A-Share Pre-Market Brief | 2026-09-02, Wednesday

Scope and caliber boundaries of this issue (please judge the strength of the conclusions accordingly) ① Scan window: 2026-09-01 (Tuesday) 15:00 close → 2026-09-02 07:00, 16 hours in total. Includes A-share post-close filings and the full U.S. trading day of Tuesday (9/1) (closing at 04:00 Beijing time on 9/2). ② Execution-price baseline: this issue is written at 07:00, so the earliest price a reader can actually execute at today is the opening price produced by the 09:25 call auction, not the 9/1 close. Unless otherwise noted, all price changes in this issue are on a 9/1 closing basis; please convert them yourself when placing orders. ③ A-share quotes are on the Tencent qt.gtimg.cn post-settlement closing caliber, timestamps 20260901161403–20260901161500, later than the 15:00 close, i.e. post-settlement data. The price triple has been reverse-computed and verified name by name (e.g. 隆平高科 (Yuan Longping High-Tech, 000998) 9.06×1.1004=9.97 ✓, 芒果超媒 (Mango Excellent Media, 300413) 16.98×1.2002=20.38 ✓). ④ Limit-ups / consecutive limit-ups / seal orders are on the East Money official limit-up pool 9/1 caliber, verified as consistent with tc=83 / len(pool)=83. ⑤ Sector price changes and sector net inflows are on the THS (Tonghuashun) 90-industry-sector closing snapshot caliber; per-stock main-force net amounts are on the THS "real-time" caliber. The two differ from East Money's "main-force net inflow" caliber and must not be subtracted across sources. ⑥ The dragon-tiger list (LHB) is the officially disclosed 9/1 data (published by the exchange after 18:00; this issue has obtained it, filling the gap left by yesterday's post-close issue). Institutional-seat data has been broken out separately and is not mixed with "LHB net buy amount" — the two calibers differ; see §2.5. ⑦ No northbound flow data: since 2024-08-19 the daily net buy amount of Stock Connect is no longer disclosed. This issue gives no northbound figure and does not cite any API value returning 0.0. ⑧ No CBOT grain futures levels are given in this issue: the corn/wheat/soybean contracts returned by the data source have last exactly equal to prev and are flagged UNCH — this is a stale snapshot, not a genuine flat print, and has been discarded. The overnight direction of overseas agricultural products is instead taken from ETF proxies + an independent second source cross-check; see §1 Note ②. ⑨ Domestic commodity futures were obtained, and they are a key link in this issue's conclusion: the DCE corn continuous main contract (Sina futures_zh_daily_sina C0) is complete and usable, closing 9/1 at RMB 2,293/tonne (−0.26%). Note the distinction: what failed is the offshore CBOT channel; the domestic DCE channel is fine — do not abandon the latter because the former failed. This data is the core evidence in §2.2.1 that overturns "grain prices up → seed stocks up".

⚠️ Data-pull and search log (this section is not sent to clients):

I. API failures that actually occurred this time

• East Money per-stock fund flow stock_individual_fund_flow_rank returned JSONDecodeError (non-JSON), same root cause as yesterday's post-close issue's clist failure, and this has now run for several consecutive trading days. Per the established fallback channel in memory, switched to THS stock_fund_flow_individual(symbol='即时'), which returned 5,207 rows and worked. All per-stock main-force net amounts in this issue are on the THS caliber.
• East Money industry-sector API failed from the same root cause; sector data continues to go through stock_board_industry_summary_ths() (90 rows, normal).
• CNBC grain futures fields are broken: for the three contracts @C.1/@W.1/@S.1, last and previous_day_closing are exactly equal and change_pct is flagged UNCH. In the same request @CL.1 crude returned +0.67% and @NG.1 natural gas +1.83%, which are normal, showing this is not a whole-request failure but individual staleness of these grain symbols — exactly the class of "CBOE's _DJI/_COMP is dead data" / "freshness must be checked down to symbol granularity" in memory. Discarded; switched to ETF proxies.
• .SSEC / .HSI / @LCO.1 likewise returned UNCH with last==prev, and were likewise discarded.

II. Three search contaminations intercepted this time (all of which would have changed conclusions)

1. The search result for "U.S. stocks September 1 close" returned Monday's (8/31) market action (East Money 202609013860550173, headline "Semiconductor and memory sectors mostly up, SanDisk +5%, Micron +2%"). That article was published on the morning of 9/1 China time, describes U.S. Monday, lies outside this issue's window, and points in exactly the opposite direction from the real Tuesday session. Copying it would have written "semis surged overnight" into the report, whereas Tuesday's SOX was in fact −2.14% and Micron −2.64%. How it was caught: colliding CNBC's previous_day_closing with the article's closing numbers — SPX 7,686.14 / IXIC 26,370.889 / DJI 53,185.90 are all three exactly equal, proving the article describes precisely the "previous day" in CNBC's caliber, while CNBC's last is the in-window Tuesday close. This is the single most critical cross-check in this issue.
2. "A-share September 2 overnight news roundup" returned an 8/30 weekend piece (mortgage tenor extended to 40 years, big-six banks' interim reports, YMTC STAR Market IPO accepted with a planned RMB 33 billion raise). All three fall outside the window; of these, the big-six banks' interim dividend was verified as disclosed on 8/28. Treating them as today's catalysts would have conjured a "major policy positive" out of thin air. All have been handled as carry-over logic with their original publication dates labeled.
3. "Russia cuts 2026 oil output to 494.2 million tonnes, a 17-year low" was listed by one 9/1 search result as a same-day headline; verification shows it is a Russian Ministry of Economy forecast dated 2026-05-12, unrelated to this window. Removed.

III. Channels that worked

• Limit-up pool push2ex getTopicZTPool (dpt=wz.ztzt, sort=fbt:asc, date=20260901) returned tc=83 / len=83, verification passed. Continue using sort=fbt:asc.
• Source for broken-board data: ak.stock_zt_pool_zbgc_em(date=20260901) returned 6 rows with a complete column structure (not an empty shell), so "6 broken boards" is usable. ⚠️ By contrast, the limit-down pool stock_zt_pool_dtgc_em returned 0 rows, and 0 rows cannot be distinguished from "genuinely no limit-downs", so this issue gives a broken-board number but no limit-down number — the same criterion is applied to both; the conclusions differ only because the returned structures differ.
• LHB ak.stock_lhb_detail_em('20260901','20260901') returned 79 rows and the pull succeeded at 07:00 today — yesterday's post-close issue failing at 15:30 was a scheduling problem (disclosure only after 18:00), not an API failure; now confirmed.
• Institutional seats ak.stock_lhb_jgmmtj_em returned 44 rows; this is the single most valuable source in this issue (see below).
• Quotes went through Tencent qt.gtimg.cn, timestamp 16:14, after the closing settlement.

IV. Two cross-checks actively performed on method, both of which changed conclusions

1. The draft was about to write agriculture as the No.1 recommended branch (strongest price +7.51%, overnight overseas ADM +4.01% confirming). It was self-rejected after checking flows: the 14 agricultural limit-up stocks had a combined main-force net amount of RMB −1.131 billion, and LHB institutional seats on 敦煌种业 (Dunhuang Seed Industry, 600354) were 0 buy / 1 sell, net sell RMB 126.57 million — on a stock that limit-upped +10.01% that day. "Price rising while institutions sell" cannot be written as "funds chasing". Agriculture was therefore demoted to the second branch with position risk explicitly flagged.
2. The draft was about to rank stocks by "main-force net amount". This was withdrawn per the lesson in memory that "single-day main-force fund-flow extremes are a contrarian indicator" — this issue uses main-force net amount only to characterize the structure of "who is buying" (the aggregate fact that 59 of 80 limit-up stocks saw net outflows), not as a per-stock ranker; the ranking in §3 still runs on evidence hardness + fundamentals as its main axis.
3. ⚠️ The most important one: 阳光股份 (Yangguang Co., 000608) went from "recommendation slot No.5, 62 pts" → "watch only" → finally Pass, 44 pts. The draft's reasoning was "the only brand-new company-level catalyst in the window + the share price did not move at all that day + RMB 633 million is meaningful against a RMB 5.4 billion market cap". All three still hold, but none of them is decisive. After a sub-agent pulled the original filing, four pieces of hard evidence overturned the narrative:
   • Pricing is "monthly comprehensive service fee × number of units", paid in arrears by calendar month, with no minimum serviced unit count and no floor service fee anywhere in the text. Therefore RMB 633 million is the nominal ceiling of five years at full load, not a committed floor. The draft treated it as "backlog", which exceeded what the filing supports. This is an upgraded form of the memory "cumulative-period order figures are old news" — this is not even a cumulative commitment.
   • The contracting entity, Shenzhen Yangguang Jinhui, was acquired on 2026-05-25 via a 60% stake purchase for RMB 0.1 million; net assets on the acquisition date were RMB 9,288.10, and net assets at end-H1 were RMB −0.7003 million.
   • Using the company's self-reported 10%–21% gross margin, the annualized gross profit attributable to parent tops out at RMB 15.05 million, while annual interest on the existing RMB 660 million shareholder loans is about RMB 19.14 million — the net contribution could be negative. The draft only computed the revenue share (36.53%) and did not carry it one layer down to profit, nearly mistaking revenue elasticity for earnings elasticity.
   • In the same filing the company itself states it "does not constitute a material impact", and in the interim report it concedes "P/B 2.65 vs the industry's 0.83; the valuation premium lacks adequate fundamental support". The issuer itself denied materiality — this is the hardest point of all. One stale figure already written into this issue's draft was also corrected: the widely circulated "only RMB 29 million on the books" caliber is as of 2026-03-31 and has been superseded by the interim report (6/30 cash and equivalents RMB 232.82 million). Although the draft had labeled it "no primary verification obtained", it still cited a stale figure as fact — this is precisely the same class of error as "silently lagging data source", except the lag here is one quarter. The corrected wording is "cash is 100% sourced from controlling-shareholder loans, with operating cash flow of RMB −82.50 million", the direction of the conclusion is unchanged, but the basis has moved from hearsay to the financial statements.
4. A risk not discussed by the market, dug out proactively by a sub-agent: in 2025 the company was placed under delisting-risk warning for "negative net profit and ex-item revenue < RMB 300 million", which was only lifted on 2026-05-21; its 2025 post-deduction revenue was RMB 332.74 million, only RMB 32.74 million above the line; and 2026H1 revenue was −10.83% year over year. If H2 continues that trend, full-year post-deduction revenue will land near the RMB 300 million line. This is a variable that decides whether the cap is re-imposed after the 2026 annual report, and it bears directly on "whether this contract can start being billed within the year". (This projection is the sub-agent's estimate, not company disclosure, and is labeled as such in the main text.)

