A-Share Pre-Market Brief | 2026-08-19, Wednesday

0. One-Sentence Summary

Today is a head-on collision between an "event day" and a "rates day." The strongest catalyst is robotics: 宇树科技 (Unitree Robotics, 688836, STAR Market) lists today, day one + the 2026 World Robot Conference opens today in Beijing Yizhuang, and the conference has designated 8/19 as "Release Day" (concentrated release of major results, product debuts, cooperation initiatives), and per the conference preview SASAC will announce on site the establishment of a Central SOE Robotics Innovation Consortium — of these, the "listing" and the "opening" are accomplished facts, while the "consortium establishment" is a pre-conference preview that has not yet happened. Even counting only the first two, this is the only catalyst cluster today with the property of "certain to occur."

But the overnight overseas session delivered a hard counterpunch. US close on 8/18: Nasdaq −1.33%, and the memory chain collapsed as a group — SanDisk (SNDK) −9.01%, Seagate (STX) −9.16%, Western Digital (WDC) −7.43%, Micron (MU) −7.02%, AMD −4.27%; the 30-year Treasury yield rose to 5.31% (5.32% intraday, the highest since 2007), and Brent crude closed above $90 for the first time in three weeks. The trigger for the memory crash was the Wall Street Journal report on "roughly $3 trillion of off-balance-sheet AI-related commitments across nine tech giants," not any memory-sector fundamental negative — this was a re-rating driven by discount rates and the AI-financing narrative, not a turn in demand.

Likely direction of flows: high-multiple growth under pressure; event-driven small-cap components and the inflation chain (oil & gas / coal chemicals / agriculture) relatively favored. The driver mix is event + overseas industry (negative) + price, all three.

Pre-market read: robotics is today's only offensive direction, but it is simultaneously today's biggest "good-news-realization" risk position. The robotics chain was already at pricing day 1 yesterday (robotics industry index +3.44%, main-force net inflow RMB 733 million, reducer concept +RMB 1.373 billion), and 埃斯顿 (Estun Automation, 002747) only first sealed its limit-up at 14:56 yesterday, then broke it, and is not in the official limit-up pool; out of all 79 limit-ups, the robotics chain had only two small-cap bearing names, 南方精工 (Nanfang Precision, 002553) and 襄阳轴承 (Xiangyang Bearing, 000678) — no anchor name. There is a theme, there is an event, there is money, but there is no ladder — that is the crack to watch today.

⚠️ This edition sets no "priority deep-dive." This is not conservatism; it is the direct result of sub-agent verification: not one name in the robotics chain simultaneously satisfies "confirmed humanoid/Unitree business + real earnings + valuation not absurd." Every single one has a substantive defect — 埃斯顿's 8/18 block-trade (LHB) disclosure shows institutional seats with a combined net sell of RMB 171 million and northbound net selling of RMB 67 million, with the only two net buyers being two hot-money seats; the circulating "RMB 3.2 billion Unitree long-term agreement" for 中大力德 is directly falsified by its own 2026Q1 revenue of only +5.12% and ex-non-recurring −43.19%; 襄阳轴承's "joint development of Unitree H1 with Unitree" originates from an investor's question on the SSE/SZSE e-interaction platform, which the company declined to answer twice; 南方精工's H1 pre-announcement puts net profit attributable to parent at −67% to −77% year on year; 绿的谐波's PS is as high as 114x; and for 卧龙电驱, the company has given no figures for either the stake ratio or the supply relationship. See §7.1 for detail.

⚠️ Also correcting one systematic basis error in this edition's first draft (found by a sub-agent's back-calculation): the "PE-TTM" displayed by most quote software for these names is in fact forward PE (Q1 annualized). Recomputed by summing the four most recent actual quarters, 埃斯顿 goes from 99x to 297.7x (ex-non-recurring TTM 1,724x), 南方精工 goes from 5,981x to 24.3x, and 秦川机床 goes from 121x to 236.5x (ex-non-recurring TTM negative). For 南方精工 and 埃斯顿 the direction flips entirely — any conclusion that ranked on those two numbers would come out backwards. All tables below have been replaced with the recomputed basis.

---

1. News Overview

📐 Two basis notes and data boundaries for this edition (stated up front, so you can judge how far each conclusion can be trusted):

1. Wherever the text says "No. N market-wide" or "highest market-wide," the figure comes from the completed whole-market statistics after yesterday's (8/18) close, not from a re-computation this morning. This morning's whole-market single-stock fund flows could not be fully retrieved and were downgraded to name-by-name spot checks of the 48 candidates in this edition — each stock's own main-force net figure is accurate, but this edition provides no whole-market fund-flow ranking on a this-morning basis.
2. The list of "interim reports disclosed today (8/19)" carries over the exchange's pre-scheduled disclosure calendar obtained yesterday (75 companies market-wide, including 兆易创新, 普冉股份 and 苏垦农发); it was not independently re-verified this morning. All conclusions and verification points involving 苏垦农发 (601952) are premised on this, and if the actual disclosure date changes, that item is automatically void.
3. Scope of the PE recomputation: this edition performed the "sum of the four most recent quarters" TTM recomputation only for the robotics-chain names covered in §3 and §7.1 plus 兆易创新 (reason in §0). PE was not recomputed for the agriculture and oil & gas lines, so those two sections quote no specific multiples.

Scan window: 2026-08-18 15:00 (previous session close) → 2026-08-19 07:15. Anything outside the window is flagged and given a separate "pricing progress" judgment (see §1.1).

1.1 Pricing Progress (the second dimension, alongside the "release window")

This section is a dimension added at the request of yesterday's recap. The release window measures when the news came out; pricing progress measures how many days the market has already digested it. The criteria use yesterday's (8/18) official limit-up pool by industry distribution plus main-force flows, not the news publication date.

1.2 Standing Output: Yesterday's Best Seal Quality and Top LHB Concentration (regardless of branch strength)

This line is a standing output explicitly required by yesterday's recap. Recording it is not recommending it.

The 3 strongest seals yesterday (excluding the "can't get filled" type that sealed on the 09:25 call auction):

The 5 that sealed straight on the 09:25 call auction with seal ratio > 1 (can't be bought, so no return opportunity): 哈森股份 (Hasen, 603958) 3.273, 桂发祥 (Guifaxiang, 002820) 2.987, 学大教育 (Xueda Education, 000526) 2.894, 日丰股份 (Rifeng, 002953) 1.783, 丰原药业 (Fengyuan Pharmaceutical, 000153) 1.553.

Top LHB net buys yesterday (8/18 list, published after 18:00 and back-filled): 中石科技 (Jones Tech, 300684) +RMB 511 million (top buy seat is a main buyer, hot-money structure), 旭光电子 (Xuguang Electronics, 600353) +RMB 440 million (Shanghai money buying), 九州一轨 (Jiuzhou Yigui, 688485) +RMB 261 million, 盈新发展 (Yingxin Development, 000620) +RMB 253 million, 诚迈科技 (ArcherMind, 300598) +RMB 183 million (top buy seat is a main buyer), 神农种业 (Shennong Seed, 300189) +RMB 217 million in total (2 institutional buy seats), 瑞丰高材 (Ruifeng New Materials, 300243) +RMB 222 million in total (5 institutional buy seats; the acquisition was announced last night), 聚和材料 (Juhe Materials, 688503) +RMB 165 million, 金字火腿 (Jinzi Ham, 002515) +RMB 269 million in total (3 institutional buyers), 金禄电子 (Jinlu Electronics, 301282) +RMB 117 million (2 institutional buyers).

