A-Share Pre-Market Brief | 2026-09-07 Monday

Caliber boundaries of this brief (judge the strength of its conclusions accordingly) ① News window = 2026-09-04 (Friday) 15:00 close → 2026-09-07 (Monday) 07:00, i.e. the entire weekend. Publication times for all filings are taken from East Money's eiTime field (consistent with the art_code timestamp); notice_date is not used (it carries the next trading day), and neither is the date on the filing's signature block (that is the board-meeting date). ② A-share quotes are on the 9/4 closing-settlement basis (Tencent qt.gtimg.cn, readings timestamped 20260904161418–20260904161457). At the time of writing the A-share market has not opened, so no pre-open auction data exists; every "pre-market status" in this text is an inference about the open, not a measurement. ③ Limit-up / limit-down / failed-limit / prior-day limit-up pools are on East Money's official push2ex basis (39 limit-ups, 8 limit-downs, 48 failed limits), with len(pool)==tc verified. ④ Sector moves are on the Tonghuashun 90-industry-sector basis; sector fund flows are on the East Money basis (obtained via the push2delay cluster). These two are not the same ruler and must not be subtracted against each other. ⑤ U.S. equities are on the 2026-09-04 close basis (CNBC, last_time=2026-09-04). The last_time on the gold / crude futures quotes is 2026-09-06T18:59 ET (the Sunday-evening futures reopen), which is not the same moment as the Friday close; both are tagged separately. ⑥ No northbound flow data (daily net buying has not been disclosed since 2024-08-19); this brief gives no northbound figures. ⑦ Limit-ups are judged by board-specific price limits: Shanghai / Shenzhen main board ±10%, ChiNext / STAR Market ±20%, BSE ±30%. Every stock in this text is tagged with its board. ⑧ The 9/4 Dragon-Tiger List was obtained for this brief (66 rows; the exchange publishes after 18:00, and last Friday's post-close recap could not get it because of scheduling — it is filled in here).

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

I. Channels that worked

• Single stocks / indices via Tencent qt.gtimg.cn, which returns PE / PB / total market cap fields in one shot.
• Limit-up pool / prior-day limit-up pool via push2ex, normal; len(pool)==tc verified per the memory item [[limit-updown-must-use-official-pool]] (39==39, 44==44).
• stock_lhb_detail_em('20260904') returned 66 rows normally this time — last Friday's recap reporting TypeError: NoneType was a scheduling problem (publication only at 18:00), not an endpoint failure; this is now closed out after three reproductions and need not be checked again.
• Filings via East Money np-anotice-stock, using market-level parameters ann_type=SHA,CYB,SZA,KCB,BJA plus begin_time/end_time to pull all 800 market-wide filings inside the window, rather than searching by topic word — this is the corrected practice from the memory item [[checklist-skipped-on-favored-branch]] (last Friday's miss of the 海通发展 (Haitong Development) placement had exactly this root cause: filings were only checked for the names it wanted to recommend).
• PDF source text via pdf.dfcfw.com plus local pypdf parsing (WebFetch returns garbage on PDFs, local parsing is mandatory). The three issuance plans from 工行 (ICBC) / 农行 (ABC) / 人保 (PICC), the 龙版传媒 (Longban Media) regulatory warning and the 宇邦新材 (Yubang New Material) tender-offer report have all been parsed to source text.
• U.S. equities via CNBC quote.cnbc.com, with last_time checked symbol by symbol.

II. Channels that failed / data-retrieval faults this time

• ⚠️ East Money's push2.eastmoney.com main cluster is still returning 502 (the same symptom as last Friday, now the second consecutive trading day). Sector fund flows were switched to push2delay.eastmoney.com successfully — this is exactly the alternative route last Friday's recap noted should be written into memory, and the first deliberate use of it worked, so there was no need to degrade to "sector-level fund flows not obtained". Recommend formally writing this route into memory.
• ⚠️ Cailianpress newsflash scraping failed: www.cls.cn/telegraph is front-end rendered and WebFetch only got the shell; both nodeapi newsflash endpoints returned non-JSON (anti-scraping). Weekend news was instead covered by cross-referencing Xinhua / CCTV / China News Service plus the market-wide filings endpoint.
• ⚠️ WebSearch indexing for this weekend (9/5–9/6) is badly lagged, repeatedly returning stale pieces from the 8/30 weekend (40-year mortgages, Earth–Moon laser communication, 长鑫 (CXMT) LPDDR6); copying them through would have written whole passages into the wrong window. This is the same shape as the memory item [[recurring-column-headlines-hide-date]]; each was falsified by publication time and removed. Weekend news in this brief is based on "the market-wide filings endpoint + Xinhua / CCTV source text," not on search summaries.