V. Two other places (besides 阳光股份) where sub-agent verification changed the conclusion

1. ⚠️ The agriculture branch was rewritten wholesale; this is the largest change in this issue. The draft gave the reason for passing on seed stocks as "indirect, spans planting seasons, expensive valuation" — that reasoning was soft and missed the point. The sub-agent pulled two pieces of hard evidence:
   • The DCE corn continuous main contract closed 9/1 at 2,293 (−0.26%), below the 2,328 of 6/30; since 8/14 it is only +3.01%, while 万向德农 (Wanxiang Denong, 600371) is +96.3% over the same span. This issue independently re-verified with futures_zh_daily_sina C0; the figures match.
   • 万向德农's 9/1 abnormal-movement filing and 登海种业 (Denghai Seeds, 002041)'s 8/26 filing independently state "a different link from the commodity corn market", "oversupply, selling prices under pressure, gross margin declining". In other words: what negates this chain is not the analyst, it is the issuers themselves. The draft's soft reasoning was therefore upgraded to "the logic is denied by primary filings"; seed stocks were uniformly re-rated from "watch only / Pass (valuation)" to "Pass (logic denied)", and §6.1c was added to list this class separately. A valuation misreading was also corrected: 隆平高科's headline P/E is 257 but TTM ex-non-recurring is a loss, and 神农种业 (Shennong Seed, 300189)'s headline 71.6x corresponds to an ex-non-recurring P/E of about 6,722x — headline P/E badly understates how expensive this group truly is.
2. ⚠️ The evidence on 中文在线 (Chinese Online, 300364) was completed, and half of the conclusion changed as a result. The draft wrote in §2.5 that "institutions bought 中文在线 rather than 芒果超媒" — that was selective citation: institutional net buying of RMB 174.73 million is real, but the LHB as a whole was a net sell of RMB 162.26 million that day (implying roughly RMB 337 million of net selling by hot money), and the tape showed a gap-up fade of −1.84% on 33.26% turnover. With that added, the true contrast is "institutions buy, hot money dumps" versus "institutions sell, hot money lifts" — neither is a followable pattern. Also found: 2026Q2 single-quarter revenue −17.03% YoY and −13.6% QoQ, the lowest in five quarters; that +3.85% in H1 came entirely from Q1, and "short drama +108.72%" is an H1-vs-H1 caliber that masks a double decline in the most recent quarter; plus the actual controller disclosed new pledges on 9/1, cumulative pledge ratio 45.18%, P/B 78.3, current ratio 0.62. The rating stays "watch only" but the basis has been completely rewritten.
3. A correction to an error in the prompt (recorded to avoid reuse): the prompt given to the sub-agent asserted that "苏垦农发 saw large institutional net buying on the 9/1 LHB". Verification shows the stock did not appear on the LHB at all that day (the Shanghai "price deviation reaching 7%" list is a top-five quota, and that day it was filled by five +10% limit-up stocks; 苏垦 at +7.39% ranked sixth or lower). This is "not found", not "no institutions", the main text has been corrected accordingly, and no inference is drawn from seat data. Lesson: the comparison framework handed to a sub-agent must itself be verified first, otherwise a non-existent fact gets propagated as a premise.

VI. What this issue did not do (listed honestly)

• No reliable read on the number of 9/1 limit-downs was obtained (stock_zt_pool_dtgc_em returned 0 rows). 0 rows cannot be distinguished from "genuinely no limit-downs", so this issue gives no limit-down figure.
• Sub-agent-level fundamental verification was not done for all 22 listed stocks; only recommendation slots and key Pass names were verified.

---

0. Today in One Sentence

Strongest catalyst: there is not a single heavyweight brand-new policy. In this 16-hour window, the only brand-new, company-level, not-yet-priced hard catalyst is the RMB 633 million server leasing contract disclosed by 阳光股份 (Yangguang Co., 000608) after the close on 9/1; the only market event occurring today is the official launch today (9/2) of the SSE STAR Market Artificial Intelligence Application Select Index. All other "big news" was verified to fall outside the window (see §1 Note ①).

Strongest branch: the flow caliber and the price caliber must be read separately today — this is the core judgment of this issue. Yesterday's strongest price action was crop farming and forestry +7.51% (30 up, 0 down), but its sector net inflow was only RMB +146 million; whereas banks +2.15% took the largest net inflow of the entire market at RMB +8.381 billion, and insurance +3.16% / RMB +2.716 billion. More importantly: of the 83 limit-up stocks on 9/1, 80 had main-force net amounts available (3 missing due to code-matching gaps, same below); those 80 totaled RMB −6.951 billion, with 59 in net outflow and 21 in net inflow, median RMB −38.66 million. That broad rally of "83 limit-ups, 6 broken boards" was structurally main-force distribution, absorbed by hot money and retail.(⚠️ All 3 stocks with missing data had turnover below RMB 100 million, not enough to change the direction of a RMB −6.951 billion magnitude.)

Likely direction of funds: low-valuation, high-dividend large financials are where the money is actually going, but its catalyst is carry-over (the RMB 220.989 billion big-six interim dividend on 8/28), not new news; agriculture is where price and popularity are, but the main force is distributing and it is already in week three of being priced. The two will most likely keep diverging today.

⚠️ The most important verification in this issue overturns the market's mainstream explanation of yesterday's agriculture rally: all three corn seed stocks limit-upped yesterday (corn seed is 95.51% of 登海种业's revenue and 93.62% of 万向德农's), while that same day DCE corn futures closed at RMB 2,293/tonne, down 0.26%, below the 2,328 of 6/30; since 8/14 corn is +3.01% while 万向德农 is +96.3%. Harder still — 万向德农's 9/1 abnormal-movement filing dismantles this chain with its own hands: "the company's main business is corn seed... a different link from the commodity corn market", "the industry as a whole is in a destocking cycle, the market shows oversupply... selling prices are under pressure and gross margin has also declined", and it states the share price has "seriously deviated from fundamentals and a reasonable valuation range"; 登海种业's 8/26 filing independently restates the same judgment. The El Niño narrative is about global wheat and rice, while these companies' wheat seed share is 0%–0.72% — the crop in the narrative and the crop of the targets are simply not the same crop. This issue therefore passes on the entire "selling seed" sub-branch, and gives agriculture recommendation slots only to companies that "sell grain" (see §2.2.1).

Driver types: overseas industry (agricultural products and semiconductors, opposite directions) + a single-company order + an index event. U.S. stocks fell on Tuesday overnight, the Philadelphia Semiconductor Index −2.14%, Micron −2.64%, AMD −2.36%, which is the second blow to A-share semis after their RMB 17.448 billion net outflow on 9/1; while agricultural stocks rose in the U.S. (ADM +4.01%, Nutrien +2.86%, Mosaic +2.74%). What the overseas session hands A-shares tonight is a directional vote of "sell tech, buy agriculture".

Pre-market stance: cautious, defensively tilted. Of yesterday's 83 limit-ups, 65 were first boards (78%), the consecutive-board tier had only 18 names, and the highest consecutive-board count in the entire market is the 7 boards of 海鸥住工 (Seagull Sanitary Ware, 002084), which is a change-of-control theme and belongs to no main line. A structure of "80% first boards + the highest consecutive board unrelated to any main line + limit-up stocks in aggregate main-force net outflow is long on sentiment and short on core, not a position for scaling up and chasing highs.

---

1. News Overview

Window: 2026-09-01 15:00 → 2026-09-02 07:00 (Beijing time)

Note ① ⚠️ Three items widely treated by the market as "today's positives" but verified to fall outside the window; this issue does not use them:

• "Personal mortgage tenor extended from 30 years to 40 years", "improvement of the commodity housing sales system" — from the 8/30 weekend roundup, not this window.
• "YMTC's STAR Market IPO accepted, planned raise RMB 33 billion" — in fact accepted by the SSE on 2026-08-21 (the entity being YMTC Holdings), outside the window. It does matter for the memory chain, but it is not an overnight addition, and directionally it is far-end supply rather than near-end demand (cf. memory: equipment/capacity news speaks to future supply, not current-period prosperity).
• "Big-six banks' interim dividend of RMB 220.989 billion, +7.98% YoY" — verified as disclosed on 8/28 together with the interim reports. This is the fundamental fulcrum of the bank branch in §2.1, but it must be labeled carry-over information and cannot be treated as a fresh overnight catalyst.
• "Russia cuts 2026 oil output to a 17-year low" — verified as a 2026-05-12 forecast by the Russian Ministry of Economy, wrongly mixed into 9/1 headlines by search results. Removed.

Note ② Data-sourcing note on the overnight direction of agricultural products (must be disclosed): this issue gives no CBOT grain futures levels — the corn/wheat/soybean contract quotes returned by the data source are stale snapshots (see caliber boundary ⑧). Two mutually independent channels are used to cross-verify instead: ① agricultural ETFs: wheat WEAT +0.68%, corn CORN +1.15%, broad agriculture DBA +0.58%; ② independent second source tradingeconomics shows wheat +0.73% on 9/1. The two channels' readings for wheat differ by 0.05 percentage points, matching in both direction and magnitude, so they can be relied on. ⚠️ But note a divergence in magnitude here, and it determines how the agriculture branch is characterized: the commodities themselves rose only 0.6%–1.2%, while agricultural stocks rose 2.7%–4.0%. The equities' gains are 3–4x the commodities', showing the overnight overseas session priced "expectations", not "spot". This does not constitute confirmation of A-share agricultural stocks' fundamentals, only confirmation of sentiment.

---

2. Strongest Positive Branches, Descending

2.1 Branch One: Large Financials — the Only Direction Where Money Is Genuinely Going In

Why it is positive: among yesterday's 90 industry sectors, banks ranked first in the entire market with RMB +8.381 billion of net inflow, 3x the second-placed insurance (RMB +2.716 billion). On price, banks rose only 2.15% (14th), a severe mismatch between price rank and flow rank — which is exactly why it deserves the first slot. 工商银行 (ICBC, 601398), 建设银行 (China Construction Bank, 601939), 中国银行 (Bank of China, 601988) and others hit all-time highs intraday; 中国银行's combined A+H market cap topped RMB 2 trillion; 邮储银行 led with +4.18%.

Logic hardness (high, but the catalyst is carry-over): three facts backed by financial statements support it — Q2 commercial-bank NIM recovered to 1.41%, up 1 basis point from Q1, the first quarter-on-quarter increase since Q1 2022; the 42 A-share listed banks posted H1 revenue of about RMB 3.14 trillion (+7.4%) and net profit attributable to parent of about RMB 1.13 trillion (+2.96%); the big-six banks' interim dividends totaled about RMB 220.989 billion (+7.98%), the largest on record. ⚠️ But per the iron rules this must be labeled: all three were disclosed on 8/28, are outside this issue's window, and represent continued fermentation of carry-over logic, not a fresh overnight catalyst. This issue ranks it first on the basis of yesterday's vote by the money, not because "new news came out".

Valuation contrast (the most striking set of numbers in this issue):

Of yesterday's 14 limit-up agricultural stocks, 6 are loss-making companies. The money went to the P/E 5.6–8.6 side; the price gains went to the P/E 70–257 and loss-making side. This is not a value judgment, it is a factual description of "who is buying".

Persistence: medium term. High dividends and insurance-capital allocation are slow variables that will not play out in a day; but it cannot deliver short-term elasticity — these are ±10% main-board names where 2%–4% intraday is the norm.

Fully priced in? Partly. Banks have already risen for a stretch and several are at all-time highs, so the position is not low; but the NIM inflection is the first since 2022, and institutional positioning is still being rebuilt.

Board distribution: all Shanghai/Shenzhen main board ±10%, no differences in price limits.

2.2 Branch Two: Agriculture / Grain — First in Price, Confirmed by Overseas Markets, but the Main Force Is Distributing

Why it is positive: on 9/1 crop farming and forestry was +7.51% with 30 up and 0 down, first in the market with not a single decliner; agricultural product processing +2.90%, agrochemicals +2.39%, animal husbandry +2.23%. Agriculture accounted for 14 of the 83 limit-ups. Overnight U.S. agricultural stocks confirmed the direction again (ADM +4.01%, Nutrien +2.86%, Mosaic +2.74%).