Three contrasts that must be named:

1. 金健米业 (Jinjian Cereals, 600127) was yesterday's benchmark 2-consecutive-limit-up name in agriculture, but its LHB net buy was −RMB 16 million (net selling) — yesterday's pre-market it was still the name with "the highest LHB net buy as a share of total turnover market-wide, 29.86%," and in one day it went from the strongest net buy to net selling.
2. 埃斯顿 (Estun Automation, 002747) — the item this edition unpacked in the most detail, and the conclusion is the exact opposite of the surface read "2 institutional buy seats." It closed at the limit-up price with main force (large-order basis) net inflow of RMB 166 million, but LHB net buy of −RMB 162.51 million (buys RMB 557.6 million / sells RMB 720.1 million). The buy side does show 2 "institutional-dedicated" seats, but once all 4 institutional seats on both sides are merged and broken out:

Institutional seats netted RMB 171 million of selling and northbound netted RMB 67 million of selling; the only two net buyers were two classic hot-money seats. Combined with "first seal only at 14:56, then broke, and not in the official limit-up pool," 8/18's 埃斯顿 was a hot-money relay while institutions and northbound distributed in step. Reading only the "2 institutional buyers" line yields the exact opposite conclusion. 3. 农发种业 (China Agri-Development Seed, 600313) had 2 consecutive limit-ups and a top-buy-seat main buyer, but LHB net buy of −RMB 31 million (net selling).

---

2. Strongest Positive Branches, Descending

Supplementary Notes by Branch

Branch 1 (robotics) — why positive, short-term or medium-term, is there an earnings path, is it already fully expected, sector distribution:

• Why positive: today is the intersection of three events (listing day one + conference opening + Release Day), the highest event density of any robotics theme this year. SASAC announcing a Central SOE Robotics Innovation Consortium is a policy-level signal, not a corporate-level one.
• Short-term or medium-term: mainly short-term. The conference runs 8/19–8/23; among the four themed days, 8/19 is Release Day and 8/20 is Procurement Day — "Procurement Day" is the only point that could produce a real order announcement, but it is tomorrow, not today.
• Is there an earnings path: extremely uneven by tier, and one magnitude ceiling caps the entire chain. 宇树's 2025 revenue was only RMB 1.708 billion and its top five suppliers accounted for 22.54% of purchases combined, implying combined purchases from the top five of roughly RMB 140–190 million — for the overwhelming majority of A-share names, "is it a Unitree supplier" is not a meaningful question on the income statement. Reducer / motor / bearing names have real revenue (though the humanoid portion is mostly tiny or undisclosed), while shadow names (0.2%–0.3% stakes) have no revenue path at all. See §7.1.
• Is it already fully expected: partly. Main-force money flowed net into every name in the chain yesterday; 埃斯顿 closed at the limit-up price; 中大力德 touched 87.00 intraday and closed 84.89 (spike and fade); 首开股份 touched 4.50 intraday and closed 4.33 (spike and fade). "Spike and fade" has already appeared, meaning some money is leaving ahead of the event; and 埃斯顿's LHB shows institutions and northbound both net selling on the same day, RMB 171 million and RMB 67 million respectively.
• Sector distribution: mainly SSE/SZSE main boards (卧龙电驱 / 埃斯顿 / 秦川机床 / 双环传动 / 金发科技 / 南方精工 / 襄阳轴承), with the STAR Market (绿的谐波, ±20%) and ChiNext (长盛轴承, ±20%) each represented, and BSE-listed 巨能股份 (Juneng Machinery, 920578, ±30%) up +8.77% yesterday.
• Risk of a one-day high-level trip: high. Today is "Release Day," and the typical A-share pricing of a "conference opening day" is that the high comes on the opening day itself.

Branch 3 (agriculture) — why it stays at No. 3 rather than higher: $90 oil hardened the "imported inflation" leg (energy → fertilizer → agricultural product costs), a new variable that did not exist yesterday. But not one structural defect has been repaired: no anchor name, poor seal quality, a high share of late-session seals, and net LHB selling in the benchmark names. Its problem today is not "is there a logic," but "how much money can it absorb."

---

3. Master Ranking of Single-Stock Catalyst Strength (Descending)

Scoring rules are in §4. Sector and price-limit bands are labeled — a ±20%/±30% name cannot be read with ±10% intuition. "Yesterday's main-force net" is labeled a reference item throughout and is not used as a ranker (this repo has observed for several consecutive days that single-day main-force extremes mean-revert significantly).

---

4. Single-Stock Scoring Model (100 points total)

Scoring example ① (卧龙电驱 600580, §3 total 72):

Scoring example ② (兆易创新 603986, §3 total 38):

This scorecard itself makes one point: earnings elasticity got 12 and fundamentals got 13, both high, and the total is still only 38 — because "expectation gap" and "theme durability" are both near zero while risk deductions are maxed out. Financial quality is not a short-term ranker.

---

5. Detailed Analysis of the Top 10

Coverage: §5.1–§5.10 cover all of the names ranked 1–10 in §3 (隆平高科 and 大北农 are merged into §5.10). The subsection order does not exactly follow the master ranking — names in the same branch are placed adjacently for easier comparison. §5.5 兆易创新 (No. 25 on the master ranking) and §5.11 机器人 (No. 11) are the two that fall outside the Top 10 but must be broken out separately, each with its reason stated in its own subsection.

Blanket note: every "holds a stake in 宇树 / supplies 宇树" statement in this section has been traced item by item to company announcements, the original Q&A on the e-interaction platform, and the wording of 宇树's prospectus by the fundamentals-analyst sub-agent. Anything the company has not confirmed is written as "not obtained" — nothing is presumed or filled in. The order matches §3.

⚠️ One magnitude fact that constrains the entire branch (read this first): 宇树科技's prospectus discloses 2025 revenue of RMB 1.708 billion, top-five suppliers at a combined 22.54% of purchases, and the mechanical suppliers are coded and unnamed in the prospectus (the only named ones are Orbbec and 上海曜励电子). Working back from a net margin of roughly 35%, combined purchases from the top five suppliers are about RMB 140–190 million, and the single largest supplier is most likely under RMB 100 million (this is an estimate, not company disclosure). This ceiling directly falsifies several circulating narratives: 中大力德's "RMB 3.2 billion three-year long-term agreement" averages RMB 1.07 billion a year, which is 1.5x 宇树's entire annual procurement; and 卧龙电驱's "60% supply share," even assuming it were 100% of 宇树's full-year purchases, would be only 5% of 卧龙's TTM revenue of RMB 15.4 billion. Within this branch, "is it a 宇树 supplier" is not a meaningful question for the income statements of the overwhelming majority of these names.

51 双环传动002472watch closely · Shuanghuan Driveline · SZSE main board ±10%

• Related news: the 2026 World Robot Conference opens today (北京经开区³); subsidiary 环动科技 is the domestic No. 1 in RV reducers, with about 25% domestic share in 2024 (GGII; 10.11%/13.65%/18.89% for 2021–2023), and its STAR Market IPO was accepted in November 2024, seeking to raise RMB 1.4 billion.
• Catalyst logic: direct (RV reducers go into humanoid lower-limb joints), but what it affects is valuation expectations, not current revenue.
• Stage of the industry branch: fermenting; it rose only 3.26% yesterday, clearly the laggard within the chain.
• Fundamental verification (the most solid in the chain): 2025 annual report segments — smart actuators RMB 798 million (8.75%, but the bulk is automotive e-drive actuators at a gross margin of only 19.46%, so it should not be counted wholesale as robotics) and reducers and other RMB 795 million (8.73%); the main body is still passenger-vehicle gears at RMB 6.005 billion (65.90%). The company does not separately break out robot reducer revenue. TTM revenue RMB 9.142 billion, attributable net RMB 1.269 billion, ex-non-recurring RMB 1.193 billion, operating cash flow RMB 2.754 billion (OCF/net profit 2.17), TTM ROE about 13.1%, PE-TTM 26.7x.
• Industry position: leader in precision gears plus No. 1 domestic share in RV reducers; its role in the chain is anchor candidate.
• Technicals and sentiment: +3.26%, open 38.86, high 39.94, close 39.91 (closed near the day's high), main force +RMB 79 million, turnover only 5.73% (among the lowest in the chain, not overheated).
• Final judgment: on a track whose average PE is in the triple digits and which just met the highest long-end rates in 19 years overnight, "26.7x PE + real earnings + OCF/net profit 2.17 + not overheated + closed near the high" is the only combination not built out of price gains.
• ⚠️ But it did not receive a "priority deep-dive," for three reasons, each of them hard:
  1. On 2026-08-16 an investor pressed on the e-interaction platform: "Please answer directly whether the company supplies reducers to 宇树科技, Tesla and related companies. Either you do or you don't. Please do not fob shareholders off with 'consult the periodic reports'" — the company did not reply.
  2. 21st Century Business Herald's systematic review of the 宇树 supply chain on 2026-08-06 does not mention 双环传动.
  3. 2026Q1 revenue was only +1.49% and ex-non-recurring −4.04%, so growth is decelerating; and robot reducer revenue has never been disclosed as a separate amount — a low valuation may be the pricing of low elasticity rather than a mispricing. The 8/26 interim report is the only resolution of that uncertainty.