III. Four cross-checks actively performed on method, all of which changed a conclusion

1. ⚠️⚠️ [The single most important one in this brief] "RMB 360 billion of capital injection = bullish for bank stocks" — rewritten by the issuance terms themselves. The first draft was set to follow the 2025 round and write "the Ministry of Finance injects capital, bank stocks rally broadly." After pulling the source text of the three issuance plans, the pricing mechanism turns out to be entirely different: the 2025 round used a fixed price at a 5%–8% premium to the closing price on the reference date (中行 (BOC) RMB 6.05, 建行 (CCB) RMB 9.27); this time all three use "the pricing reference date is the first day of the issuance period, and the issue price shall be no lower than the average of the prior 20 trading days" — i.e. issuance at market, with no premium cushion. And with 工行 (ICBC) at PB 0.73 and 农行 (ABC) at PB 0.85 both below book, issuing at market dilutes book value per share; only 人保 (PICC) at PB 1.04 is broadly non-dilutive. Without reading the source text, this brief would have produced a conclusion that was directionally right but wrong on both strength and stock selection.
2. ⚠️⚠️ [The first branch that was nearly missed] Overnight the Philadelphia Semiconductor Index was +3.38%, while A-share semis were −2.82% the same day. The first draft's branch table did not contain semiconductors at all (because it was the weakest A-share sector on Friday). After checking CNBC: all three major U.S. indices fell that session (Dow −0.51%, S&P −0.38%, Nasdaq −0.29%), and only the SOX was +3.38%, led by equipment (KLAC +7.32%, LRCX +5.12%, AMAT +4.31%, ASML +4.17%) and memory (MU +6.10%, STX +6.34%, WDC +5.86%), while NVDA was only +0.84% and AVGO +0.21%. This is a cross-market timing gap of "A-shares already sold, the U.S. only just bought," and Monday's A-share open is the first time it faces this price. Per the memory item [[premarket-breadth-before-depth]], sweeping breadth first is what kept the weakest sector from hiding the whole line.
3. ⚠️ ["Change of control = good news" killed by arithmetic] 宇邦新材 (Yubang New Material, 301266) resumes trading Monday, having filed a "change of control + partial tender offer." But both the negotiated transfer price and the offer price are RMB 20.08 per share, against a pre-suspension close of RMB 24.97 — a 19.58% discount. With the offer below market, free-float holders have no reason to tender; and of the 5% offer quota, 3% is already taken up by the former controlling shareholder's pre-acceptance undertaking. "Tender offer" here is not price support; the direction is inverted.
4. ⚠️ [Seal-order quality on the agriculture limit-ups was computed name by name, and the conclusion runs against "the theme is strong"] 16 agriculture names hit the limit, but the seal-order / turnover ratio varies enormously: 海欣食品 (Haixin Foods) 0.449 and 播恩集团 (Boen Group) 0.341 are acceptable, whereas 敦煌种业 (Dunhuang Seed) is only 0.008 (27.93% share turnover), 新希望 (New Hope) only 0.003, and 平潭发展 (Pingtan Development) 0.062 (17.80% share turnover) — the very names drawing the most capital attention are precisely the sloppiest seals. Per the memory item [[seal-ratio-and-lhb-seat-breakdown]], a seal order is only meaningful once divided by turnover value.

IV. What this brief did not obtain / did not do (listed honestly)

• No same-day attribution obtained for A-share semis −2.82% and RMB 16.038 billion of net outflow on 9/4. Searches turned up no negative catalyst that day; the attribution for 浪潮信息 (Inspur Electronic Information, 000977) −9.99% was likewise not obtained in last Friday's recap. This brief keeps to "not found ≠ not there" and does not write "no negative found" as "it was an unjustified sell-off."
• The sponsors / lead underwriters for the three issuances were not obtained (not set out in the body of the issuance plans; they should appear in other filing documents). The "brokers benefit" chain is therefore not given as a branch in this brief.
• No primary source obtained for Micron's HBM capacity expansion: so far only the Korean outlet ETNEWS's local-time 9/3 report plus Chinese reposts; no Micron official statement found. This brief treats it as "media relay" and weights it down.
• The A-share market had not opened at the time of writing, and there is no auction data; all of §8 consists of signals to be verified, not measurements.