Logic hardness (medium): the driver is expectations of rising grain prices — J.P. Morgan warned that a global food crisis could erupt next year (five Ws: War/Weather/Warehousing/Water/Waste), NOAA raised the probability of a historic super El Niño in October–December 2026 to 69% and "very strong" to 81%, and HSBC warned that global grain output in 2026/27 would see the first supply-demand inversion since 2020/21. ⚠️ But these warnings were published 8/16–8/21; they are carry-over information with nothing new inside this window.

2.2.1 ⚠️ The Most Important Verification in This Issue: Seed Stocks Have Already Decoupled From Their Own Commodity Price

This section overturns the market's mainstream explanation of this agriculture rally. There are two pieces of evidence, mutually independent, and both from primary sources.

Evidence one: all three corn seed stocks limit-upped yesterday, while Chinese corn futures fell that same day.

DCE corn continuous main contract (futures_zh_daily_sina C0, independently re-verified in this issue):

⚠️ Caliber reminder: the continuous main contract rolled over between end-June and early September, so the price series contains roll gaps; the cross-contract comparison "1.50% below 6/30" is therefore only indicative and this issue draws no conclusion from it. What actually supports the conclusion is the 8/14 → 9/1 stretch within the same main contract: corn +3.01%, and the −0.26% on 9/1 itself.

Meanwhile 万向德农 (Wanxiang Denong, 600371) rose +96.3% from 8/14 to 9/1. Over the same period the underlying commodity rose 3.01% and the stock rose 96.3%, a gap of about 32x. Yesterday's limit-ups 登海种业 (corn seed 95.51% of revenue), 万向德农 (93.62%), 神农种业 (15.84%) all saw the price of their main crop fall on the very day they limit-upped. What the El Niño narrative actually corresponds to is global wheat and rice (FAO July cereal index +3.4% MoM, wheat futures +8.60%), while these companies' wheat seed shares are: 登海 0.72%, 万向德农 0%, 神农 0%. The crop in the narrative and the crop the targets grow are not the same crop.

Evidence two: the two companies concerned dismantled the "grain prices up → seed stocks up" chain with their own hands, in their own regulatory filings.

万向德农 (Wanxiang Denong, 600371) 2026-09-01 "Announcement on Abnormal Share Price Movement" (No. 2026-033), original text: "Recently, the market has discussed climate topics such as 'super El Niño' extensively; the related topics mainly concern price fluctuations of agricultural products such as commodity corn, whereas the company's main business is the R&D, production and sale of corn seed, which belongs to the upstream of the agricultural industry chain and is a different link from the commodity corn market. At present, the domestic corn seed industry as a whole is in a destocking cycle, the market shows oversupply, intensified competition has put corn seed selling prices under pressure, and the company's corn seed gross margin has also declined." The same filing also states: cumulative gain of 78.48% from 8/17 to 8/31, which has "seriously deviated from fundamentals and a reasonable valuation range, with risks of overheated market sentiment, irrational speculation and a rapid pullback in the share price"; the company "is not involved in any matters such as material asset restructuring, share buybacks, equity incentives, major business cooperation, or introduction of strategic investors".

登海种业 (Denghai Seeds, 002041) 2026-08-26 abnormal-movement filing independently restates the same judgment: "At present the corn seed market as a whole shows oversupply, industry supply-demand contradictions remain prominent, and market competition is severe."

Two listed companies, two independent filings, telling the market in the same direction: this transmission chain does not hold at the current point in time, and the industry is in a destocking cycle with falling prices. 隆平高科's statements corroborate this too — RMB 4.661 billion of inventory sits on the books (of which corn seed RMB 2.964 billion), with inventory turnover of 772.86 days.

⚠️ The clock does not line up either. The commercial rhythm of seed is fixed: summer seed production (seed for the 2027 spring planting is already in the ground, output and cost locked) → autumn October–December ordering conference sets prices → shipments and revenue recognition in Q1 of the following year. Therefore the El Niño narrative can at the earliest affect the autumn 2026 ordering conference, showing up in the 2026 annual report and 2027 Q1 report; it is zero in the 2026 interim report and close to zero in the Q3 report as well.

⚠️ Transmission-chain grading (iron rule 2), rewritten after this round of verification:

• Grain prices ↑ → own-farmland cultivation / grain reserve and trading: direct, current period. Selling prices float with the market straight into the income statement, and land contracting costs are locked by contract. Representative: 苏垦农发 (Suken Agriculture Development, 601952). ⚠️ Note two easily misclassified exceptions: 深粮控股 (Shenzhen Cereals Holdings, 000019) also runs food processing, and the processing segment is on the cost side of grain prices, not the beneficiary side; 北大荒 (Beidahuang, 600598) derives about 72% of revenue from land contracting fees, which is "collecting rent" rather than "selling grain" — the contracting fee is a fixed price and grain prices do not enter its statements in the current period. Both have been moved out of this tier; see §5.3 and §5.3b.
• Grain prices ↑ → seed companies: ⚠️ downgraded from "indirect and lagging" to "denied by company filings at the current point in time". Beyond the lag across planting seasons, the more critical point is: the price of the underlying crop (Chinese corn) is not rising, and the industry is in an oversupplied destocking cycle with seed selling prices under pressure. Yesterday's limit-ups 隆平高科, 登海种业, 神农种业, 农发种业, 万向德农 all fall into this tier.
• Grain prices ↑ → forestry: far-end mapping, essentially does not hold. 福建金森 (Fujian Jinsen Forestry, 002679) is mainly in forest cultivation and timber production and sales; there is no direct causality between grain prices and timber prices. It has the most remote logic of this agriculture rally, yet the best seal quality (see §2.3).

⚠️ Persistence and position (the most important passage in this branch):

• Already in week 3 of pricing. Among the 88 limit-ups on 8/31 there were already 3 agricultural names (万向德农 (600371) 5 consecutive boards, 新赛股份 (Xinsai, 600540) 3 consecutive boards, 福建金森 (002679) 2 consecutive boards), spreading to 14 on 9/1. This is not the launch point, it is the diffusion point.
• The main force is distributing: the 14 agricultural limit-ups had a combined main-force net amount of RMB −1.131 billion. Of these, 新赛股份 (600540, 4 boards) was RMB −229 million.
• Institutions are selling: LHB institutional seats on 敦煌种业 (Dunhuang Seed Industry, 600354) were 0 buy / 1 sell, net sell RMB 126.57 million, on a stock that limit-upped +10.01% that day with 24.22% turnover. RMB 127 million of one-sided institutional selling on a limit-up name is the hardest de-risking signal in this branch.
• Poor quality of the vehicles: 6 of the 14 are loss-making, with the highest P/E at 257x.
• ⚠️ This line already had one collective limit-down on 8/21: that day 万向德农 −10.00%, 神农种业 −11.83%, 登海种业 −9.04%. A collective limit-down washout within 12 trading days shows the chip structure itself is fragile.
• ⚠️ The truth about the four seed companies' ex-non-recurring earnings (verified against interim reports; very different from what P/E appearances suggest):

Headline P/E already badly understates this group's true valuation level. 隆平 and 万向德农 have negative TTM ex-non-recurring earnings, and 神农种业's ex-non-recurring P/E is about 6,722x.

Conclusion: the branch direction is questionable, almost all the vehicles fail to qualify, and the position is elevated. This issue gives agriculture three recommendation slots, all to names that "did not limit-up, grow the crop the narrative points to, and make money by selling grain rather than selling seed", and not a single seed stock.

2.3 Inside Agriculture: Seal Quality Separates the Hottest From the Most Solid

This issue quantifies seal quality with "seal order amount ÷ turnover" (seal ratio) — this ratio answers "how much it cost to seal it, relative to how many are running", which is more reliable than looking at seal order size alone. The market-wide median is 0.183.

This table yields one directly actionable judgment in this issue: the two biggest gainers in agriculture yesterday (神农种业 +20.03%, 敦煌种业 +10.01%) have the two worst seal qualities in the branch, and turnover rates of 37.26% and 24.22% show chips are churning violently rather than being locked up. Meanwhile the most solidly sealed name, 福建金森, has the most remote industrial logic (forestry). "Highest trading recognizability" and "hardest logic" are separated in this branch, and neither equals "worth buying" — these are three orthogonal conclusions.

2.4 Branch Three: AI Compute / STAR AI — the Only Brand-New Company-Level Catalyst in the Window, but Facing an Overseas Headwind

Two catalysts:

1. 阳光股份 (Yangguang Co., 000608)'s RMB 633 million server leasing contract — signed by controlled subsidiary Shenzhen Yangguang Jinhui with "Company A", service period 5 years, total price including tax RMB 633 million. The company closed 9/1 at RMB 7.25 (−0.41%, it did not move that day), with a total market cap of only RMB 5.437 billion. This news only came out after the close on 9/1 and has not been priced. ⚠️ But after checking the original filing line by line, this issue has demoted it from a recommendation-slot candidate to Pass. The three most critical points are: pricing is "monthly comprehensive service fee × number of units", paid monthly in arrears, with no minimum serviced unit count or floor clause anywhere in the text, so RMB 633 million is a full-load nominal ceiling rather than locked-in revenue; the contracting entity Shenzhen Yangguang Jinhui is a company in which 60% was acquired on 2026-05-25 for RMB 0.1 million, with net assets of RMB 9,288 on the acquisition date and negative net assets now; in the same filing the company states that "the new business does not constitute a material impact on the company's overall production and operations". The full nine pieces of evidence are in §5.10. This catalyst is still the newest in the window, but it cannot support a recommendation slot.
2. The SSE STAR Market Artificial Intelligence Application Select Index officially launches today (9/2) — selecting AI application software and hardware companies from the STAR Market as constituents. An index launch itself brings no incremental money; what it brings is the expectation of subsequent ETF product allocation, so it is event-driven with limited strength.

⚠️ But overnight is a headwind: Philadelphia Semiconductor −2.14%, Micron −2.64%, AMD −2.36%. The A-share compute chain is highly correlated with U.S. semis, which suppresses the probability of a gap-up today.

There is one positive piece of flow evidence: among yesterday's 83 limit-up stocks, the largest main-force net inflow was 常山北明 (Changshan Beiming, 000158) at RMB +167 million — the only limit-up stock in the entire market with main-force net inflow above RMB 150 million, with a volume ratio as high as 5.24. Its subsidiary Beiming Software is a core strategic partner of Huawei. ⚠️ But its P/E is as high as 1,701x, so valuation provides no support.

2.5 ⚠️ Media: the Best Relay Quality in the Market, but Institutions Did Not Buy the One That Limit-Upped

Relay data (computed exclusively in this issue): grouping the 88 limit-up stocks of 8/31 by industry and averaging their 9/1 performance, media is the strongest tier in the market — digital media +11.27% (n=3), film and cinema +8.27% (n=4), publishing +5.82% (n=3), TV broadcasting +3.99% (n=2), while real estate development −5.71% (n=4) and decoration −4.98% (n=4) are at the bottom. Crop farming +10.04% (n=2) ranks second. ⚠️ The sample sizes must be read alongside: each group has only 2–4 samples and the mean is dominated by individual names (e.g. the digital media group contains 芒果超媒 at +20.02%), so this cannot be taken as a robust industry strength ranking, only as a directional observation. A more robust fact at the full-sample level is: of the 88 names, 43 rose and 45 fell on 9/1, a relay win rate below fifty percent.

