Robotics Falsification Checklist for Morning Review v1

Date: 2026-06-12 Owner: Hugo / Genius Team Agent: Finance / Charlie AGT-002 Status: public-safe research artifact for morning review and Codex packaging guardrails Visibility: PUBLIC Related files:

• knowledge/robotics-research-harness.md
• knowledge/robotics-public-claim-bank-v1.md
• knowledge/robotics-tesla-figure-unitree-slide-evidence-v1.md
• knowledge/robotics-oem-three-evidence-curves-public-brief-v1.md
• knowledge/robotics-leaderdrive-case-study-slide-evidence-v3.md

Public-safety: yes. No Hugo private portfolio data. No trade recommendation.

0. One-line answer

截至 2026-06-12，机器人研究的核心结论不应该是“谁赢了”，而应该是：哪些证据足以把行业从 S3/S4 推向 S5，哪些证据只是更漂亮的 demo / customer logo / capacity story。

Current label: S4-heavy evidence environment, not S5 scaled-commercial-economics proof. 🟠 synthesis based on primary-source packets through 2026-06-12.

1. Why this artifact exists

前面的 artifacts 已经把 Tesla / Figure / Unitree / 绿的谐波 / why-now catalysts 做厚。早审和 Codex 包装最大的风险不是材料不够，而是压缩时把四类不同证据混成一个“商业化已完成”的叙事：

1. designed capacity 被误读成 actual production；
2. customer KPI 被误读成 full customer economics；
3. low price 被误读成 reliability / gross margin；
4. supplier filing revenue 被误读成 named humanoid OEM revenue。

这份 checklist 的用途：

• 给 morning review 一个 5 分钟审稿表；
• 给 Codex 一个包装前的反 hype gate；
• 给后续研究一个“什么会改变结论 / 什么不会”的更新规则。

2. The S5 test: what would actually change the thesis

S5 requires a chain, not a single headline

A humanoid robotics claim becomes thesis-changing only when multiple links become public and verifiable:

Current conclusion as of 2026-06-12: public evidence has several S4 links, but not the full S5 chain. 🟠 synthesis.

3. Company / layer falsification gates

3.1 Tesla Optimus gate

Current public evidence:

• Tesla disclosed Fremont first-generation Optimus line designed for 1M robots/year and Texas second-generation line long-term designed annual capacity of 10M robots. Tesla Q1 2026 Form 8-K Exhibit 99.1, filed 2026-04-22. 🟢
• Reviewed Tesla Optimus filing sequence through Q1 2026 does not disclose external paid Optimus customers, actual Optimus quarterly output, line yield, robot ASP, robot revenue, robot gross margin, or customer payback. 🟢/🟠 as of 2026-06-12.

Current signal grade: S4 manufacturing-infrastructure / capacity-intent signal.

Upgrade toward S5 if:

1. Tesla discloses actual Optimus quarterly production output, not only designed capacity. 🟢
2. Tesla discloses line yield, utilization, cost per robot, or production learning curve. 🟢/🟠
3. Tesla discloses internal factory task KPI: hours worked, tasks, uptime, intervention rate, labor/cost savings. 🟢/🟠
4. Tesla discloses external paid deployment, price, revenue, or customer contract. 🟢
5. Tesla links Optimus to financial statement materiality, capex efficiency, or segment economics. 🟢

Downgrade / caution if:

• Future updates keep emphasizing capacity design while actual output and task KPIs remain absent. 🟠
• Production-line buildout slips without quantitative explanation. 🟢/🟠
• Optimus demos improve but remain detached from runtime, intervention, cost savings, or customer economics. 🟠

Do-not-say:

• “Tesla has 1M / 10M current annual Optimus production.”

Use instead:

• “Tesla disclosed designed capacity / long-term designed annual capacity; actual production and economics remain undisclosed.” 🟢/🟠

3.2 Figure AI gate

Current public evidence:

• Figure BMW deployment disclosed 11-month deployment, 10-hour shifts Monday-Friday, 90,000+ parts loaded, 1,250+ runtime hours, and contribution to 30,000+ BMW X3 vehicles. Figure official post, 2025-11-19. 🟢
• Figure BotQ disclosed 350+ Figure 03 delivered, production cadence from 1/day to 1/hour, EOL first-pass yield >80%, battery-line first-pass yield 99.3%, 500+ battery packs, and 9,000+ actuators. Figure official post, 2026-04-29. 🟢
• Figure Catalyst Brands commercial agreement disclosed logistics deployment start at Reno, Nevada Distribution Logistics Center, but did not disclose robot count, contract value, ROI/payback, repeat-order terms, uptime/intervention, revenue, or margin. Figure official post, 2026-05-26. 🟢/🟠

Current signal grade: S4 deployment + manufacturing signal; Catalyst remains S3/S4 until economics appear.