---

0. Today in one sentence

The strongest catalyst today is not an A-share one — it is that RMB 360 billion released on Saturday afternoon. Between 15:31 and 16:14 on September 6, eight central financial enterprises disclosed capital-increase plans together, replenishing RMB 360 billion of core tier-1 capital in aggregate, with the Ministry of Finance to issue RMB 300 billion of special treasury bonds in support (Xinhua, CCTV, 9/6). This is the largest concentrated capital replenishment of the financial system in years; the driver type is "policy + event," impact grade S.

But the direction and strength of this positive turn on one clause almost nobody will read — the pricing mechanism. The three A-share issuers (工商银行 ICBC 601398, 农业银行 ABC 601288, 中国人保 PICC 601319) all use "the pricing reference date is the first day of the issuance period, and the issue price shall be no lower than the average of the prior 20 trading days," i.e. issuance at market, and not the 2025 round's "fixed price + premium" (建行 CCB +6.34%, 中行 BOC +7.82%, 交行 BoCom +15.62%, 邮储 PSBC +19.42%). Strip away that layer of premium subsidy and the economic substance of the transaction reverts to what it actually is: an equity raise at PB<1, in which existing shareholders use their book value per share and their dividend per share to buy capital adequacy for the company.

And measured out, what actually gets diluted is not net assets — it is the dividend. The three have almost identical dilution of book value per share (ICBC −0.897%, ABC −0.913%, PICC +0.159%), but dividend-per-share dilution is −3.34% / −6.16% / −4.07%. The 2025 round left behind a clean natural experiment: dividend per share fell at all four participants, by margins almost exactly equal to their share-count dilution, while ICBC and ABC, which did not participate at the time, held flat or rose — this year those two move from the control group into the treatment group. So this is not a simple "buy the banks" story; it simultaneously creates a counter-intuitive comparison: the names that are genuinely undiluted and carry the highest dividend yields are the two that got no money this time — 交通银行 (Bank of Communications, 601328, 4.38%) and 邮储银行 (Postal Savings Bank of China, 601658, 4.03%).

The second strongest is an overnight divergence A-shares have not yet priced. On 9/4 all three U.S. indices fell (Dow −0.51% / S&P −0.38% / Nasdaq −0.29%), and only the Philadelphia Semiconductor Index was +3.38%, led by semiconductor equipment and memory (KLAC +7.32%, MU +6.10%, STX +6.34%, LRCX +5.12%), while NVDA managed only +0.84% and AVGO +0.21% — what got priced that night was the supply-side chain "capacity expansion → equipment orders," not AI compute demand. The same day A-share semis were −2.82%, 13 up / 172 down, with RMB 16.038 billion of net main-force outflow — the weakest sector on the board. The driver (Micron's HBM expansion) was published at 19:00 Beijing time on 9/4, after the A-share close — so Monday's A-share open genuinely is the first time it meets it. But two things must be said in the same breath: this is only media citing anonymous supply-chain people, not a company statement; and A-share equipment makers are not in Micron's supply chain — Micron is in fact lobbying to restrict equipment exports to 长鑫 (CXMT) / 长存 (YMTC) / 中芯 (SMIC). This is a valuation / sentiment channel; the orders will never arrive.

On flows, what was genuinely going in last Friday was media (+RMB 6.174 billion) and agriculture, forestry, animal husbandry & fishery (+RMB 3.016 billion), with agriculture accounting for 16 of the 39 limit-ups. The driver of the agriculture line has been corrected in this brief: it is not "losses so deep you bet on state stockpiling," it is that hog prices are genuinely rebounding sequentially (up from the April low of RMB 8.61/kg to RMB 10.97 on 9/6, +27.4%, with three consecutive up days from 9/4 to 9/6) plus CBOT wheat and corn at 3-year highs. But the "stockpiling expectation" chain can be falsified — the most recent state-reserve action was the 8/18 rotational release (supply added), which runs the other way. And 7 of the 11 limit-up names post losses attributable to parent, with seal-order quality wildly dispersed (敦煌种业 0.008 and 新希望 0.003 are sloppy limits, and those two happen to be exactly the ones drawing the most capital attention).

Pre-market read: expect a higher open, but the structure will tear. Financial heavyweights have the RMB 360 billion event driver, tech has overnight SOX repair momentum, agriculture has theme inertia — the three lines are unrelated to one another, which is a good thing (they are not three copies of the same variable), but it also means capital will be split, and single-stock dispersion may worsen even as the index rises. The Shanghai Composite at 3930.12 is only 1.78% below 4000; if the heavyweights are pulled up by the event driver, an index-level attempt at a breakout is the single thing most worth watching this week.