⚠️ But institutional seats cast exactly the opposite vote:

⚠️ Caliber note (caliber boundary ⑥ of this issue requires that the two not be mixed, so it must be explained here why subtraction is permissible in this case): institutional-seat statistics are aggregated once per stock, whereas LHB details are compiled separately by listing reason. 中文在线 was verified to have only one listing reason (top five by daily turnover rate reaching 30%), so the back-out "overall RMB −162.26 million − institutional RMB +174.73 million = about RMB −337 million for non-institutional seats" holds; 芒果超媒 has two listing reasons whose figures cannot be added, so this issue gives only direction for it and no implied figure.

⚠️ This table has two more columns than the draft, and those two columns changed half of the conclusion. The draft only wrote "institutions net sold 芒果超媒 and net bought 中文在线", which was selective citation. Adding the LHB overall caliber, the true contrast is: 芒果超媒 = institutions sell, hot money lifts (overall net buy); 中文在线 = institutions buy, hot money dumps (overall net sell of RMB 162.2 million). Neither is an "institutions are buying so you can follow" pattern. The tape confirms it: 中文在线 on 9/1 opened at 28.84 and closed at 28.31, a gap-up fade of −1.84%, with 33.26% turnover, an enormous figure (it had already limit-upped +20.02% on 8/31, so 9/1 was huge churn on the day after a limit-up). "Institutional net buying of RMB 174.7 million" is a fact, but it coexists with a gap-up fade and RMB 337 million of hot-money net selling that day, and cannot be used on its own as a reason to buy.

⚠️ The caliber must be spelled out or it will be read backwards: 芒果超媒's LHB "net buy amount" is positive (RMB +188 million / RMB +158 million, compiled separately by the two listing reasons), but that is an aggregate that includes hot-money seats. Broken out by seat, the institution-dedicated seats were 0 buy 1 sell, net sell RMB 108.34 million. This is exactly in the same direction as what yesterday's post-close issue found — "Shenzhen Connect net sell of RMB 117.7 million, institutions 1 buy 2 sell with a net sell of RMB 16.38 million, absorbed by three hot-money desks" — layered on top of the THS-caliber main-force net amount of RMB −1.542 billion (outflow RMB 2.168 billion vs inflow RMB 626 million, 3.5:1). ⚠️ The wording must be tightened (the draft wrote "three mutually independent channels", which was an overstatement): LHB seats and Shenzhen Connect seats are both subsets of "full-day large orders", not three independent samples, so directional agreement is structurally inevitable; and the three figures are not from the same day — institutions RMB −108.34 million and main force RMB −1.542 billion are 9/1, while Shenzhen Connect RMB −117.7 million is 8/31 data found in yesterday's post-close issue. The accurate statement is: on the same day (9/1) there are two levels of evidence (institutional seats RMB −108.34 million, full-day large orders RMB −1.542 billion), plus one historical corroboration from 8/31 (LHB Shenzhen Connect dedicated seat net sell of RMB 117.7 million — note this is seat-level data, not the daily northbound net buy amount that is no longer disclosed). All three point the same way, supporting the judgment that "institutions and foreign capital are distributing while hot money absorbs", but it is two levels + one historical data point, not three independent confirmations. Yesterday's pre-market issue downgrading 芒果超媒 to "watch only" gains further support today.

The policy must be read correctly: the "Administrative Measures for the Development of Micro Short Dramas" took effect on 9/1; its nature is regulatory tightening, not industrial stimulus — Category I/II micro short dramas are subject to filing publicity + distribution licensing, while Category III are reviewed by broadcasters and labeled with program numbers; broadcasters must establish an editor-in-chief content accountability system, verify licenses, review before broadcast, and build a credit rating system. It raises the compliance threshold, directionally favoring large platforms with licenses and compliance capability while suppressing small producers, but it was already published on 7/31 with extremely full expectations, so it constitutes no expectation gap.

2.6 ⚠️ Explicitly Avoid: Semiconductors / Memory

This is the clearest stance in this issue. On 9/1 A-share semiconductors were −2.64%, with net outflow of RMB 17.448 billion (the largest in the market), 21 up and 165 down; electronic chemicals −3.83%, components −3.19%. Overnight the Philadelphia Semiconductor Index fell another 2.14% and Micron −2.64%. This is the second consecutive external negative.

⚠️ Note one piece of information that is easily read backwards: 兆易创新 (GigaDevice, 603986) announced a RMB 654 million buyback after the close on 9/1, and its share price was −3.20% that day. A large buyback is a statement of management's attitude, but it does not change industry conditions or fund flows and cannot be treated as a bottom-fishing signal. Likewise, 中际旭创 (Innolight, 300308)'s first buyback of RMB 318 million is only 0.03% of its RMB 1,012.178 billion market cap, so the amount carries no signal value.

Conclusion: this issue does not list semiconductors as a direction to watch; let it stop bleeding on its own. Yesterday's post-close issue gave the same judgment, and today's overnight data supports it further.

---

3. Master Ranking of Per-Stock Positive Strength

The main ranking axis is "evidence hardness + fundamentals + position", not yesterday's price gain, and not single-day main-force net amount (single-day flow extremes are a contrarian indicator; this issue uses them only to characterize structure, not as a ranker).

---

4. Per-Stock Scoring Model (Total 100)

Example (苏垦农发 601952, 76 pts): authority 12 (company interim report + investor-relations filing + overseas quotes) + directness 18 (grows grain on its own farmland, raw grain priced "at prevailing market prices" entering the current period) + elasticity 10 + fundamentals 12 (P/B 1.89, clean balance sheet, but H1 ex-non-recurring −33.85% so not full marks) + expectation gap 6 + persistence 8 + trading attributes 10 (no limit-up, turnover 4.51%, allows unhurried accumulation, full marks) − risk 0 = 76. Example (神农种业 300189, 40 pts): authority 10 (company abnormal-movement filing + primary LHB) + directness 6 (seeds, spans planting seasons and denied by a peer's filing) + elasticity 6 + fundamentals 7 (headline P/E 71.57 but ex-non-recurring P/E about 6,722x, P/B 11.28, capped per "P/E>100 caps at ≤7") + expectation gap 3 (already +20%) + persistence 8 + trading attributes 3 (seal ratio 0.077, turnover 37.26%, capped per "seal ratio <0.1 or turnover >30% gets ≤3") − risk 3 (institutional seats net sold consecutively on 8/28 and 8/31; 13.45% restricted-share unlock date to be verified) = 40.

⚠️ Two scoring disciplines: ① no sub-item may exceed the cap set in the table above (the draft once gave 苏垦 "trading attributes 12", corrected to the cap of 10); ② the same risk is not deducted twice — 神农种业's institutional net selling is already counted in the risk deduction, so it is not deducted again under trading attributes.

---

5. Detailed Analysis of the Top 10 Stocks

51 苏垦农发601952SH main ±10% | Priority deep-dive | 76 pts · Suken Agriculture Development

• Related news: no new company filing. The driver is the overnight rise in overseas agricultural stocks (9/2 04:00, ADM +4.01%, Nutrien +2.86%, Mosaic +2.74%) plus the continued fermentation of grain-price expectations.
• Positive logic (the core reason this issue puts it first, verified against the interim report and investor-relations filing as primary sources):
  • What it sells is grain itself, not seed. In its 2026-06-04 "Announcement on Receiving Institutional Research" the company spells out the pricing mechanism: "raw grain sales mainly reference spot market prices in the producing area, following prevailing market prices". A rise in grain prices enters its revenue in the current period, 1:1.
  • The cost side is locked by contract. The "Supplementary Agreement to the Land Contracting Agreement" with Jiangsu Agricultural Reclamation Group sets the contracting fee at RMB 361.64 per mu per year from 2021-11-01. Revenue floats up with grain prices while costs do not move, so incremental revenue drops almost entirely into gross profit.
  • The farmland is real and scarce: about 955,000 mu of self-operated farmland, and about 1.362 million mu including transferred and entrusted land, all located in premium producing areas of Jiangsu (interim report lines 375–376, mutually corroborated with the 6/4 research filing's "own + transferred exceeding 1.3 million mu").
  • ⚠️ And the crop it grows is precisely the crop in the narrative. The El Niño and food-crisis narrative corresponds to wheat and rice; 苏垦 grows Jiangsu rice and wheat. Compare §2.2.1: for yesterday's limit-up corn seed stocks, the underlying commodity (Chinese corn) fell that day. This is the fundamental reason this issue ranks it first in the whole agriculture branch while passing on every seed stock.
• ⚠️ But it must be disclosed honestly: its 2026 interim report shows double-digit YoY declines, not growth.

Item	2026H1	YoY
Total operating revenue	RMB 4.332 billion	−5.59%
Net profit attributable to parent	RMB 144 million	−32.48%
Ex-non-recurring net profit	RMB 116 million	−33.85%
Gross margin	11.31%	−0.72pp
Net operating cash flow	RMB +167.2 million	−76.84%
Gearing ratio	46.48%	−4.2pp

Table scrolls horizontally

• ⚠️ So a P/E of 28.45x is not cheap — it is "28x on earnings that have already been compressed by a third", implying the market is already pricing a recovery. Ranking cyclicals by P/E is a classic misuse (when profits are compressed to a minimum the denominator tends to zero); what should be looked at instead is the P/B of 1.89x and the market cap per mu of farmland.
• ✅ But the transmission is already showing up in the statements, and this is the most critical product-level table in this issue (2026H1 vs 2025H1 gross margins):

Product segment	2026H1 revenue share	Change in gross margin	How to read
Agricultural products (own-grown grain sold externally)	11.2%	+5.99pp (12.29%→18.28%)	The farming end is improving
Raw grain	13.9%	+3.22pp (0.65%→3.87%)	The farming end is improving
Rice	11.0%	−3.85pp	Processing end squeezed by rising input prices
Malt	7.4%	−4.34pp	Processing end squeezed
Edible oil	30.5% (share jumped from 20.4%)	+1.08pp	Low-margin business scaled up, diluting the blended gross margin

Table scrolls horizontally

This table resolves the apparent contradiction between "profit decline" and "grain price transmission": farming-end gross margins have clearly improved (+5.99pp / +3.22pp), but the processing end (rice, malt) deteriorated because grain is its cost, and on top of that low-margin edible oil rose from 20.4% to 30.5% of revenue; the three offset each other and the blended gross margin instead fell 0.72pp. In other words: transmission is happening, it just has not reached the blended statement level yet.