52 卧龙电驱600580watch closely · Wolong Electric Drive · SSE main board ±10%

• Related news: ① the World Robot Conference opens today; ② 宇树科技 (688836) lists today (东方财富¹); ③ the company has established a "Unitree–Wolong joint robot electric-drive laboratory" with 宇树 (21 财经¹⁹).
• Catalyst logic: indirect; there is real segment revenue but it is tiny. The 2025 annual report broke out for the first time "robot components and system applications" at RMB 515.8 million, 3.34% of revenue.
• ⚠️ A widely overlooked contrary signal: that segment's gross margin is 26.21%, below explosion-proof drives at 32.18% and industrial drives at 30.00% — the robot business currently dilutes the company's gross margin rather than lifting it.
• Verification results on the stake and the supply relationship (the part most in need of clarity here):
  • Stake: the company only confirmed on the interaction platform on 2025-09-02 that it "indirectly holds shares in 宇树科技 through 金石投资 (CITIC Goldstone)," giving no ratio whatsoever. Four mutually exclusive values circulate — 0.083% / 0.13% / 0.1525% / 0.1613% — none first-hand confirmed, and none adjusted for the 10% new-share dilution in this IPO. This edition's conclusion: the ratio is not obtained.
  • Supply: "supply share to 宇树 reaches 60%" appears only in self-media with no first-hand source; the original wording from 金融界 / 环球老虎财经 is "according to market rumor, some of the externally purchased joint motors used in 宇树's quadrupeds and humanoids are supplied by 卧龙." And it directly conflicts with 宇树's own public statement — 宇树 says it designs the electromagnetic solutions for the motor stage itself and has built its own production line, and "does not purchase finished motors, motor modules, or semi-finished stators and rotors," buying only raw materials such as copper wire and magnets upstream. If that statement holds, the direct-supply-of-finished-motors path is excluded, leaving only R&D cooperation through the joint laboratory.
• Fundamental verification: TTM revenue RMB 15.400 billion, attributable net RMB 1.134 billion, ex-non-recurring RMB 806 million, operating cash flow RMB 1.656 billion (OCF/net profit 1.46), TTM ROE about 10.6%, PE-TTM 52.2x / ex-non-recurring 73.4x, debt-to-asset ratio 53.91%. ⚠️ Revenue has been negative or flat year on year for 5 consecutive quarters (−5.78% / −6.64% / −13.97% / −1.31%), and 2026Q1 ex-non-recurring was −7.56%.
• Technicals and sentiment: +7.40%, open 35.59, high 38.00, close 37.87 (closed near the day's high), turnover 7.49%, main force +RMB 202 million, turnover value of RMB 4.295 billion, the largest in the robotics chain. No limit-up, so no consecutive-limit-up altitude risk.
• Final judgment: it is the name in the chain with the largest money flow and it does have real robot segment revenue — that is why it sits at No. 2. But three demerits are equally clear: the segment is only 3.34% of revenue and is dilutive to gross margin, the company gives no number for the stake, and the supply claim conflicts with 宇树's own statement. Its interim report is not due until 8/31, the latest in the chain.

53 绿的谐波688017watch closely · Leaderdrive · STAR Market ±20%

• Related news: the World Robot Conference opens today; the company is the A-share "first stock" in harmonic reducers.
• ⚠️ Customer verification result: management has publicly said its share of harmonic reducers for domestic humanoid robots may be about 90%, but the lead customers the company mentions publicly are Agibot and UBTech, not 宇树 (新浪财经²⁰). Supply relationship with 宇树: not obtained; it appears only as a "potential supplier" listing in broker research.
• Catalyst logic: medium-high. Harmonic reducers are the core component of humanoid rotary joints, and industry shipment growth flows straight into its revenue — and its customer mix shows it is already delivering for real, just not to the company listing today.
• Fundamental verification: the 2025 annual report's "industrial and embodied-intelligence robot components" was RMB 423 million, 74.04% of revenue (on a product basis, harmonic reducers and metal parts were RMB 476 million / 83.49%, gross margin 36.77%). ⚠️ Basis note: the same segment was named "industrial and service robot components" in the 2024 annual report and was only renamed to add "embodied intelligence" in 2025 — that is a change in disclosure wording, not evidence of humanoid revenue.
  • TTM: revenue RMB 612 million, attributable net RMB 137 million, ex-non-recurring RMB 106 million, operating cash flow RMB 141 million (OCF/net profit 1.03), debt-to-asset ratio only 9.44% (cleanest in the chain), TTM ROE about 3.9%.
  • 2026Q1: revenue RMB 140 million (+42.96%), attributable net RMB 32.63 million (+61.17%), but gross margin of 33.62% was down year on year and quarterly operating cash flow was only about RMB 5.5 million — cash conversion badly lagged net profit that quarter.
• Valuation (the most extreme in the chain, and the extremity is on the revenue side): market cap RMB 69.8 billion / TTM revenue RMB 612 million = PS 114x; PE-TTM 510.5x / ex-non-recurring 656.7x, PB 19.77. A PS of 114x means that even at a 30% net margin, revenue would have to grow more than 10x to land at a 40x PE.
• Technicals and sentiment: +5.14%, main force +RMB 221 million, the largest net inflow in the chain; open 362.00, high 385.37, close 380.80, no limit-up. STAR Market ±20%, so the theoretical single-day swing is twice that of the main board.
• Final judgment: robot revenue at 74% of the total is the second-highest share in the chain, and that is real; a PS of 114x is also real. The 30-year Treasury at 5.31% overnight is the highest since 2007 — mathematically, a rising discount rate does far more damage to a 510x PE than to a 26.7x PE, which is exactly the mechanism behind last night's −7% to −9% in the US memory chain. The 8/27 interim report is the first real test of that 114x PS.

54 神农种业300189watch closely · Shennong Seed · ChiNext ±20%

• Related news: the J.P. Morgan food-crisis report (published 8/16, outside the scan window, but already at pricing day 3).
• Catalyst logic: indirect leaning medium; what it affects is valuation expectations, not current revenue — the report is a forecast about 2027 and changes none of this quarter's statements.
• Stage of the industry branch: climax turning to divergence. The agriculture-food chain contributed 28 of yesterday's 79 limit-ups, but 7 of them sealed only after 14:25.
• Fundamental verification: free float RMB 5.52 billion, PE negative. A niche seed-industry name.
• Industry position: one of the market benchmarks of this agriculture move (its 44.12% turnover was among the highest market-wide yesterday).
• Technicals and sentiment: +20.04% limit-up, turnover 44.12%, first sealed 10:21, 1 break, seal ratio only 0.044, main force +RMB 277 million; the LHB shows 2 institutional buy seats totaling +RMB 217 million net — the only agriculture name with substantive institutional net buying.
• Final judgment: institutional net buying is the most substantive thing separating it from the other agriculture names (yesterday's recap explicitly required back-filling the LHB to judge "institutional accumulation or hot-money relay," and the answer is there are institutions). But 44% turnover + seal ratio 0.044 + ChiNext ±20% stacked together put the risk dial extremely high. "Institutions are present" does not mean "you can chase it."