---

1. News overview

Time window: 2026-09-04 15:00 → 2026-09-07 07:00. Publication times for filings are taken from East Money eiTime.

There were 800 market-wide filings inside the window (East Money np-anotice-stock, begin_time=2026-09-05 / end_time=2026-09-07 plus the 9/4 post-close segment), and the overwhelming majority of those actually published on Saturday and Sunday were routine items — earnings-briefing notices, buyback / shareholder-reduction progress, convertible-bond notices and the like. Beyond the table above, no other filing was found that could constitute a branch-level catalyst. This sentence was written only after scanning them one by one; it is not a "we didn't find anything by searching."

1.1 Previous session (9/4) A-share close review — Monday's starting point

39 limit-ups / 8 limit-downs / 48 failed limits, a 55.2% failure rate. 2,249 advancers, 2,773 decliners. The structure is "heavyweights hold, tech collapses" — the SSE 50 and the STAR 50 are 2.25pp apart.

⚠️ This line matters a great deal today: capital was already rotating out of tech and into the heavyweights last Friday, and then Saturday delivered a RMB 360 billion heavyweight-stock event. Directionally this is a relay, not a reversal.

1.2 Overnight offshore (U.S. 9/4 close, CNBC, last_time=2026-09-04)

Semiconductor single stocks (9/4 close):

⚠️ The key to this table is not "semis went up," it is "which half went up." NVDA +0.84% and AVGO +0.21% barely moved, while equipment and memory rose +4% to +7% as a group. That says what got priced that night was not AI compute demand but the supply-side chain "memory capacity expansion → equipment orders." On that reasoning, what this line can drag along in A-shares on Monday is semiconductor equipment first, and memory only second — and the direction of transmission into memory needs a separate argument (see the risk column of branch 2 in §2).

⚠️ Commodity-caliber reminder: gold futures (COMEX Dec'26) are quoted at 4,475.40 and crude (WTI Oct'26) at 92.06, but the last_time on both quotes is 2026-09-06T18:59 ET (the Sunday-evening futures reopen), not the Friday close, so they cannot be read side by side with Friday's equity indices. On non-farm day spot gold fell as much as 2% intraday to US$4,382.62/oz (a 9/4 intraday reading).

---

2. Strongest positive branches, in descending order

2.1 Branch 1 in detail: who exactly does the RMB 360 billion benefit?

This is the passage in this brief that most deserves being read slowly.

The plan itself (Xinhua 9/6, cross-checked company by company against the issuance plans):

⚠️ The first easy mistake: treating 中国人寿 (China Life, 601628) as a beneficiary. The recipient of the RMB 35 billion injection is China Life Insurance (Group) Company, i.e. the unlisted parent. This brief has verified: 中国人寿 (601628) published no filing at all inside this window (after 9/4 15:00); its most recent filing is the 9/2 17:00 notice of an extraordinary general meeting. The listed company received none of this money and is issuing no shares. This one goes onto the §6 Pass list.

Subscriber detail (taken from the source text of the issuance plans, not from media relay):

• 工商银行 (ICBC, RMB 100 billion): Ministry of Finance RMB 70 billion + China Tobacco RMB 10 billion + Shanghai Tobacco RMB 5 billion + Yunnan China Tobacco RMB 5 billion + Hunan China Tobacco RMB 5 billion + Hunan Tobacco RMB 5 billion. Shares issued not to exceed 10% of pre-issuance total share capital.
• 农业银行 (ABC, RMB 160 billion): Ministry of Finance RMB 130 billion + China Tobacco RMB 10 billion + Jiangsu Tobacco RMB 5 billion + Zhejiang Tobacco RMB 5 billion + Hubei Tobacco RMB 5 billion + Beijing Tobacco RMB 3 billion + Shuangwei Investment RMB 2 billion. Shares issued not to exceed 15% of pre-issuance total share capital.
• 中国人保 (PICC, RMB 15 billion): fully subscribed by the Ministry of Finance. Shares issued not to exceed 10% of pre-issuance total share capital.
• All three carry a 5-year lock-up.

⚠️⚠️ The decisive clause — the pricing mechanism, completely different from the 2025 round:

⚠️ This section corrects two numerical errors in the first draft (per the source text of the four 2025 "Announcement of Issuance Results and Change in Share Capital" filings): CCB's issue price was 9.06 and not 9.27; BOC's was 5.93 and not 6.05; and that round raised RMB 520 billion in total, not RMB 500 billion (the RMB 500 billion figure is the Ministry of Finance's subscription amount, not the total). In addition, the widely circulated "a 5%–8% premium" covers only CCB and BOC — the premiums at BoCom (+15.62%) and PSBC (+19.42%) are 2–3 times that.