• Why H1 has not delivered yet (judgment): the summer grain harvest only came in June, and "prevailing market prices" means sales are distributed across the second half. The H1 income statement mainly reflects carry-over from the previous sales season (2025's low grain prices). Meanwhile 2026H1 summer grain output was 1.401 billion jin, up 10.5%, an all-time high, and barley yield per mu on its own bases hit a record since collective cultivation began — the "volume" variable has been delivered with certainty, leaving only the "price" variable.
• Inventory is the vehicle for H2 profit: inventory rose from RMB 3.084 billion to RMB 3.469 billion (+RMB 385 million), which is also the main reason operating cash flow fell 76.84% YoY. The grain has been harvested into the warehouse and not yet sold at higher prices.
• Chips and balance sheet (the cleanest of the four compared): no equity pledges, no restricted shares pending unlock (fully circulating), goodwill of only RMB 83.3 million, 1.2% of net assets attributable to parent; current ratio 3.47 / quick ratio 1.82; cash of RMB 1.456 billion vs short-term borrowings of only RMB 30.3 million. ⚠️ Do not be frightened by the 46.48% gearing ratio: RMB 4.098 billion of it is lease liabilities, 57.1% of total liabilities, the result of capitalizing land contracting rights under the new lease standard, not financial borrowings. True interest-bearing debt is extremely low.
• Dividend: 2026 interim dividend of RMB 0.4 per 10 shares, totaling RMB 55.12 million, 38.29% of H1 net profit attributable to parent; the 2025 full-year payout ratio was 52.82%.
• Technical sentiment: SH main board ±10%. On 9/1 it opened at 9.21 / closed at 9.88, closing near the high, open→close +7.27%, turnover only 4.51%. ⚠️ One correction: it did not appear on the 9/1 LHB (the Shanghai "price deviation reaching 7%" list is a top-five quota and was filled that day by five +10% limit-up stocks), so it has no seat data at all — this is "not found", not "no institutions". This issue draws no inference from it.
• Final judgment: priority deep-dive. It is the only name in this issue that simultaneously satisfies "shortest transmission chain and effective in the current period + the crop matches the narrative + costs locked by contract + cleanest balance sheet + lowest position with the cleanest chips". ⚠️ But two things must be stated honestly: ① its current earnings are declining, and the recommendation rests on the forward judgment of "clear mechanism + farming-end gross margin already improving", not on delivered statements; ② a P/E of 28.45x is not cheap. Real verification must wait for the 2026 Q3 report (late October), watching three things — whether agricultural product / raw grain gross margins keep improving, whether inventory starts drawing down from RMB 3.469 billion, and whether single-quarter net profit attributable to parent stops falling. Only if all three hold is this logic confirmed.

52 邮储银行601658SH main ±10% | Watch closely | 74 pts · Postal Savings Bank

• Related news: the bank sector saw net inflow of RMB 8.381 billion on 9/1, first among all 90 sectors; 邮储 was +4.18%, first in the sector by price gain, volume ratio 2.09.
• Positive logic: carry-over logic being cashed in by flows. The fulcrums are Q2 commercial-bank NIM recovering to 1.41% (the first quarter-on-quarter increase since Q1 2022) and the big-six interim dividend of RMB 220.989 billion (+7.98%, the largest on record). ⚠️ Both were disclosed on 8/28 and are outside this window; this issue does not treat them as new catalysts.
• Fundamentals: closed 9/1 at RMB 5.48, P/E 7.34, total market cap RMB 658.1 billion. One of the six state-owned large banks.
• Technical sentiment: its volume ratio of 2.09 is the highest among the big six, and its turnover of 0.60% is also above ICBC (0.13%), BOC (0.15%) and ABC (0.12%), showing incremental money picked it first.
• Final judgment: watch closely. The reason for giving it a recommendation slot is "yesterday's money is here", not "new news came out". Banks' intraday elasticity is only 2%–4%, so it suits allocation rather than board-chasing; and several banks are already at all-time highs, so the position is not low.

53 深粮控股000019SZ main ±10% | Watch only (downgraded from a recommendation slot this time) | 58 pts · Shenzhen Cereals Holdings

⚠️ The draft of this issue placed it in recommendation slot No.3. It was downgraded before publication after the interim report was reviewed; the process follows.

• Related news: same as 5.1, overnight overseas agriculture confirmation. No new company filing.
• Technical sentiment (the draft's reason for the recommendation slot, and this part still holds): limit-up on 9/1, seal ratio 0.653, 2nd in quality among the 14 agricultural limit-ups; turnover of only 6.01%, the lowest tier among agricultural limit-ups. Compared with 神农种业's 37.26% and 敦煌种业's 24.22% — its chips are locked, not churned out. That trading fact has not changed.
• ⚠️ But after the fundamentals were filled in, two pieces of evidence keep it out of the recommendation slots:

Item	2026H1	2025H1	YoY
Total operating revenue	RMB 2.3876 billion	RMB 2.3842 billion	+0.14%
Net profit attributable to parent	RMB 111.87 million	RMB 176.02 million	−36.4%
Ex-non-recurring net profit	RMB 85.81 million	RMB 158.01 million	−45.7%
Net operating cash flow	RMB −52.88 million	RMB +737.3 million	Turned from positive to negative

Table scrolls horizontally

• ⚠️ More importantly, the transmission direction may be reversed. Its main business includes food processing and manufacturing as well as wholesale and retail, and the processing business is on the cost side of grain prices, not the beneficiary side. This is the same phenomenon as the product-level data for 苏垦农发 in §5.1: rice gross margin −3.85pp, malt −4.34pp, precisely the result of being squeezed by rising input prices. This issue cannot write "the processing end is squeezed" on one page and count a processing company as a direct beneficiary of grain prices on another. ⚠️ A technical statement in the draft is also corrected: the draft wrote "rising grain prices are directly reflected in trading spreads and inventory revaluation" — inventory is measured at the lower of cost and net realizable value, so a price rise generates no revaluation gain; it is only realized on sale. That statement has been deleted.
• Final judgment: watch only. The seal quality is genuinely good, but ex-non-recurring is −45.7%, operating cash flow has turned negative, and grain prices are a cost rather than revenue for its processing business. A trading signal cannot substitute for an industrial judgment — this is the same error as §5.8 福建金森 in a different form.

53b 北大荒600598PassSH main ±10% (downgraded from a recommendation slot this time) | 30 pts · Beidahuang

⚠️ The draft of this issue placed it in recommendation slot No.5, on the grounds of "the same logic as 苏垦农发 (own-farmland cultivation)". That reasoning is wrong, wrong in a very specific way, and worth listing separately.

• ⚠️ It is not a company that "sells grain", it is a company that "collects rent". Land contracting fee income accounts for about 72% of its operating revenue, and the contracting fee is a fixed price — rising grain prices do not enter its income statement in the current period at all.
• ⚠️ It and 苏垦农发 are not the same logic; they are the two ends of the same contract. 苏垦 pays RMB 361.64 of contracting fee per mu; 北大荒 collects a contracting fee per mu. Writing the two as "the same logic" is the direct consequence of pushing the "sell grain vs sell seed" dichotomy too far — it does neither.
• Fundamentals (badly understated in the draft): 2026H1 net profit attributable to parent RMB −536.5 million, ex-non-recurring RMB −208.7 million, operating revenue RMB 2.128 billion, −29.39% YoY. The main cause of the loss is back taxes and late fees of about RMB 1.41 billion (a one-off item). ⚠️ The draft only wrote "P/E −69.45, loss-making on the current caliber", without stating that this is a loss on the order of RMB 500 million, a nearly 30% revenue decline, and the back-tax event behind it. This is the biggest disclosure shortfall caught by quality control in this issue.
• ⚠️ There is a knock-on correction as well: §8.2 originally set "the appearance in the limit-up list of names with real grain businesses such as 苏垦农发 / 北大荒 / 深粮控股" as a confirmation signal for branch quality improvement — 北大荒 does not meet that definition, and the signal has been corrected in §8.2.
• Final judgment: Pass. Its trading picture on 9/1 (+7.44%, no limit-up, turnover 6.16%) is not bad, but the "beneficiary of rising grain prices" logic does not hold for it in the current period, while its current statements show a loss on the order of RMB 500 million and revenue −29.39%.

54 中国太保601601SH main ±10% | Watch closely | 70 pts · China Pacific Insurance

• Related news: the insurance sector on 9/1 was +3.16% with net inflow of RMB 2.716 billion (2nd in the market), 5 up and 0 down. 太保 +3.34%, volume ratio 1.86.
• Positive logic: an equity market recovery lifts investment returns + repair of an extremely low valuation. Also carry-over logic.
• Fundamentals: closed 9/1 at RMB 33.07, P/E 5.64, the lowest valuation among the 22 listed stocks in this issue, total market cap RMB 318.1 billion.
• Final judgment: watch closely. Same "direction the money is going" as banks, but with slightly higher elasticity. The risk is that if the equity market weakens today, the insurance investment-side narrative weakens with it.

55 中国银行601988SH main ±10% | Watch closely | 68 pts · Bank of China

Closed 9/1 at RMB 6.67, +2.46%, combined A+H market cap topped RMB 2 trillion, an all-time high intraday. P/E 8.63. Same logic as 5.2. The position is no longer low; after a record high it needs fresh money to take the baton, while the catalyst is carry-over.

56 中国平安601318SH main ±10% | Watch closely | 67 pts · Ping An

Closed 9/1 at RMB 57.23, +2.53%, P/E 6.50, total market cap RMB 1,036.3 billion. Same logic as 5.4, larger in size and lower in elasticity.

57 福建金森002679SZ main ±10% | Watch only | 60 pts · Fujian Jinsen

• Technical sentiment: the strongest tier in the market. 3 consecutive boards, first sealed at 09:25:00 (sealed at the open), seal orders of RMB 353 million against turnover of only RMB 79 million, seal ratio 4.469, 3rd in the entire market.
• ⚠️ But its industrial logic is the most remote in the branch. Its main businesses are forest cultivation and afforestation, forest maintenance, and timber production and sales; it holds the largest FSC-certified timber-forest stock volume among southern collective forest areas. There is no direct causal chain between rising grain prices and timber prices — it falls in the "pure concept mapping" tier of iron rule 2.
• Final judgment: watch only. Seal quality and logic hardness are separated to the extreme in this name. It is the strongest on trading, but this issue cannot give it a recommendation slot just because "it seals well" — that would be substituting a trading signal for an industrial judgment.

58 常山北明000158SZ main ±10% | Watch only | 58 pts · Changshan Beiming

• Limit-up on 9/1, main-force net inflow of RMB 167 million, the only one of the 83 limit-up stocks above RMB 150 million, volume ratio 5.24. Subsidiary Beiming Software is a core strategic partner of Huawei (Kunpeng/Ascend/HarmonyOS).
• ⚠️ P/E 1,701x, total market cap RMB 22.189 billion. Against yesterday's backdrop of "limit-up stocks in aggregate main-force net outflow of RMB 6.951 billion", it has the best flow structure of any limit-up stock, and that fact is real and rare.
• ⚠️ But the overnight Philadelphia Semiconductor Index at −2.14% is a direct headwind, and the valuation offers no support whatsoever.
• Final judgment: watch only. The flow evidence is strong and the valuation evidence is extremely weak; when the two point in opposite directions, this issue chooses not to give a recommendation slot.

59 中文在线300364ChiNext ±20% | Watch only | 56 pts · Chinese Online

• Related news: the "Administrative Measures for the Development of Micro Short Dramas" took effect on 9/1. ⚠️ The measures were approved on 7/27 and reported by Xinhua on 7/31; 9/1 is the "effective date", not the "disclosure date". Traced back to the first disclosure date, the expectation gap should be heavily discounted.
• ✅ Short drama is real, already-delivered revenue, not a concept: 2026H1 short drama and IP derivatives revenue was RMB 411.3 million, 71.16% of total revenue, +108.72% YoY, with a gross margin of 48.62% (+13.59pp). This was verified against the interim report's revenue-breakdown table and holds up.
• ⚠️ But "AI revenue" does not exist. Searching the full interim report, AI's role in the business descriptions is entirely that of a production tool and a cost-reduction means (script writing, film production, precision distribution); there is no "AI business" revenue line in the revenue-breakdown table. The correct statement is: AI is a delivered cost-side lever for it (gross margin 31.87%→49.58%, +17.7pp is the evidence), not a revenue line. Using "AI revenue share" as a reason to recommend is unfounded.
• ⚠️ In the same report AI also works in reverse: the digital reading main business (online literature) saw revenue −54.91%, which the company attributes to "the dual test of platform rule changes and a surge in AI-generated content". AI helped short drama on the cost side but cannibalized the online-literature main business on the revenue side.
• ⚠️ The most important point: shift the caliber from "H1" to "single quarter" and the picture completely reverses.

