own sentence is "H1-2 supports OPTIONAL dexterous hands INCLUDING the Dex5-1 (5 DOF anthropomorphic) OR other ambidextrous configurations from the Unitree portfolio" — one item from a catalogue of choices, attached to no variant. The quote never says "active" either, which is the whole point of this key against handJointCount.

    1 value · recorded by cleanup-2026-09-27

  30. Not a language of the official SDK: GitHub's org page lists "Rust interface for unitree_sdk2 with C++ SDK2 bindings" — a third-party/wrapper repo around the C++ SDK, i.e. the row's REJECT "languages of an unrelated tool". CLAUDE.md already records RUST as a known alias hole on this exact host (it also matched "Trust center"). PYTHON and CPP survive from the vendor and two directories.

    1 value · recorded by cleanup-2026-09-27

  31. 3.3 m/s is H1's MAXIMUM, not its walking speed, and the row's REJECT says so in its own words: "running / maximum / top speed, including a speed the page itself calls a record ('Moving speed of 3.3m/s(world record)' is maxSpeedMps, not walking)". Nothing is lost: 3.3 m/s is already stored under maxSpeedMps from Unitree's own page plus 12 directories, and Unitree's own "移动速度 ＜ 2m/s" (up to 2 m/s) survives here.

    14 values · recorded by cleanup-2026-09-27

40 reasons per page. A robot's own withdrawals are listed on its page, under “Corrected values”. How a value gets published.