Upgrade toward S5 if:

1. BMW, Catalyst, or another named customer confirms ROI/payback or productivity economics. 🟢
2. Figure discloses robot count by customer and repeat-order / multi-site rollout terms. 🟢/🟠
3. Figure discloses uptime, intervention rate, reset count, maintenance burden, and task success distribution across many shifts. 🟢/🟠
4. BotQ output is linked to customer revenue, backlog, utilization, and gross margin rather than only delivered robot count. 🟢/🟠
5. The same deployment playbook repeats across at least 2-3 customer contexts. 🟢/🟠

Downgrade / caution if:

• BMW remains a one-off showcase and Catalyst stays economics-undisclosed. 🟠
• Manufacturing output grows faster than field utilization/customer conversion. 🟠
• Autonomy claims remain video-centric without long-duration reliability metrics. 🟠

Do-not-say:

• “Figure has proven scaled humanoid unit economics.”

Use instead:

• “Figure has unusually strong public deployment/manufacturing KPI for the sector, but S5 still requires customer economics, repeatability, intervention, and margin evidence.” 🟢/🟠

3.3 Unitree / 宇树 gate

Current public evidence:

• Unitree official product pages list R1 from US$4,900, G1 from US$13.5K, and H2 at US$29,900, tax/shipping excluded. Accessed 2026-06-10. 🟢
• Unitree H2 Plus / G1-D pages reference NVIDIA Jetson T5000, Isaac GR00T / TeleOp / Sim, and data/training workflows. Accessed 2026-06-10. 🟢
• Reviewed official pages do not disclose humanoid unit shipments, industrial customer runtime/task KPI, repeat orders, gross margin, maintenance cost, customer payback, or warranty/service burden. 🟢/🟠 as of 2026-06-12.

Current signal grade: S3 cost/access + platform signal; possible S4 if platform adoption becomes verifiable.

Upgrade toward S5 if:

1. Unitree discloses humanoid shipment volume by product line and customer type. 🟢
2. Unitree discloses industrial customer deployments with runtime, uptime, task KPI, and repeat orders. 🟢/🟠
3. Low-cost hardware is linked to sustainable gross margin or software/service revenue, not only lower ASP. 🟢/🟠
4. H2 Plus / G1-D becomes a measurable developer/data platform with adoption metrics, paid tools, dataset scale, or model ecosystem traction. 🟢/🟠
5. Support, maintenance, warranty, and reliability data show low price is not offset by high service burden. 🟢/🟠

Downgrade / caution if:

• Low-cost humanoids mainly expand demos/research but not paid deployment. 🟠
• Hardware gross margin is weak or support costs overwhelm ASP advantage. 🟠
• Value migrates to third-party AI/software/deployment layers while Unitree captures commoditized hardware margin. 🟠

Do-not-say:

• “Unitree low price proves it wins.”

Use instead:

• “Unitree low price changes experiment cost and ecosystem access; it does not by itself prove reliability, margin, or customer ROI.” 🟠

3.4 Leaderdrive / 绿的谐波 supplier gate

Current public evidence:

• 绿的谐波 2025 annual report disclosed revenue RMB 570.714m (+47.31% YoY), net profit RMB 124.367m (+121.42% YoY), harmonic reducer sales 425,158 units (+72.48% YoY), and “工业及具身智能机器人零部件” revenue RMB 422.528m with 34.88% gross margin. Disclosed 2026-04-23. 🟢
• 绿的谐波 2026 Q1 report disclosed revenue RMB 140.134m (+42.96% YoY) and net profit RMB 32.634m (+61.17% YoY). Disclosed 2026-04-23. 🟢
• 2026-06-04 to 2026-06-08 share price +36.00% vs SSE Composite -3.05% and automation-equipment index +2.13%; company said no undisclosed major matter. Abnormal trading announcement, 2026-06-09. 🟢
• Reviewed filings do not disclose named humanoid OEM customer, humanoid-specific revenue split, humanoid-specific margin, order volume, single-robot value content, or customer ROI/payback. 🟢/🟠 as of 2026-06-12.

Current signal grade: S4 filing-backed supply-chain validation, not S5 humanoid-specific economics.