	2025Q2	2026Q1	2026Q2
Revenue (RMB million)	323	310	268
YoY	—	+33.12%	−17.03%

Table scrolls horizontally

2026Q2 revenue of RMB 268 million was −17.0% YoY and −13.6% QoQ, the lowest single quarter since 2025Q1, and 58.5% below the 2025Q4 peak of RMB 646 million. All of that +3.85% H1 growth came from Q1; Q2 has already turned negative. The same is true of the short drama business: 2026H1's RMB 411 million against 2025H2's implied RMB 699 million is −41.2% half-on-half. "Short drama +108.72% in H1" is an H1-vs-H1 caliber that masks the fact that the most recent quarter is already down both YoY and QoQ. The essence of the so-called "narrowing losses and jumping gross margin" is "shrinking in exchange for margin" — revenue is contracting and costs are being cut faster, not scale expansion.(Single-quarter figures are derived by differencing cumulative figures quarter by quarter and can be recomputed; the company's interim report offers no explanation of the single-quarter decline.)

• ⚠️ The balance sheet is the worst on the whole list: P/B about 78.3x (net assets attributable to parent shrank within a year from RMB 765.5 million to RMB 263.4 million, −65.6%); P/E (TTM) is N/A (TTM net profit attributable to parent RMB −488 million); gearing ratio 74.76%, current ratio 0.62 / quick ratio 0.44, net current liability gap RMB −367.1 million; short-term interest-bearing debt of RMB 359 million already exceeds cash of RMB 276.7 million; net goodwill of RMB 116.64 million = 44.3% of net assets attributable to parent, 100% from a single target (Beijing Hanmu Chunhua), with no impairment recognized this period.
• ⚠️ Actual controller pledges, disclosed exactly on 9/1: Tong Zhilei added a pledge of 8.51 million shares on 8/31 (the day of the limit-up), bringing cumulative pledges to 39.40 million shares = 45.18% of their holdings, with the stated purpose "to support the listed company's production and operations". This points to the same thing as the current ratio of 0.62 and the net current liability gap of RMB −367 million: working capital at the listed-company level is tight.
• Final judgment: watch only. The short drama revenue is real, the AI cost reduction is real, and the RMB 174.7 million of institutional net buying is real; but Q2 has already turned to contraction, P/B is 78x, the current ratio is 0.62, goodwill is 44.3% of net assets, the actual controller's pledge ratio is 45.18%, and the LHB that day was a net sell of RMB 162.2 million with a gap-up fade on the tape. The only premise on which a 78x P/B is payable is that the market is pricing the option value of future AI short drama, and the latest observation of that option (2026Q2) is contraction, not expansion.

510 [Special Feature] 阳光股份000608PassSZ main ±10% (further downgraded this time from "watch only") | 44 pts · Yangguang Co.

⚠️ It ranks 14th by total score and is no longer in the Top 10. This issue still gives it its own section because the judgment process behind "it is the only brand-new company-level catalyst in the window, yet it was refuted point by point by primary filings" is more worth writing out clearly than any stock on the list.

It is the only brand-new, company-level hard catalyst in this window whose share price has not reacted, and on the "expectation gap" dimension it should have earned a recommendation slot. What this issue ultimately gives is a Pass. All nine pieces of evidence below come from the company's own statutory disclosures (the 9/1 contract filing, the 8/3 outbound investment filing, the 8/27 interim report), not outside speculation — this re-rating process is more worth reading than the conclusion.

• Related news: post-close filing on 9/1; controlled subsidiary Shenzhen Yangguang Jinhui Technology signed a comprehensive server leasing service contract with "Company A", with a service period of 5 years from the billing commencement date and a total price including tax of RMB 633 million. (Original filing⁵ · Jiemian flash¹)

① First, get the annualization right: the revenue elasticity is real, but it must be computed excluding tax. The filing states a tax-inclusive total of RMB 633 million with a VAT rate of 6%, so the tax-exclusive figure is RMB 597.17 million, i.e. about RMB 119.43 million a year excluding tax. Against TTM revenue of RMB 326.94 million, that is an annualized share of 36.53% — a genuinely meaningful magnitude, larger than the initial estimate.

② ⚠️ But RMB 633 million is a nominal ceiling, not locked-in revenue. The filing's pricing method is "monthly comprehensive service fee × number of units", "paid in arrears by calendar month". Nowhere in the text is there any minimum serviced unit count or floor service fee. In other words, revenue is recognized month by month according to units actually deployed, and RMB 633 million is the ceiling under five years of full load, not a locked-in minimum commitment. Any characterization of it as "backlog" exceeds what the filing supports — this is an upgraded version of the "cumulative-period order figures" trap in memory: it is not even a cumulative commitment.

③ ⚠️ The profit contribution could be near zero or negative. In the filing the company states the business has a gross margin of about 10%–21%. Annualized gross profit = RMB 119.43 million × 10%~21% = RMB 11.94 million ~ RMB 25.08 million; and the listed company holds only 60% of Yangguang Jinhui, leaving RMB 7.17 million ~ RMB 15.05 million attributable to parent. ⚠️ A control group is needed here, and the draft used the wrong one; the correction follows: the draft compared it against "the company's existing RMB 660 million of shareholder loans, whose annual interest is about RMB 19.14 million" — but that RMB 660 million is pre-existing debt from before signing, and its interest is not generated by this contract and should not be charged to it. The correct comparison is "the incremental depreciation plus incremental financing cost from the server purchases needed to perform this contract", and the company has not disclosed that increment. In addition, whether the company's self-reported 10%–21% gross margin already includes server depreciation is also undisclosed; if it does, deducting interest again would be double-counting. The retained conclusion is: the contract's annualized gross profit attributable to parent tops out at RMB 15.05 million, a scale insufficient to cover the financing cost of the capital expenditure added for it (the incremental caliber pending disclosure); the company itself also wrote that "financial expenses will increase to some extent". The revenue elasticity is real; the earnings elasticity is not.

④ ⚠️ The contracting entity is a newly purchased company with negative net assets. Shenzhen Yangguang Jinhui was acquired on 2026-05-25 via a 60% stake purchase for RMB 0.1 million, with net assets of only RMB 9,288.10 on the acquisition date, and at end-H1 net assets of RMB −0.7003 million (negative), revenue of RMB 0.8818 million and net profit of RMB −0.5036 million. A five-year contract with a nominal value of RMB 633 million is being signed by an entity bought three months ago for RMB 0.1 million with negative net assets.

⑤ ⚠️ The counterparty is completely anonymous, with no performance safeguards whatsoever. The filing says that for reasons of commercial secrecy it "has gone through the exemption-from-disclosure procedure and will not disclose the specific details of the counterparty" — no name, no industry, no background, no statement of related-party relationships. Payment is monthly in arrears, with no prepayment, no deposit, no guarantee, no performance bond, and no minimum purchase volume. The listed company buys the equipment at its own expense first, then chases payment monthly, bearing the credit risk one-sidedly. Risk item 2 in the company's own risk warnings concedes the counterparty faces a risk of "losing the ability to perform".

⑥ ⚠️ The servers have not been bought, the money is not in place, and the project only goes live in October 2027. The filing says the servers "still need to be procured on a coordinated basis", and the funding source for the RMB 980 million intelligent computing center is "own funds or external financing (including shareholder loans)", with no signed credit facility or approved loan disclosed; the company itself lists three risks: "fundraising risk", "excessive debt pressure risk" and "liquidity risk". The filing also gives no billing commencement date (it only says "5 years from the service billing date").

⑦ ⚠️ The issuer itself denies the materiality of this contract. From the 9/1 filing: "the company's overall operating fundamentals have not changed materially, and the new business does not constitute a material impact on the company's overall production and operations". Intelligent computing revenue in the first half was only RMB 0.9347 million, 0.78% of revenue.

On "only RMB 29 million on the books" — this issue must correct a widely circulated stale figure. That number does come from the company's 8/3 filing, but its caliber is March 31, 2026 and it has been superseded by the interim report: cash and equivalents at 6/30 were RMB 232.82 million (RMB 224.66 million usable). ⚠️ But the source of that money must be spelled out: H1 operating cash flow was RMB −82.50 million (RMB +44.75 million in the prior-year period, turning from positive to negative), while net cash from financing was RMB +332.92 million, which the interim report notes was "mainly due to an increase in shareholder loans this period" — the cash is 100% from controlling-shareholder loans, not earned from operations. So the "light-versus-heavy inversion" still holds; it just needs restating as: usable cash of RMB 225 million + about RMB 190 million of remaining shareholder credit ≈ RMB 410 million of funds on hand must simultaneously cover the RMB 980 million intelligent computing center and the server purchases required by the RMB 633 million contract, while operations themselves are in net outflow.

⑧ Fundamental backdrop and valuation: five consecutive years of losses attributable to parent (2022 RMB −384 million / 2023 RMB −227 million / 2024 RMB −129 million / 2025 RMB −218 million / 2026H1 RMB −51.93 million); H1 revenue RMB 120.20 million (−10.83%), ex-non-recurring RMB −57.99 million; gearing ratio 46.38%. In 2025 the company was placed under delisting-risk warning for "negative net profit and post-deduction revenue < RMB 300 million", and it was only lifted on 2026-05-21 (2025 post-deduction revenue was RMB 332.74 million, only RMB 32.74 million above the line). ⚠️ If H2 continues H1's −10.83%, full-year post-deduction revenue will land near the RMB 300 million line — a variable almost nobody is discussing yet one that decides whether the cap is re-imposed after the 2026 annual report. (This is an estimate, not company disclosure.) The hardest sentence on valuation comes from the company itself — from the interim report: "the latest P/B of the real estate industry is 0.83 and the company's P/B is 2.65... the valuation premium lacks adequate fundamental support and there is a risk of valuation retracement".

⑨ ⚠️ And the "cross-industry into intelligent computing" narrative is not new; the market has already priced one round of it and is giving it back. On 8/3 the disclosure of the RMB 980 million intelligent computing center produced a limit-up that day, and on 8/13 the price hit an intraday high of RMB 8.27; on 9/1 it closed at RMB 7.25, already 12.3% below that high, with volume shrinking from 564,000 lots on 8/14 to 245,000 lots. What is new is only this specific contract, not the story.