55 兆易创新603986Passthe item this edition most needs to spell out · GigaDevice · SSE main board ±10%

This section was reconciled item by item by the fundamentals-analyst sub-agent, with all data taken from the original 巨潮 / SSE announcements, including the 178-page full interim report, the Q1 report, the 7/10 pre-increase announcement, the 6/30 risk-warning announcement, the abnormal-movement announcements and the buyback plan. It also corrected two basis errors in this edition's first draft, both flagged below.

• Related news: the 2026 interim report was disclosed the evening of 8/18 — revenue RMB 11.56576 billion (+178.67%), attributable net RMB 6.85679 billion (+1091.50%), ex-non-recurring RMB 4.88303 billion (+796.90%) (interim report full text⁹).

(1) Line-by-line reconciliation against the pre-announcement: all three land within ±0.7%, no expectation gap in either direction

Metric	2026-07-10 pre-increase announcement	8/18 interim actual	Deviation
Revenue	RMB 11.500 billion (+177%)	RMB 11.56576 billion (+178.67%)	+0.57%
Net profit attributable to parent	RMB 6.900 billion (+1099%)	RMB 6.85679 billion (+1091.50%)	−0.63%
Ex-non-recurring attributable net	RMB 4.850 billion (+791%)	RMB 4.88303 billion (+796.90%)	+0.68%

Table scrolls horizontally

⚠️ Correction: both the exchange disclosure date and the announcement date of the pre-announcement were 2026-07-10 (announcement number 2026-054); the previously circulating "July 9" is wrong.

(2) The hardest pricing fact is not "how much the stock fell" but "which way it moved the day the pre-announcement landed"

Date	Close	Change	Event
06-29	840.00	+9.09%	Peak of the range, total market cap RMB 589.5 billion
06-30	815.00	−2.98%	Company-initiated risk-warning announcement (self-declaring the price top)
07-10	612.00	−7.76%	The day the pre-increase announcement (+1099%) landed
07-13	550.80	−10.00%
07-16 / 17	514.61 / 463.15	−6.57% (versus 7/13, spanning the two unlisted sessions 7/14–7/15) / −10.00%	Triggered the 7/21 abnormal-movement announcement
07-27	434.03	−5.03%	The day 长鑫科技 (CXMT, 688825) listed
07-28 / 29	390.63 / 364.03	−10.00% / −6.81%	Triggered the 7/30 abnormal-movement announcement
08-03	340.74	−10.00%	Low of the range
08-18	436.74	−1.64%	Interim report date, intraday high of 455.81 leaving a long upper shadow, main force net outflow RMB 1.615 billion (4th largest market-wide)

Table scrolls horizontally

A +1099% pre-announcement fell 7.76% on the day it was disclosed; from the pre-announcement date 7/10 (612.00) to the 8/3 low (340.74) the cumulative decline was −44.32%, including 4 limit-downs (7/13, 7/17, 7/28, plus 1 inside the unlisted 7/14–7/15 window); from the 6/29 peak to 8/18 the cumulative decline is −48.01%.

⚠️ The table above is a selection of key nodes, not a continuous session series (7/11–7/12, 7/14–7/15, 7/20–7/24, 8/04–8/14 and others are not listed). You cannot back out the change between two adjacent rows by dividing their prices. Price moving against the information after it became public shows expectations had long exceeded the facts. But the other side must also be stated: from the 8/3 low of 340.74 to 8/18 it has rebounded +28.2%, so there was a round of front-running in the two weeks before the interim report.

(3) Account-level breakdown of the RMB 1.97375 billion gap between attributable net and ex-non-recurring net — and it is precisely the part that should not be priced

Page 7 of the interim report shows that RMB 2.23140 billion (pre-tax) of that gap comes from fair-value changes and disposal gains on financial assets, with government grants at only RMB 8.58 million. Drilling down to the security level (interim report p.18):

Security	Listing date	Fair-value change taken to P&L	Share
联讯仪器 (Lianxun Instruments, 688808)	2026-04-24	RMB 1.26273 billion	58.4%
臻宝科技 (Zhenbao Technology, 688797)	2026-06-24	RMB 511.93 million	23.7%
强一股份 (Qiangyi, 688809)	2025-12-30	RMB 327.70 million	15.2%
中科仪 (KYKY Technology, 920186)	2026-04-29	RMB 43.76 million	2.0%
曦智科技 (Lightelligence, 01879.HK)	—	RMB 16.72 million	0.8%
Subtotal (five securities)		RMB 2.16284 billion	100%
Others (wealth-management products, FX derivatives, etc.)	—	about RMB 69 million	—
Total		RMB 2.23140 billion	—

Table scrolls horizontally

(The denominator of the share column is the RMB 2.16284 billion five-security subtotal, not RMB 2.23140 billion.) Of that, RMB 1.818 billion (84.1% of the five securities) comes from one-off IPO revaluation of three associates newly listed within 2026H1 (臻宝科技 listed 6/24 and RMB 512 million of paper gain was recognized by 6/30, six trading days apart). This is a one-off, non-cash, non-repeatable gain.

(4) The clarification of the 长鑫 accounting treatment is true and is cross-confirmed by three independent pieces of evidence in the interim report — but the conclusion is the opposite of market intuition

The company holds 1,085,672,415 shares of 长鑫科技 = 1.80%, at a cost of RMB 2.300 billion, with a 6/30 fair value of RMB 10.062 billion and cumulative fair-value changes taken to equity of RMB 7.762 billion; in the consolidated income statement that account sits under other comprehensive income, in the "not to be reclassified to profit or loss" line (interim report p.24 / p.52). That is: not one cent of 长鑫's RMB 7.762 billion paper gain has ever entered net profit, and none of it ever will. And 长鑫科技 already listed on the STAR Market on 2026-07-27 — the shadow-stock logic was settled in its listing week by way of a "cumulative three-day decline deviation exceeding 20%" (the 7/30 abnormal-movement announcement). "长鑫 listing" is neither the source of this RMB 1.97 billion nor still an unrealized expectation.

(5) Valuation formation: low PE percentile + high PB percentile, the textbook shape of a cyclical earnings peak

⚠️ Correction (the single most important basis fix in this edition): the "PE 22.35x" shown by quote software is not TTM but annualized interim results (RMB 306.48 billion ÷ RMB 13.714 billion).

Basis	PE	3-year percentile
H1 annualized (commonly displayed by quote software)	22.35	0.0%
True rolling TTM (updated through 26H1)	38.65	0.0% (below the three-year minimum of 39.18)
Ex-non-recurring TTM	52.77	0.7%
PB (8.22x)	—	81.7% (82.7% over 5 years)

Table scrolls horizontally

A PE percentile of 0% against a PB percentile of 82% is divergence taken to the limit — the low PE comes from the denominator (E) being at a peak, not from the numerator being cheap. Stripping out RMB 7.691 billion of other comprehensive income lifts PB to 10.36x; on an SOTP basis, after carving out RMB 16.020 billion of external investments, the core business is at about 13.0x PB and an implied PE of about 29.9x (and that still uses cycle-peak earnings annualized). For reference: full-year 2025 diluted ROE was only 8.67%, whereas the sustainable ROE implied by a PB of 8.22 is far above that.