Upgrade toward S5 if:

1. Filing or official disclosure identifies humanoid-specific revenue/order contribution. 🟢
2. A named OEM/customer confirms supplier relationship, volume, and program timing. 🟢
3. Gross margin stays resilient while reducer / actuator volume grows. 🟢
4. Product roadmap shows verified fit beyond harmonic reducers if humanoid architecture shifts toward integrated actuators / screws / linear modules. 🟢/🟠
5. Inventory and receivables quality remain healthy during robot-related growth. 🟢

Downgrade / caution if:

• Revenue grows but gross margin compresses sharply or receivables/inventory quality deteriorates. 🟢/🟠
• Public market narrative infers Tesla/Figure/Unitree/AgiBot/UBTECH exposure without company/customer confirmation. 🔴/🟠
• Humanoid architecture reduces harmonic reducer value content faster than volume grows. 🟠

Do-not-say:

• “绿的谐波已被证明是某个 humanoid 龙头供应商。”

Use instead:

• “绿的谐波报表信号强于概念叙事，但 named customer / humanoid-specific economics 仍是缺口。” 🟢/🟠

4. Morning review 5-minute scorecard

Use this table before approving any slide/site wording.

5. Signal vs noise rules

Strong signals

• Filing-backed actual output, revenue, margin, or segment exposure. 🟢
• Named customer + quantified runtime/task KPI + repeat order / expansion. 🟢
• Customer-confirmed ROI/payback or productivity improvement. 🟢
• Production yield/cadence tied to customer deployment or revenue. 🟢/🟠
• Supplier customer relationship confirmed by both sides or by filing-backed revenue/order data. 🟢

Monitoring signals

• Official product price/spec changes that materially lower experimentation cost. 🟢
• Commercial agreement with named customer but without count/value/economics. 🟢/🟠
• Manufacturing capacity/design plan without actual output/yield. 🟢/🟠
• Official AI-stack / model / simulation tooling integration. 🟢 for existence; 🟠 for commercial implication.

Noise unless upgraded

• Viral demos without reset count, intervention rate, runtime, or customer context. 🟠/🔴
• TAM-only charts without adoption/economics evidence. 🟠
• Customer logos without robot count, contract value, or task KPI. 🟢/🟠
• Supply-chain rumors without filing/customer confirmation. 🔴
• Share-price moves interpreted as order evidence. 🟢 for price move if filing-backed; 🔴/🟠 for order inference.

6. Public site section draft

What would prove the robotics thesis wrong — or right?

Robotics now has better evidence than a demo-only cycle. But the right question is not “which robot looks best?” It is “which claim survives falsification?”

A Tesla capacity claim survives if it stays clearly labeled as designed capacity, not current production. A Figure deployment claim survives if it says BMW runtime and BotQ manufacturing metrics are strong S4 evidence, but not complete economics. A Unitree cost claim survives if it says low price lowers experimentation cost, not that reliability or margin is proven. A supplier claim survives if it uses filings for what they actually say, without inferring unnamed humanoid customers.

The sector upgrades toward S5 only when these separate curves connect: paid deployment, repeat orders, uptime/intervention, customer ROI, production yield, robot revenue, and gross margin. Until then, the honest framing is: evidence is improving, but commercialization remains a testable hypothesis.

7. Slide-ready compression

Title: Robotics morning review: what would upgrade S4 to S5?

Main message: Evidence is improving, but S5 needs a full chain: paid deployment -> repeatability -> runtime/intervention -> customer ROI -> production economics -> financial materiality.

Four cards:

1. Tesla: capacity ambition

   • Proven: Fremont 1M/year designed line; Texas 10M/year long-term designed capacity. 🟢
   • Missing: actual output, yield, task KPI, robot revenue/margin. 🟢/🟠

2. Figure: deployment + manufacturing KPI

   • Proven: BMW 1,250+ hours / 90,000+ parts; BotQ 350+ robots / >80% EOL FPY / 99.3% battery FPY. 🟢
   • Missing: contract value, customer ROI, repeat orders, intervention, margin. 🟢/🟠

3. Unitree: cost-access curve

   • Proven: R1 US$4.9K; G1 US$13.5K; H2 US$29.9K; H2 Plus / G1-D platform tooling. 🟢
   • Missing: deployment KPI, shipments, uptime, support cost, margin. 🟢/🟠

4. Leaderdrive / 绿的谐波: filing-backed supplier signal

   • Proven: 2025 revenue RMB 570.714m, reducer sales 425,158 units, robot-parts revenue RMB 422.528m, 34.88% gross margin. 🟢
   • Missing: named humanoid OEM, humanoid-specific revenue/margin/order volume. 🟢/🟠

Footer: Evidence map only. No winner ranking. No trade recommendation. S4-heavy does not equal S5 scaled-commercial economics.