• Final judgment: Pass (a further downgrade from "watch only" under yesterday's framework). This is not a "position too high" type of Pass, it is a "narrative refuted by primary evidence" type of Pass — historical reconciliation shows the former averaged +6.18% and often outperformed, while the latter averaged −0.21%, the two mean entirely different things, please do not read them interchangeably. ⚠️ But to be honest: it may well gap up or even limit-up today. "Pass" is not a forecast that it will fall; it says that per the original filing, this contract has no minimum commitment, no prepayment, an unverifiable counterparty, unpurchased equipment, unsecured funding, a project that only goes live in October 2027, and an annualized gross profit attributable to parent capped at RMB 15.05 million that cannot even cover interest on existing borrowings — and the company itself says it "does not constitute a material impact". Betting a recommendation slot on a contract like this does not meet this issue's evidence standard. One thing to track intraday today: whether the Shenzhen Stock Exchange issues a letter of concern regarding the filing. Given the combination of "anonymous counterparty + cross-industry move + three abnormal-movement filings by the company in June", this is a realistic possibility; if a letter is issued, the inquiry will most likely focus on the counterparty's identity, the revenue-recognition cadence and the server-purchase funding, and its reply would directly change the judgment above. As of the morning of 9/2 none has appeared.

---

6. Downgrade and Avoidance List

⚠️ Naming note: this section keeps the function of the "Pass list" but has been renamed, because some of the stocks collected in 6.2/6.3/6.4 (福建金森 60 pts, 深粮控股 58 pts, 常山北明 58 pts, 中文在线 56 pts) are rated "watch only" rather than "Pass" in this issue — they are downgraded, not excluded. This issue's rating is marked in the last column of each row.

6.1 Category One: Institutions and the Main Force Are Distributing (hardest evidence, avoid first)

6.1b A Special Category: the Only Brand-New Catalyst in the Window, but the Narrative Is Refuted Point by Point by Primary Filings

6.1c ⚠️ The Companies Concerned State in Their Filings That There Is "No Transmission This Period" (newly added in this issue)

⚠️ Wording calibration (important): the draft titled this category "refuted by company filings", which was one notch too strong. The phrase "a different link from the commodity corn market" in an abnormal-movement filing is a standard disclaimer produced under exchange verification pressure, and a company has a statutory incentive to play down associations when its share price rises abnormally. What it strictly proves is that "the company is unwilling to endorse this price move, and there is no transmission this period", not that "this chain will never exist". §2.2.1 of this issue itself writes that this is a timing question (it could at the earliest show up at the autumn 2026 ordering conference). One possibility in the opposite direction should also be added: rising grain prices could in fact accelerate seed destocking — "oversupply" is a current-period state, not an immutable variable. The Pass rationale for this category should therefore be read as "no current-period transmission at the current point in time; the verification window is the autumn 2026 ordering conference and the annual report", not "the logic is permanently falsified".

⚠️ How this differs from the "position too high" type of Pass: the two have historically performed in opposite ways — the "position too high" group averaged +6.18% with 7/7 outperforming, while the "logic does not hold" group averaged −0.21%. ⚠️ But the sample size of these statistics must be stated alongside: n=7 for each group, taken from a single trading day's one-off retrospective reconciliation; it does not constitute a cross-period statistical prior and can only be treated as a directional hint. ⚠️ Also: where a single stock qualifies for both categories (e.g. 万向德农, which both "states no transmission this period in its filing" and is "extremely extended"), this issue always files it under the category with the stronger evidence and does not double-count it, so readers are not left unsure which prior to apply.

6.2 Category Two: Position Too High / Already Priced (downgraded only, not removed)

⚠️ Special note (from historical lessons): stocks passed on for "position too high" are not necessarily wrong on direction. Historical reconciliation shows the group passed on for "logic does not hold" averaged −0.21%, while the group passed on for "position too high" averaged +6.18% with 7/7 outperforming. This category is only downgraded, not removed from the list; please do not read "Pass" as "it will fall".

6.3 Category Three: Valuation Does Not Support (downgraded only, not removed)

6.4 Category Four: Logic Does Not Hold / Far-End Mapping

6.5 Directions Explicitly Avoided (Not Individual Stocks)

1. Semiconductors / memory / electronic components — net outflow of RMB 17.448 billion on 9/1 (the largest in the market), 21 up and 165 down; overnight the Philadelphia Semiconductor Index fell another 2.14%, the second consecutive blow. No bottom-fishing today.
2. Relaying yesterday's limit-up tier as a whole — of the 88 limit-ups on 8/31, 9/1 saw 43 up and 45 down, a relay win rate below fifty percent; and of the 83 limit-ups on 9/1, 65 were first boards with aggregate main-force net outflow of RMB 6.951 billion. The board-chasing strategy has low expected value under the current structure.
3. First-day speculation in new listings — 马矿股份 (601123) +290.98%, 贝特利 (Beiteli, 301697) +196.69%, with institutions net selling in both.

---

7. Intra-Branch Ranking

7.1 Large Financials

Hardest name: 中国银行 (Bank of China) (core name, the steadiest on fundamentals and market cap). Highest trading recognizability: 邮储银行 (Postal Savings Bank) (first in both gain and volume ratio). Who is following: 工商银行 (ICBC, 601398) (a gain of only 0.99%, clearly lagging).

7.2 Agriculture / Grain

The main ranking axis of this branch has changed after the verification in §2.2.1: the first criterion is no longer "did it limit-up" or "seal quality", but "how does it actually make money". Grain sellers benefit directly in the current period with costs locked; seed sellers are stated by two of the companies concerned to have no transmission this period; and the rent collector (北大荒) is neither.

⚠️ Positions 2 and 3 in this table are the only two recommendation slots downgraded during this issue's quality-control stage. 深粮控股's processing business is on the cost side of grain prices (the same thing as "rice −3.85pp, malt −4.34pp squeezed by input prices" written in §5.1 of this issue), and its H1 ex-non-recurring is −45.7%; 北大荒 derives about 72% of revenue from land contracting fees, which is collecting rent rather than selling grain, standing at exactly the opposite end of the same kind of contract as 苏垦. The "sell grain vs sell seed" dichotomy was pushed too far on these two names, and this has been corrected.

Hardest name: 苏垦农发 (Suken Agriculture Development) (sells grain + costs locked + grows exactly the rice and wheat the narrative points to + lowest position). Highest trading recognizability: 福建金森 (Fujian Jinsen) (seal ratio 4.469) — but it has the most remote logic (forestry); the two are completely separated in this branch. Who is following: 敦煌种业 (Dunhuang Seed), 金健米业 (Golden Health Rice) (hollow seals, high turnover). ⚠️ Who is passed and why: all 5 seed stocks in the top 9 are either Pass or watch only — not because they have risen too much, but because "Chinese corn futures fell on 9/1" + "万向德农 and 登海 state in their filings that the industry is oversupplied with selling prices under pressure". This is a conclusion completely at odds with the market's mainstream explanation; the evidence is in §2.2.1.

7.3 AI Compute / Media

Not a single name in this branch earns a recommendation slot, and this is the gap in this issue that most needs explaining. The reason is that the evidence structure of the four names is mutually misaligned: the one with the newest catalyst and the most unmoved share price (阳光股份) has its contract refuted point by point by the original filing; the one with the best flow structure (常山北明, first in the market by main-force net inflow) trades at a P/E of 1,701x; the one institutions actually bought with real money (中文在线, institutional net buy of RMB 175 million) has a P/B of 78.29 and is loss-making; and the biggest gainer (芒果超媒) is being net sold by institutions. "Every piece of evidence sits on a different stock" — this is precisely the sign that a branch has not yet formed a coherent force, and the reason this issue places no recommendation slot here.

---

8. Opening Verification Signals for Today

8.1 Call Auction Signals (09:15–09:25)

• 阳光股份 (Yangguang Co., 000608): the RMB 633 million contract is the only brand-new catalyst in the window, so the auction will most likely gap up. This issue has already given it a Pass (§5.11); it is treated here only as an observation sample, not as a tradable name.
  • Observational significance: it is the best sample today for testing "whether the market reads the original filing". If the gap-up is priced off the RMB 633 million headline total (i.e. gap-up >7% or a one-line board), the market is reading the headline; if it gaps up <3% and then weakens, the market has read "no floor clause + anonymous counterparty".
  • ⚠️ Watch one thing intraday: whether the Shenzhen Stock Exchange issues a letter of concern. The combination of an anonymous counterparty + a cross-industry move + three abnormal-movement filings in June makes this probability non-trivial, and if it happens it will directly change the pricing.
• 苏垦农发 (Suken Agriculture Development, 601952): with a gap-up of 0%–3%, the premise of §5.1 still holds (a non-limit-up stock absorbing the overnight overseas move). A gap-up >5% worsens the odds and downgrades it to observation.
• 深粮控股 (Shenzhen Cereals Holdings, 000019): already limit-up yesterday. A gap-up <3% with a quick re-seal = the logic continues; a gap-up >5% means no participation (§5.3 has preset this condition).
• Agriculture as a whole: if 神农种业 (300189) and 敦煌种业 (600354) gap down, it confirms that yesterday's 37% / 24% turnover really was distribution, and the branch is downgraded for the day.

8.2 Sector Signals (09:30–10:30)

• Confirmation signal: the agriculture branch sees 3 or more names limit-up quickly, and among them a name like 苏垦农发 (601952) that "owns farmland and makes money selling grain" — showing money is rotating from seed concepts to farming entities, and the branch's observation level is upgraded. ⚠️ This item has been corrected per quality control: the draft also listed 北大荒 (600598) and 深粮控股 (000019) as confirmation samples of "real grain businesses", which is wrong — 北大荒 gets about 72% of revenue from land contracting fees (collecting rent), and 深粮控股 also runs food processing (the cost side of grain prices). Gains in either cannot be used to confirm the "grain price transmission" logic.
• ⚠️ Falsification signal: only seed stocks and loss-makers limit-up while 苏垦农发 (601952) lags — showing this is still concept rotation rather than logic deepening, and this issue will not follow up.
• Large financials: if banks/insurance keep seeing net inflows with 邮储 and 太保 rising on volume, then §2.1's "the money is here" holds and positions can be maintained; if banks spike and fade, it means that batch of 8/28 carry-over positives is exhausted.
• Semiconductors: a gap-down is expected (overnight SOX −2.14%). The key is whether it sells off on volume — if semiconductors see another net outflow above RMB 10 billion, market-wide risk appetite will be dragged down and the observation level of the entire list is downgraded in step.

8.3 Per-Stock Signals (All Day)

• A limit-up with seal ratio >0.5 and turnover <15% = an effective seal (cf. 深粮控股 yesterday, 0.653 / 6.01%).
• A limit-up with seal ratio <0.1 and turnover >25% = a distribution-style limit-up, not included in this issue's watchlist (cf. 神农种业 0.077 / 37.26%, 敦煌种业 0.076 / 24.22%).
• Limit-up stocks with positive large-order net inflow numbered only 21/80 yesterday; if that ratio recovers above 40% today, the main force is back; if it stays below 25%, stay defensive.