(6) Financial quality itself is beyond reproach — which is exactly why the test of "realization" has to fall elsewhere

Q2 quarterly ex-non-recurring RMB 3.473 billion (+146.30% quarter on quarter), quarterly gross margin 66.57% (+9.49pp quarter on quarter); H1 net operating cash flow RMB 6.048 billion (+531.47%), OCF / ex-non-recurring attributable net = 1.24; cash-collection ratio 1.035 (flat against 1.039 a year earlier); inventory grew only 36.47% while revenue grew 178.67%, and inventory days fell from 177 to 153; receivable days 4.89; debt-to-asset ratio 11.47%, with zero interest-bearing debt at period end. "Is there a divergence between inventory appreciation and cash collection?" — clear answer: there is none. But memory chips are already 85.0% of revenue at a 67.57% gross margin, so the revenue mix is tightly bound to a single cyclical product currently at a price peak.

(7) The strongest contrary evidence comes from the company itself

The 6/30 risk-warning announcement (巨潮 2026-052²¹) reads: "Product prices are already at historic highs, and the trend of continued sharp increases is not sustainable... downstream demand has already been suppressed to a degree. As marginal capacity is added in the niche memory market, prices will subsequently fall considerably."

The 8/18 interim report neither withdrew that judgment nor softened it — it merely compressed it into a milder passage of risk language on p.23, while the "outlook" chapter says nothing about prices at all.

(8) Layered on top are three overnight external negatives: ① the US memory chain, SNDK −9.01% / STX −9.16% / WDC −7.43% / MU −7.02%; ② the 30-year Treasury at 5.31%, the highest since 2007; ③ main force net outflow of RMB 1.615 billion in this stock yesterday, 4th largest market-wide.

• Final judgment: explosive numbers ≠ an expectation gap; this is a second realization. An interim report that deviates less than 0.7% from the pre-announcement has near-zero variance in information content; the only item that could be misread as a "new positive," the RMB 1.97 billion, is precisely a one-off IPO revaluation; and the valuation formation gives a cycle-top signal rather than an undervaluation signal. This edition sets no recommendation on the memory line.
• ⚠️ Falsification conditions for this judgment (must be stated): overturning the "realization" conclusion requires new information from outside the interim report — ① whether spot quotes for niche DRAM (DDR4 4Gb/8Gb) and SLC NAND have kept rising since 7/1; ② what management says at today's results briefing about Q3 sequential prices and second-half gross margin; ③ the shareholder-meeting approval and actual execution progress of the RMB 1–2 billion buyback (capped at RMB 750 per share, for cancellation). The interim report answers none of these three, and this edition has not obtained them either.
• Same-day accompanying information (recorded, no conclusion drawn): the company simultaneously announced a change of board secretary — the incumbent 董灵燕, whose term ran to 2027-12-16, left early for "personal career planning," with the announcement stating there was no disagreement with the board; no interim dividend was paid for this reporting period.

Supplementary note (must be given to clients): the trigger for the overnight memory-chain plunge was the Wall Street Journal report on roughly $3 trillion of off-balance-sheet AI commitments across nine tech giants, and that report contains nothing negative about memory supply or demand; the 24/7 Wall St. piece explicitly states "there was no memory-industry-specific bad news today." This means that if the A-share memory chain follows lower today, it is transmission of a valuation/financing narrative, not evidence of a turn in demand. The subsequent paths of the two are completely different and must not be conflated.

56 埃斯顿002747watch only · Estun Automation · SZSE main board ±10%

• Related news: the World Robot Conference opens today; domestic leader in industrial robot bodies.
• Catalyst logic: the most direct business basis in the chain. The 2025 annual report's "industrial robots and smart manufacturing systems" was RMB 3.997 billion, 81.77% of revenue (gross margin 29.23%), with core automation components at RMB 891 million (18.23%) and total revenue of RMB 4.888 billion. This is the largest robot revenue base in the chain.
• ⚠️ But neither the "humanoid" layer nor the "宇树" layer holds: on the e-interaction platform on 2026-05-27 the company said "the company's associate 埃斯顿酷卓 (Estun Codroid) conducts humanoid robot business" — that is at the associate level, not consolidated and with no humanoid revenue disclosed; relationship with 宇树: not obtained, with no confirmation of any kind on the e-interaction platform.
• Stage of the industry branch: divergence has appeared.
• Fundamental verification (a passage that runs counter to the "leader" intuition): PE-TTM 297.7x, ex-non-recurring TTM as high as 1,724x (the previously circulating "99x" is the Q1-annualized basis, not TTM); 2026Q1 attributable net was RMB 97.8367 million while ex-non-recurring was only RMB 19.3586 million.
  • 2026 interim pre-announcement (7/15): attributable net RMB 150–180 million (+2144.74% to +2593.68%), but ex-non-recurring only RMB 60–75 million. The roughly RMB 90–105 million difference comes from "completion of the asset restructuring of associates 南京工艺 (Nanjing Gongyi) and 南京化纤 (Nanjing Chemical Fibre), with the stake swapped into listed-company equity and the fair value of that equity increasing" — non-recurring, non-cash and non-sustainable.
  • Legacy baggage: a RMB 744 million loss in 2024Q4 alone; the debt-to-asset ratio fell from 82.38% to 69.18% but is still in the highest tier of the chain.
• Industry position: leader (among the top domestic industrial robot body shippers).
• Technicals and sentiment (the most critical passage): it closed at RMB 40.04 yesterday = the limit-up price, up 10.00%, but it first sealed only at 14:56 and then broke, so it is not in the exchange's official limit-up pool (yesterday's official pool checked out with tc=79 and len(pool)=79, and this stock is not among them). Judging by "close = limit-up price" on your own would mistakenly read it as a clean limit-up name.
  • Fund conclusion after breaking out the seats (see §1.2): main force (large orders) net inflow +RMB 166 million, while LHB net buy was −RMB 162.51 million; 4 institutional-dedicated seats netted RMB 171 million of selling and Shenzhen Connect netted RMB 67 million of selling, with the only two net buyers being hot-money seats.
• Final judgment: late-session seal + broken seal + institutions and northbound distributing in step + profit resting on equity fair value — all four hold at once. It is the name with the hardest business in this chain and also the dirtiest tape yesterday. Rated "watch only": the business is worth tracking long-term, but yesterday's tape does not support chasing. The 8/22 interim report is the nearest real test.

57 石化机械000852watch closely · SINOPEC Oilfield Equipment · SZSE main board ±10%

• Related news: Brent closed above $90 for the first time in three weeks, up about 3% intraday; the US–Iran 60-day negotiating window expired 8/17 and Trump refused to extend (新浪财经⁸).
• Catalyst logic: indirect and lagged. Oil price → oil-company capex → oilfield equipment orders normally takes 1–2 quarters and does not affect this quarter's revenue.
• Stage of the industry branch: launch phase (pricing day 1) — its only advantage over agriculture (day 3) and robotics (event day).
• Fundamental verification: free float RMB 5.50 billion, PE negative (the company is loss-making). The positive cannot convert into profit in the near term.
• Industry position: the SINOPEC-system oilfield-services equipment platform; its role in the chain is an elastic name.
• Technicals and sentiment: +10.08% limit-up, first sealed 09:48 (morning-session consensus, not a late-session chase), 0 breaks, seal ratio 0.247, main force +RMB 47 million.
• Final judgment: "morning seal + zero breaks + pricing day 1" was the cleanest tape in yesterday's oil & gas chain. But the company is loss-making and the chain is long-lagged, so it can only be rated "watch closely," no higher.

58 天山生物300313watch only · Tianshan Bio · ChiNext ±20%

• Related news: the food-security main line.
• Catalyst logic: indirect (livestock breeding, further from grain prices than seeds are).
• Stage of the industry branch: divergence point (2 consecutive limit-ups).
• Fundamental verification: free float only RMB 2.54 billion, PB 84.06 — the highest on this list.
• Industry position: far more a themed name than an industrial one.
• Technicals and sentiment: +20.04% limit-up (2nd consecutive), first sealed 09:31, 0 breaks, seal ratio 0.522, best in the agriculture group, LHB shows 1 institutional buyer +RMB 30 million.
• Final judgment: yesterday's recap designated it the criterion stock for "can agriculture produce the market's first 3rd consecutive limit-up" — a criterion stock is not a recommended stock. PB 84x + a RMB 2.54 billion float + a ±20% band puts the risk dial too high. Its value today lies in the answer it will give us about the agriculture line, not in itself.