8. Common misconceptions

1. “如果行业证据变多，商业化就已经完成。”

   • Correction: evidence becoming measurable is not the same as complete paid repeat deployment and economics. 🟠

2. “设计产能、客户 KPI、低价、供应链收入可以直接加总成一个 winner score。”

   • Correction: they answer different questions and cannot be summed without double-counting or category error. 🟠

3. “只要有 customer logo，就一定有收入和 ROI。”

   • Correction: logo, pilot, paid deployment, repeat order, and customer ROI are different evidence levels. 🟢/🟠

4. “股价上涨可以证明订单来了。”

   • Correction: 绿的谐波 2026-06-09 abnormal-trading announcement explicitly says no undisclosed major matter while the stock had moved +36.00% over 2026-06-04 to 2026-06-08. 🟢

9. Think Deeper questions

1. If Tesla Optimus is deployed internally first, should S5 be defined by internal cost savings, external revenue, or both?
2. For Figure, which missing metric matters most: repeat order, customer ROI/payback, uptime/intervention, or gross margin?
3. For Unitree, does low-cost hardware shift value toward software/data/services, or simply accelerate hardware commoditization?
4. For suppliers, how do we distinguish real humanoid exposure from broad industrial/robotics segment growth?
5. Which evidence type is easiest for public markets to overprice before S5: capacity, customer logo, low price, supplier revenue, or viral demo?

10. Source list

Primary / official sources:

• Tesla Q1 2026 Form 8-K Exhibit 99.1, filed 2026-04-22: https://www.sec.gov/Archives/edgar/data/1318605/000162828026026551/exhibit991.htm 🟢 Used for Fremont 1M robots/year designed line and Texas 10M robots/year long-term designed capacity.
• Figure official post, “F.02 Contributed to the Production of 30,000 Cars at BMW”, 2025-11-19: https://www.figure.ai/news/production-at-bmw 🟢 Used for BMW deployment KPI.
• Figure official post, “Ramping Figure 03 Production”, 2026-04-29: https://www.figure.ai/news/ramping-figure-03-production 🟢 Used for BotQ manufacturing metrics.
• Figure official post, “Figure Signs Agreement with Catalyst Brands to Scale Humanoid Operations”, 2026-05-26. 🟢 for announcement; 🟠 for economics because robot count, contract value, ROI/payback, and margin are undisclosed.
• Unitree R1 product page: https://www.unitree.com/R1 🟢 Accessed 2026-06-10; used for R1 price.
• Unitree G1 product page: https://www.unitree.com/g1/ 🟢 Accessed 2026-06-10; used for G1 price.
• Unitree H2 product page: https://www.unitree.com/H2 🟢 Accessed 2026-06-10; used for H2 price.
• Unitree H2 Plus and G1-D product pages. 🟢 Accessed 2026-06-10; used for Jetson / Isaac / TeleOp / Sim / data-training references.
• 绿的谐波 2025 annual report, disclosed 2026-04-23: https://static.sse.com.cn/disclosure/listedinfo/announcement/c/new/2026-04-23/688017_20260423_EU45.pdf 🟢 Used for revenue, net profit, robot-parts revenue, gross margin, reducer production/sales/inventory, and product lines.
• 绿的谐波 2026 Q1 report, disclosed 2026-04-23: https://static.sse.com.cn/disclosure/listedinfo/announcement/c/new/2026-04-23/688017_20260423_TB4A.pdf 🟢 Used for Q1 continuation.
• 绿的谐波 2026-06-09 abnormal-trading announcement: https://static.sse.com.cn/disclosure/listedinfo/announcement/c/new/2026-06-09/688017_20260609_MZU2.pdf 🟢 Used for price-move anti-hype caveat.

Internal synthesis sources:

• knowledge/robotics-public-claim-bank-v1.md 🟠 synthesis artifact, dated 2026-06-11.
• knowledge/robotics-tesla-figure-unitree-slide-evidence-v1.md 🟠 synthesis artifact, dated 2026-06-11.
• knowledge/robotics-oem-three-evidence-curves-public-brief-v1.md 🟠 synthesis artifact, dated 2026-06-11.
• knowledge/robotics-leaderdrive-case-study-slide-evidence-v3.md 🟠 synthesis artifact, dated 2026-06-11.

11. Public-safety check

Safe to publish if used as written:

• No Hugo private portfolio weights.
• No buy / sell / hold language.
• No paid-report excerpts.
• No private-channel or unsourced claims.
• All material claims have source grade and as-of / filing / post date.

Do not publish:

• Any winner ranking.
• Any claim that designed capacity equals current production.
• Any claim that customer KPI equals full economics.
• Any claim that low price proves reliability or margin.
• Any claim that supplier filing revenue proves named humanoid OEM exposure.