8.4 Risk Signals (Downgrade on Sight)

1. Gap up then dive: 苏垦农发 / 深粮控股 gap up and turn negative within half an hour — the agriculture branch is void for the day.
2. The core does not follow: the bank sector turns to net outflow — §2.1's sole pillar (flows) disappears, and the large-financials recommendation slots are downgraded.
3. A lone limit-up: fewer than 5 agricultural limit-ups on the day — branch fade confirmed.
4. Yesterday's strong themes fade: the 8/31 limit-ups were already 43 up and 45 down on 9/1; if that ratio deteriorates further to below 30% today, the historical win rate of relay samples under such a structure is low.
5. A second semiconductor sell-off dragging down the index — the observation level of the entire list is downgraded.
6. VIX has risen to 16.34 (+9.52%), external volatility is picking up; if U.S. equity futures weaken further overnight, every gap-up name today should be discounted.
7. ⚠️ Trading-halt verification risk for 万向德农 (Wanxiang Denong, 600371) (listed separately in this issue; please confirm before the open). It rose about 96.3% cumulatively from 8/17 to 9/1, 8 boards in 11 days, with 3 abnormal-movement filings + 1 risk warning already issued, and the company itself says the price has "seriously deviated from fundamentals and a reasonable valuation range". It also appeared on the LHB 5 times between 8/19 and 8/31, yet does not appear in the 9/1 LHB data (on the same day 金健米业 and 敦煌种业 were both listed for "cumulative three-day price deviation of 20%") — the reason is unconfirmed and this issue labels it "to be verified". ⚠️ Note: the reason for its absence is different from 苏垦农发's and cannot be explained the same way — 苏垦's case is the top-five quota for "price deviation of 7%" being filled, whereas 万向德农, on 6 consecutive boards, should have triggered the "cumulative deviation of 20% over three consecutive trading days" condition, which has no quota limit (金健米业 and 敦煌种业 were listed on exactly that basis the same day). A quota explanation cannot account for its absence, so it is more likely that this issue's 79-row LHB dataset itself has a gap. ⚠️ Entering an exchange verification procedure cannot be ruled out (the public quantitative thresholds are not verifiable, so this issue gives no unfounded quantitative statement such as "close to triggering"). It is recommended to re-verify once before the open using the SSE website's original disclosures: this directly determines whether it can trade normally today. ⚠️ This produces a report-wide downgrade caveat: the institutional-seat conclusions in §2.5 (芒果超媒 institutional net sell of RMB 108.34 million), §6.1 (敦煌种业 institutional net sell of RMB 126.57 million) and elsewhere all rest on this same dataset. If that dataset does have a gap, those conclusions must be discounted accordingly.
8. ⚠️ Signal of a collective turn in seed stocks: if 2 or more of 万向德农, 神农种业 and 登海种业 gap down or break their seals, it means the market has started reading 万向德农's 9/1 filing — at which point the whole seed sub-branch will fade fast, while 苏垦农发 (601952), which actually sells grain, may not follow (the drivers differ). This is the divergence most worth watching today.

---

9. Final Recommendation Conclusions

9.1 The 5 Stocks Most Worth Watching Today

⚠️ Five must-read notes about this list:

⓪ This list was changed once in the final quality-control pass before publication, and the change was large, so it must be stated first. The draft had 深粮控股 (Shenzhen Cereals Holdings, 000019) in slot 3 and 北大荒 (Beidahuang, 600598) in slot 5, both on the grounds of "sells grain rather than seed". After quality control pulled both companies' interim reports, both slots were rejected:

• 北大荒 (600598) is simply not a company that "sells grain"; it is a company that "collects rent". Its land contracting fee income accounts for about 72% of operating revenue, and the contracting fee is a fixed price — rising grain prices do not enter its income statement in the current period at all. It and 苏垦农发 stand at exactly the two ends of the same kind of contracting agreement (苏垦 pays RMB 361.64 per mu, 北大荒 collects it), yet the draft wrote them as "the same logic". This is the direct price of pushing this issue's "sell grain vs sell seed" dichotomy too far — it is neither. And its 2026H1 net profit attributable to parent was RMB −536.5 million, ex-non-recurring RMB −208.7 million, revenue RMB 2.128 billion, −29.39% YoY, with the loss mainly caused by back taxes and late fees of about RMB 1.41 billion (one-off). The draft only wrote "P/E −69.45, loss-making on the current caliber", a severe understatement.
• 深粮控股 (000019)'s fundamentals were not done in the draft, and once done they do not support a recommendation slot: 2026H1 net profit attributable to parent RMB 111.87 million (−36.4% YoY), ex-non-recurring RMB 85.81 million (−45.7% YoY), and net operating cash flow swinging from RMB +737 million in the prior-year period to RMB −52.88 million. Moreover its food processing business is on the cost side of grain prices, not the beneficiary side — the same thing this issue itself writes in §5.1 as "rice −3.85pp, malt −4.34pp, the processing end squeezed by rising input prices". The same report cannot say the processing end is squeezed and count a processing company as a direct beneficiary at the same time. Both have been moved out of the recommendation slots (relisted in §6.6). This is also why 4 of this issue's 5 recommendation slots are large financials and agriculture is left with only 苏垦农发 — this is not deliberate allocation, it is what remains after verification.

① Not one of the 5 recommendation slots is a strong limit-up stock from yesterday. This is deliberate. Of the 83 limit-ups yesterday, the 80 with main-force net amounts available totaled RMB −6.951 billion with 59 in net outflow; in a structure where the main force is distributing across the board, chasing yesterday's strongest names is the lowest win-rate approach. Four names on the list did not limit-up, and one (邮储) is a large cap. ② Agriculture ends up with only 1 slot (苏垦农发) and not a single seed stock — the reason is not valuation, it is two pieces of primary evidence. First: all three corn seed stocks limit-upped yesterday, while that same day DCE corn futures closed at RMB 2,293/tonne, down 0.26%, and below the 2,328 of 6/30. Since 8/14 corn is +3.01% while 万向德农 is +96.3%. The El Niño narrative points to global wheat and rice, while these companies' wheat seed shares are: 登海 0.72%, 万向德农 0%, 神农 0% — the crop in the narrative and the crop the targets grow are not the same crop. Second: 万向德农's 9/1 abnormal-movement filing states verbatim "the company mainly does corn seed... a different link from the commodity corn market", "the industry as a whole is in a destocking cycle, the market shows oversupply... selling prices are under pressure and gross margin has also declined", and states the share price has "seriously deviated from fundamentals and a reasonable valuation range"; 登海种业's 8/26 filing independently restates the same industry judgment. The issuers themselves dismantled this chain in regulatory filings. Whereas 苏垦农发, 深粮控股 and 北大荒 sell grain itself, with selling prices set at prevailing rates entering the current period and a contract-locked contracting fee on the cost side — these are two completely different businesses that nonetheless rallied under the same concept yesterday.

③ The only brand-new catalyst in the window (阳光股份's RMB 633 million contract) not only failed to earn a recommendation slot, it was downgraded to Pass. It has the highest expectation gap and its share price did not move at all on 9/1 (−0.41%), but after checking the original filing point by point, every prop underneath it collapsed: the contract is priced as "monthly comprehensive service fee × number of units" paid monthly in arrears, with no minimum unit count and no floor clause, so RMB 633 million is a full-load nominal ceiling rather than locked-in revenue; the contracting entity is a newly purchased entity in which 60% was acquired on 2026-05-25 for RMB 0.1 million and whose net assets are negative; the counterparty is completely anonymous, with no prepayment and no guarantee; on the company's self-reported 10%–21% gross margin, annualized gross profit attributable to parent tops out at RMB 15.05 million, which cannot even cover the roughly RMB 19.14 million of annual interest on the existing RMB 660 million of shareholder loans; and in the same filing the company itself writes "does not constitute a material impact". It may still gap up or even limit-up today — "Pass" is not a price forecast, it says this contract is not the thing it is being described as. The full nine pieces of evidence are in §5.10. ④ On valuation, one reminder is required: the headline P/Es of yesterday's limit-up seed stocks badly understate how expensive they truly are. 隆平高科's headline P/E is 257x, but TTM ex-non-recurring is RMB −61.30 million (a loss); 神农种业's headline is 71.6x, with an ex-non-recurring P/E of about 6,722x (of the RMB 107 million 2025 net profit attributable to parent, RMB 103 million came from disposal gains and fair-value changes); 万向德农's headline is −741x, and its TTM net profit attributable to parent is also a loss. For cyclicals at a price trough, a high or negative P/E is the norm, and ranking agricultural stocks by P/E level is itself a misuse; this issue instead uses P/B and "market cap per mu of farmland" as the main criteria.

9.2 The 3 Strongest Branches Today

9.3 Directions Not Recommended for Chasing Today

1. Semiconductors / memory / electronic components — net outflow of RMB 17.448 billion on 9/1 (the largest in the market), 21 up and 165 down; overnight the Philadelphia Semiconductor Index −2.14% and Micron −2.64%, the second consecutive external negative. GigaDevice's RMB 654 million buyback is management's attitude and does not change industry conditions or fund flows; it is not a reason to bottom-fish.
2. ⚠️ The entire seed sub-branch (not just the extended names) — 隆平高科 (000998), 登海种业 (002041), 神农种业 (300189), 万向德农 (600371), 敦煌种业 (600354), 农发种业 (600313). The reason is not position, it is logic: Chinese corn futures fell on 9/1 and are below where they were half a year ago, while 万向德农 and 登海种业 state in their own filings that the industry has "oversupply, selling prices under pressure, declining gross margins", and 万向德农 further states its share price has "seriously deviated from fundamentals and a reasonable valuation range... risk of irrational speculation". See §2.2.1 and §6.1c. ⚠️ 万向德农 (600371) additionally carries exchange verification-procedure risk; please re-check SSE disclosures before the open (§8.4 item 7).
3. Yesterday's extended consecutive boards and high-turnover limit-ups — 万向德农 (6 boards), 海鸥住工 (7 boards, main-force net outflow RMB 625 million), 神农种业 (seal ratio 0.077 / turnover 37.26%), 敦煌种业 (institutional net sell of RMB 127 million). This line already had one collective limit-down on 8/21 (万向德农 −10.00%, 神农 −11.83%, 登海 −9.04%), so the chip structure is fragile.
4. 芒果超媒 (Mango Excellent Media, 300413) — two levels of evidence on the same day 9/1 (LHB institution-dedicated seats 0 buy 1 sell, net sell RMB 108.34 million; full-day large-order main-force net amount RMB −1.542 billion) point the same way, showing institutions are distributing, with the 8/31 LHB Shenzhen Connect seat net sell of RMB 117.7 million as historical corroboration; the catalyst is 8/31 carry-over. Yesterday's downgrade is maintained. (⚠️ These are not three independent samples; see the caliber note in §2.5.)
5. First day of new listings — 马矿股份 (601123, +290.98%, institutions 0 buy 5 sell), 贝特利 (Beiteli, 301697, +196.69%).
6. Pure board-chasing relays — the 8/31 limit-ups were only 43 up and 45 down on 9/1, and 65 of the 83 limit-ups on 9/1 were first boards with aggregate main-force net outflow; the tier structure does not support a relay.

9.4 Final One-Sentence Judgment

Two separations must both be recognized today: one is the separation of flows from prices — yesterday's broad rally of "83 limit-ups with only 6 broken boards" was, in flow terms, a distribution structure with main-force net outflow of RMB 6.951 billion, and the price leaderboard (crop farming +7.51%) points somewhere completely different from the flow leaderboard (banks RMB +8.381 billion); the other is the separation of narrative from targets — the corn seed stocks that limit-upped saw their underlying commodity fall that very day, and the two companies concerned have already written in their own filings that the industry is oversupplied with selling prices under pressure. This issue therefore stands on "the side the money is going to" and "the side where the chain actually works": allocate to large financials, in agriculture buy only entities that sell grain while passing on the entire seed sub-branch, explicitly avoid semiconductors, and downgrade the window's only new order (阳光股份's RMB 633 million) to Pass because it has no floor clause and an anonymous counterparty.

---

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