59 中曼石油603619watch closely · Zhongman Petroleum · SSE main board ±10%

• Related news: Brent closed above $90 for the first time in three weeks, up about 3% intraday; the US–Iran 60-day negotiating window expired 8/17 (新浪财经⁸).
• Catalyst logic: the most direct name in this branch. The company owns oilfields (the Wensu block), so rising oil prices flow straight into oil and gas sales revenue with no transmission lag — fundamentally different from 石化机械 (equipment orders, lagged 1–2 quarters) and 海油工程 (engineering orders).
• Stage of the industry branch: launch phase (pricing day 1), and it has barely been priced at all.
• Fundamental verification: free float RMB 10.51 billion, turnover 7.00%. Note: earnings in the oil & gas E&P sector are highly non-linear in the oil price, and this edition did not obtain unit costs or output for its oil and gas segment, so no quantified conclusion on elasticity is given.
• Industry position: one of the few names combining private oilfield services with owned oilfields; its role in the chain is a niche name.
• Technicals and sentiment: on 8/18 it was only +0.80%, high 23.18, close 22.74, main force +RMB 42 million. On a day when oil & gas and refining engineering rose 3.38%, oilfield-service engineering rose 2.32% and 石化机械 hit the limit-up, it basically did not move.
• Final judgment: the combination "hardest logic + least movement yesterday" cuts both ways. Either it has not yet been priced (opportunity), or money judges its oilfield elasticity to be below expectations (risk). These two readings can be told apart within the first 30 minutes of today's session — see the criteria in §8.

510 隆平高科000998and 大北农 (Dabeinong, 002385, SZSE main board ±10%) — both watch only · Longping High-Tech · SZSE main board ±10%

• Related news: the J.P. Morgan food-crisis report continues to ferment; commercialization of GMO corn / soybean varieties (in March 2026 the Ministry of Agriculture and Rural Affairs approved 27 GMO corn varieties and 3 GMO soybean varieties).
• Catalyst logic: these two are the pair with the most direct products in the agriculture branch — 隆平高科 leads hybrid rice and corn seed, and 大北农 is one of the GMO trait leaders. Rising grain prices → seed demand and pricing power is the shortest transmission path on this chain.
• Stage of the industry branch: pricing day 3, a divergence point.
• Fundamental verification: both have negative PE (loss-making), with free floats of RMB 12.12 billion and RMB 12.05 billion respectively.
• Technicals and sentiment (the sole reason they are pressed down to "watch only"):
  • 隆平高科 +10.02% limit-up, but first sealed only at 14:40, seal ratio 0.177, main force +RMB 110 million;
  • 大北农 +10.10% limit-up, but first sealed only at 14:50, seal ratio only 0.026 — essentially no seal orders, main force +RMB 201 million, LHB shows 1 institutional buyer +RMB 40 million.
• Final judgment: the two with the hardest products happen to be the two whose tape looks most like a late-session chase. Yesterday, "stocks that hit the limit the previous day" gapped up +2.33% on average but went −1.12% from open to close, and the late-session sealers were the main contributors to that shape. ⚠️ 隆平高科 discloses its interim report on 8/21, a two-way risk.

5.11 (Outside the Top 10 but broken out separately) 机器人 (SIASUN Robot & Automation, 300024, ChiNext ±20%) — watch only

• Related news: the World Robot Conference opens today.
• Why it is broken out despite being outside the Top 10: its defining feature is "trading recognizability" — the company's short name literally is "Robot," which makes it the natural first stop for money in a conference-driven move. It rose 6.45% yesterday with main force +RMB 186 million.
• Fundamental verification: PE negative (the company is loss-making), free float RMB 26.08 billion. A veteran domestic robotics firm whose profitability has been under sustained pressure in recent years.
• Final judgment: being loss-making dictates that it can only be a sentiment name. High recognizability is double-edged: money enters fastest and exits fastest. Its role today is closer to a "thermometer" — if even it does not rise, the conference opening is simply not being treated by money as a reason to trade.

---

6. Pass List

---

7. Intra-Branch Ranking

7.1 Robotics / Humanoid Robots (today's strongest branch)

"Concept ≠ beneficiary" grading note: the "directness of benefit" column in this table follows evidence strength strictly across four tiers — direct order positive > industry-trend positive (has a product, no named order) > indirect via supply chain > pure concept mapping (a 0.x% stake). This edition obtained no first-hand evidence of a signed supply contract between any A-share name and 宇树, so none enters the top tier.

Tier 1 · Robot-related revenue the company itself breaks out in its periodic reports (ranked by amount and evidence strength):

Tier 2 · Essentially "shadow names holding a 0.x% stake" (ratios recomputed on a uniform basis, including dilution from the 10% new shares issued in the IPO):

⚠️ Dilution basis note: 首开 0.270% → 0.243% and 金发 0.249% → 0.224%, both computed as "post-issue ×90%" (40.4464 million shares issued, post-issue total share capital 404 million shares); the base is the fund's 4.1520% stake per the registration draft of 宇树's prospectus, multiplied by each company's subscribed fund share. For the 卧龙 row there is no company-disclosed fund share as a base, so no number is given.

⚠️ The three circulating ratios are not even internally consistent. All three sit in the same fund (the 金石成长 growth fund), and the ratios among the rumored values are 1 : 0.797 : 0.304, while the ratio of their fund shares is 1 : 0.922 : 0.469 — the rumored numbers cannot be reproduced by any point-in-time combination of "fund share × fund stake." And 金发科技's ratio has been falling all along (0.42% in 2025-02 → 0.32% in 2025-09 → 0.25% in 2026-03), as the fund kept raising capital and diluting LP shares while successive 宇树 financing rounds diluted the fund's stake. 0.4229% is a figure from a year and a half ago.

Tier 3 · Pure concept mapping (no robot revenue + relationship unconfirmed by the company):

• 襄阳轴承 (000678) — the sentence "joint development with 宇树 of a fit for the Unitree H1, currently in sample validation" originates from a question posted by an investor on the e-interaction platform on 2026-08-04, ending with "is the above true?", which the company did not answer; asked again on 8/13 whether it has any cooperation with 宇树, the company again did not answer. This is investor-authored content that the market has treated as company disclosure.
• 南方精工 (002553) — no robot segment; asked on the e-interaction platform "can the precision components be used in humanoid robots?" and "progress on planetary roller screws," it answered "see the periodic reports" both times.
• 中大力德 (002896) — there is a real industrial reducer business, but every claimed link to 宇树 (the RMB 3.2 billion long-term agreement / 63% of purchases / a 10% indirect stake via Shenzhen Capital Group) has zero company confirmation, and it is directly falsified by its own 2026Q1 revenue of +5.12% and ex-non-recurring of −43.19%; four 宇树-related questions on 2026-06-09 plus one on 8/04 were all unanswered on the e-interaction platform.

Composite ranking:

Four sentences to conclude this branch:

1. Hardest single name: 双环传动 (002472) — a 26.7x PE + OCF/net profit 2.17 + ROE 13.1% make it the only valuation in the chain not propped up by concept. But the company refused to answer the 宇树 supply question, robot revenue has never been separately disclosed, and 2026Q1 ex-non-recurring was −4.04%, so it did not get a "priority deep-dive" either.
2. Highest trading recognizability: 埃斯顿 (002747) and 机器人 (300024) — the former has the dirtiest tape (late-session seal + broken seal + institutions and northbound distributing in step), the latter is loss-making. The two most recognizable names happen to be the two you cannot chase.
3. The four most absurd valuations (by "market cap ÷ real earnings power"): 绿的谐波 PS 114x → 埃斯顿 ex-non-recurring TTM 1,724x → 中大力德 PE 296 / ex-non-recurring 404 with ex-non-recurring falling four straight quarters → 秦川机床 PE 236 with TTM ex-non-recurring negative.
4. The full divergence in one sentence: the reasonably valued ones have no story (双环 refused to answer, 金发 has no robot business at all), and the ones with a story have no earnings (绿的谐波 PS 114x, 埃斯顿 ex-non-recurring 1,724x, 襄阳轴承 6 consecutive loss-making quarters). That is the entire reason this edition sets no "priority deep-dive."

⚠️ A timing fact that determines the entire pricing basis of this branch today: none of the 10 names in Tiers 1 and 2 above has disclosed its 2026 interim report (rows 11–15 of this table — 长盛轴承, 机器人 300024, 巨能股份 and the three large-cap watch slots — were not within the scope of the scheduled-disclosure list check). The schedule is: 8/22 埃斯顿, 8/25 金发科技 / 首开股份, 8/26 双环传动 / 中大力德, 8/27 绿的谐波 / 襄阳轴承, 8/29 南方精工 / 秦川机床, 8/31 卧龙电驱. That is: all of today's pricing of "robot revenue scaling up" rests on 2026Q1 or 2025 annual figures, and the latest hard data lands in a cluster from 8/22 to 8/31. None of 绿的谐波, 中大力德 or 双环传动 issued an earnings pre-announcement, so whether humanoid revenue is materializing will be settled 8 days from now.

7.2 Agriculture / Food Security

⚠️ Basis note: this section and §7.3 quote no specific PE multiples. This edition performed the "sum of the four most recent quarters" TTM recomputation only for the robotics chain and 兆易创新 covered in §3 and §7.1; PE was not recomputed for the agriculture and oil & gas lines, and the default display in quote software has been confirmed by this edition to be forward PE (Q1 annualized), which must not be quoted directly, let alone used for cross-stock comparison.

Conclusion: the hardest names are 隆平高科 and 大北农 (most direct products), the most recognizable is 神农种业, and the two sets do not overlap — the two with the hardest products both sealed late in the session, while the two with the best tape (天山生物, 金健米业) have the softest products. This line's structural problem has not improved by one day. And there is still not a single agriculture name market-wide that qualifies as an anchor (the largest turnover, 北大荒 at RMB 1.349 billion, is an order of magnitude below tech names).

7.3 Oil Price / Inflation Chain

Conclusion: the hardest name is 中曼石油 (owns oilfields, oil price flows straight into revenue) and the most recognizable is 石化机械 (the only limit-up), and the two do not overlap. 海油工程 shows a price-volume divergence, Pass tier. The whole chain produced only 1 limit-up yesterday, with no ladder.

---

8. Today's Opening Verification Signals

Every verification point in this section is written in the falsifiable form "is X > threshold," with the read moment stated. Compound formulations that depend on a prior condition holding are forbidden. ⚠️ Read-moment discipline: yesterday's recap taught that "the same criterion gives opposite conclusions 5 minutes after the open versus at the close." Every criterion in this edition that involves cross-market or whole-day conclusions is assigned a closing read; only tape-structure criteria use morning reads, and those are explicitly marked as unusable for whole-day conclusions.

8.1 Call-Auction Signals (09:15–09:25, read moment: end of the 09:25 call auction)

8.2 Sector Signals (read moments: ① 10:30 ② the close, judged separately)

8.3 Single-Stock Signals

8.4 Risk Signals (any one triggering downgrades all of today's watch tags)

---

9. Final Recommendation Conclusions

① The 5 Stocks Most Worth Watching Today

⚠️ All five are "watch closely"; this edition sets no "priority deep-dive." "Worth watching" means they will produce information today, not that they are worth buying today.

⚠️ This table is not the top five of the §3 master ranking; it is "one representative from each of the three branches + the names that can produce the most information today." 神农种业 ranks No. 9 with 60 points on the §3 master ranking — it is included because it is the agriculture line's only evidence point of substantive institutional net buying today, not because its score is high. Its score is held down by §4's hard constraints: PE negative → both fundamentals and earnings elasticity are capped at 8 points, on top of risk deductions for 44.12% turnover, a seal ratio of 0.044 and 2 consecutive limit-ups.

② Today's 3 Strongest Branches

③ Directions Not to Chase Today + Reasons

1. Memory / semiconductor design (兆易创新 603986, 长鑫科技 688825, 通富微电 002156) — four negatives in resonance: ① the US memory chain was −7% to −9% overnight; ② the 30-year Treasury is at its highest since 2007; ③ 兆易创新's interim report deviates less than 0.7% from the 7/10 pre-announcement on all three metrics, i.e. zero expectation gap, and the stock already saw main force net outflow of RMB 1.615 billion yesterday (4th largest market-wide); ④ the company itself wrote in its 6/30 risk-warning announcement that "product prices are already at historic highs... and will subsequently fall considerably," the interim report did not withdraw that judgment, and the "outlook" chapter says nothing about prices at all.
2. 宇树 "shadow names" (首开股份 600376, 深信服 300454, 金发科技 600143) — recomputed on a uniform basis from the companies' own statements, post-issue look-through stakes are only 0.22%–0.24%, with no observable impact on profit whatsoever; the core businesses of 首开 and 深信服 are real estate and network security respectively, with zero business link to robotics; and on material supply 金发科技 explicitly said it is "not convenient to disclose on commercial confidentiality grounds" (i.e. unconfirmed). 首开股份's H1 pre-announcement guides to a loss of RMB 1.7–2.2 billion against attributable equity of only RMB 7.374 billion — the only name in the chain with solvency pressure.
3. The three whose narratives are unconfirmed by the company and are falsified by the company's own financials or by silence on the interaction platform (中大力德 002896, 襄阳轴承 000678, 南方精工 002553) — see §6. Among them, 襄阳轴承's "joint development of the Unitree H1 with 宇树" originates from an investor's question that the company declined to answer twice, the most textbook case this edition found of "a question being treated as disclosure."
4. Yesterday's high-level consecutive limit-ups and late-session sealers — 神奇制药 (600613) on 4 consecutive limit-ups but with a seal ratio of 0.165; 旭光电子 (600353) on 3 but with 10 breaks; 中石科技 (300684) on 3 but with 7 breaks, a seal ratio of 0.018 and a serious abnormal-movement flag already triggered; 一鸣食品 (605179) sealed 14:56 with a seal ratio of 0.001; 国投丰乐 (000713) sealed 14:56 with a seal ratio of 0.012. Those seals were tested into existence repeatedly, not built on real absorption.
5. Auction-locked limit-ups (日丰股份 002953 had a seal ratio of 1.783 yesterday on all-day turnover of only RMB 63 million; 骏成科技 301106 will most likely be locked today) — a limit-up you cannot buy produces no return and should not be counted in any expectation.
6. The lithium / nonferrous chain with strong interim results but a falling share price first (多氟多 002407, 天华新能 300390, 融捷股份 002192, 株冶集团 600961) — results and share prices point in opposite directions, and this edition has not yet traced these companies' first pre-announcement disclosure dates. Until it is confirmed "whether these numbers were already given in a pre-announcement," no recommendation is given.
7. Optical communications / CPO (中际旭创 300308) — the WSJ report on "$3 trillion of off-balance-sheet AI commitments" that hit memory overnight targets exactly the same AI capex narrative that optical-module valuations depend on; and this stock saw main force net outflow of RMB 2.761 billion yesterday, the largest market-wide.

④ Final One-Sentence Judgment

Today is the highest-event-density day of the year for the robotics theme, and also the day it is most likely to see its high intraday — all three events were settled before the open, and in A-shares a "certain event" is usually priced the instant the bell rings. More importantly, here is the sentence that came out of the sub-agent verification: on this chain, the reasonably valued ones have no story (双环 refused to answer the 宇树 question, 金发 has no robot business at all), and the ones with a story have no earnings (绿的谐波 PS 114x, 埃斯顿 ex-non-recurring TTM 1,724x, 襄阳轴承 6 consecutive loss-making quarters with negative operating cash flow). So this edition sets no "priority deep-dive," demotes the hottest name 埃斯顿 (002747) to "watch only," and demotes 中大力德 (002896), 襄阳轴承 (000678), 南方精工 (002553) and 首开股份 (600376) to Pass — not because they have no story, but because their stories are either falsified by the companies' own quarterly reports or have never been confirmed by any company at all. If the robotics chain still has fewer than 3 limit-ups by 10:30 today, this line should immediately be downgraded from "strongest branch" to "event-driven pulse"; and this line's real answer is not today, it is in those ten interim reports from 8/22 to 8/31.

---

⚠️ Risk warning: this list is pre-market information mapping and observation only and does not constitute investment advice. A-share volatility risk is extremely high, and automatically generated content may contain timeliness gaps or errors in supply-chain mapping, so it must not be used directly as a basis for trading.

Data-Retrieval and Run Log (not sent to clients)

1. Execution on a full-universe basis (guarding against one-sided pagination bias)

• 8/18 official limit-up pool: declared tc=79, actually retrieved len(pool)=79, check passed. Every limit-up determination in this edition uses the official pool and does not screen on "close = limit-up price" — which is exactly why it identified that 埃斯顿 (002747) closed at the limit-up price yet is not in the pool (it is in the broken-seal pool, first seal 14:56).
• Concept boards: declared total=504, actually retrieved 504, complete.
• Industry boards: declared total=496, actually retrieved 496, complete.
• 8/18 LHB: ak.stock_lhb_detail_em returned 86 rows, back-filled (the list had not yet been published when yesterday's recap was generated at 15:30; §6.4 item 4 required back-filling today, now done). 埃斯顿's seat breakdown comes from this table.

2. Data-retrieval failures

• ⚠️ Whole-market single-stock fund flows (clist pagination, declared total≈5,560) timed out and did not complete, so it was downgraded to calling the single-stock endpoint name by name, covering this edition's 48 candidates. Consequence: this edition has no "market-wide main-force net inflow/outflow TOP N" table, and the main-force nets in §1.2 and in the various tables are spot checks, not whole-market rankings. This does not affect any single-stock conclusion (the per-stock nets are themselves accurate), but it means this edition cannot claim "this stock is No. N market-wide." Wherever the text cites a ranking such as "4th market-wide," it comes from the full dataset already obtained in yesterday's recap, not from a self-computation this morning.
• ⚠️ East Money's RPT_PUBLIC_BS_APPOIN query for the 8/19 scheduled-disclosure list returned 0 rows (the response field names differ from yesterday's; the SECURITY_CODE column does not exist). So this edition's "interim reports today" list carries over the result already obtained in yesterday's recap (75 companies market-wide on 8/19, including 兆易创新, 普冉股份 and 苏垦农发), without independent re-verification this morning.
• push2.eastmoney.com was intermittently unstable; all requests have a push2delay.eastmoney.com fallback added.
• Search engines returned large amounts of same-period 2025 content for queries like "August 19 morning news" (e.g. "Shanghai Composite 3727.29"); this was identified and discarded wholesale, and none of its market figures were adopted.

3. Data traps caught (three this time)

1. ⚠️ The overnight US session was nearly taken from the wrong trading day. The first search returned "8/18 close: the three major US indices fell broadly, SanDisk +8.88%, Micron +4.13%," which was in fact the 8/17 US session (Chinese media headline by Beijing time). Adopting it would have produced the exact opposite conclusion, "memory soared overnight." Switched to pulling the CNBC quote endpoint directly, with last_time = 2026-08-18T16:00:00-0400, so freshness was verified down to symbol granularity, yielding SNDK −9.01% / STX −9.16% / WDC −7.43% / MU −7.02%.
2. ⚠️ "Robot concept +0.17%, main force net outflow RMB 10.283 billion" cannot be used as is. BK1090 "robot concept" is a broad basket containing many heavyweight electronics names, and the outflow is dominated by unrelated large caps; the genuinely core robot names all saw main-force net inflows yesterday (绿的谐波 +RMB 221 million, 卧龙 +RMB 202 million, 机器人 +RMB 186 million, 秦川 +RMB 177 million, 金发 +RMB 177 million, 埃斯顿 +RMB 166 million, 南方精工 +RMB 133 million), and the narrow sub-concepts were also net inflows (reducers +RMB 1.373 billion, robot actuators +RMB 1.193 billion). The main text uses the narrow basket and never cites that −RMB 10.283 billion.
3. ⚠️ 襄阳轴承 (000678) "joint development of the Unitree H1 with 宇树" — the most dangerous item this time. It circulates through multiple channels in the form of "the company has confirmed it," when the original text is an investor's question on the e-interaction platform ending with "is the above true?", which the company did not answer, and a repeat question on 8/13 that again went unanswered. Without tracing back to the original Q&A on the platform, this would have been written up as a concrete supply fact. This is the mirror image of "'no news found' ≠ 'no news': you can find the content, but its author is not the company."

4. Items in this edition's first draft corrected by sub-agents (must be recorded)

6 of these 10 are "the number itself looks perfectly normal in the quote software, but the basis is wrong." By this repo's records, the worst errors in the pre-market slot have all taken this shape — the main slot cannot find them on its own; it takes a sub-agent's back-calculation. The two fundamentals-analyst runs took about 17–20 minutes each this time. Worth it.

5. Not done / not obtained in this edition

• Not obtained: 卧龙电驱's exact subscribed share in the 金石成长 fund, and any first-hand company confirmation of its "supply to 宇树"; the actual amount and customer list behind 绿的谐波's "humanoid robot revenue"; 长盛轴承's robot business share; 中曼石油's unit costs and output for its oil and gas segment; the size of "other equity instruments (perpetual bonds)" inside 首开股份's attributable equity (reported attributable equity of RMB 7.374 billion implies PB≈1.51 while the vendor shows PB=4.74, and the gap of about RMB 5 billion is suspected to be perpetual bonds, unverified).
• Not re-verified: the 8/19 scheduled interim-report disclosure list (see the §2 failures); yesterday's result was carried over.
• Not done: no tracing of the "first disclosure date" for the strong-interim names 多氟多 (002407), 天华新能 (300390), 株冶集团 (600961), 图南股份 (300855) and others. This is exactly the action required by this repo's "trace to the first disclosure date before discussing an expectation gap" record; it was not executed this morning for lack of time, so all of these names go only onto the Pass list with no recommendation, and §6 states "needs the pre-announcement check first." This is an honest downgrade, not an omission.
• bull-bear-debater was not used. This run called 2 fundamentals-analysts + 1 risk-auditor.

6. Methodology notes

1. The "pricing progress" dimension required by yesterday's recap was implemented for the first time today (§1.1), and it paid off immediately: it corrected robotics from "a brand-new in-window catalyst" to "pricing day 1 + event-landing day," and corrected agriculture from "8/16, outside the window, should be downgraded" to "a divergence point at pricing day 3." Scoring on the release window alone would have rated robotics higher today and thrown agriculture out entirely.
2. A new failure mode appeared today: "an investor's question treated as company disclosure." It belongs to the same family as this repo's already-recorded "a search summary conjuring an earnings report out of nothing" and "a recurring column's headline hiding the publication date," but it is more insidious — because the text really is on the company's official platform (the e-interaction platform); it is just that the author is the questioner, not the company. The test: check whether the entry has an "answer" field, not whether it sits on the official site.
3. The "low PE trap" appeared twice today in opposite directions: 兆易创新 is the cycle-top shape of "PE percentile 0% + PB percentile 82%," while 南方精工 is displayed by quote software as 5,981x when it is actually 24.3x. Both say the same thing: before any PE is used for ranking, you must first ask "which four quarters is this E summed from?"