⚠️⚠️ But the measured result is far more complicated than the intuition "below book means dilution" — the three have almost identical dilution of book value per share:

(Issue price assumed = the 2026-09-04 close; BVPS taken from the 2026 interim report, disclosed 2026-08-29)

(The CET1 uplift figures are the static estimates in ICBC's and ABC's own plan documents; PICC's plan provides no estimate table, and the +6.15pp is self-computed against interim-report minimum capital of RMB 243.943 billion — please treat it as a self-computed value.)

Three points that must all be made together:

① The dilution of book value per share is in fact very small, and nearly identical across the three (−0.897% / −0.913% / +0.159%). The first draft's line that "ABC's dilution is the largest of the three" is wrong on a BVPS basis. What actually separates them is not book value per share but earnings and dividends per share: ICBC −3.34%, ABC −6.16%, PICC −4.07% — ABC is close to twice ICBC.

② "ABC's 15% cap = bigger dilution" is a misreading. Triggering the 10% / 15% caps would require the share price to fall to RMB 2.806 / RMB 3.048 (−65.5% / −56.2% from current levels). The cap is simply not a binding constraint; issuing at market, ABC uses only 43.8% of its cap. ABC asked for 15% to leave room for price error, not to issue more.

③ ⚠️⚠️ PICC's "accretion" is extremely fragile — this is the most important sentence in the section. PICC's BVPS is RMB 7.69 against a 9/4 close of RMB 7.99, a margin of safety of only 3.9%. More to the point: the assumed issue price PICC itself uses in its plan is RMB 7.41 (the floor of the 20-day average), which is below the BVPS of RMB 7.69 — on the company's own assumption PICC is diluted by 0.159%, not accreted. The "only one not diluted" claim holds only if the final issue price lands above RMB 7.69. This conclusion hangs on a 3.9% thread and cannot be used as a hard conclusion.

⚠️⚠️ On the dividend dimension there is a clean 2025 natural experiment — not a deduction, but something that has already happened:

Dividend per share fell at all four participants, and the size of the fall tracks the share-capital dilution closely (at BoCom and PSBC it is almost exactly the dilution ratio); over the same period the two non-participants were flat or higher. Under an unchanged payout-ratio policy, share dilution passes through to dividend per share 100%.

And this year, ICBC and ABC move from being the "control group" in that experiment to being the "treatment group."

Worse still are the three "Shareholder Return Plans for the Next Three Years (2026-2028)" published the same day as the capital-increase plans, which have been read word for word: they commit only to "annual cash dividends of no less than 10% of net profit," while the actual payout ratio is 30%–31%. Nothing in the disclosed text prevents dividend per share from falling in proportion to the dilution. PICC's dividend filing even states explicitly that it "intends to keep the total distribution unchanged and adjust the per-share distribution ratio accordingly."

Static estimates: ICBC's dividend yield 3.82%→3.69%, ABC's 3.58%→3.36%, PICC's 2.75%→2.64%.

⚠️ Which yields a counter-intuitive but fully evidenced inference — on the dividend dimension, the names that got no injection are actually cleaner:

What is more, BoCom's 4.38% and PSBC's 4.03% are themselves already post-dilution readings from 2025 — their dividends per share were already cut by 14%–17% and the share prices have finished digesting it. Comparing a diluted BoCom / PSBC against an undiluted ICBC / ABC is using two rulers at the same point in time. The comparable basis should be: ICBC 3.69% vs BoCom 4.38%, PSBC 4.03%.

One corroborating point: the current pricing of the six is clearly dividend-driven rather than capital-driven — there is almost no positive correlation between PB and CET1 (BoCom's CET1 is 11.25% yet its PB is only 0.561; CCB's CET1 is 14.24% with a PB of 0.802; PSBC has the lowest CET1 at 10.04% yet a PB of 0.622, higher than BoCom's). In a sector priced off dividends, trading dividends for capital adequacy need not be net positive.

So the full conclusion for this branch — direction and stock selection must be stated separately:

• Direction: still most likely up. The Ministry of Finance and the tobacco system are taking RMB 275 billion of paper at market prices with a 5-year lock-up, which is a very strong signal; the day after the 2025 round was announced (3/31) the large state banks rallied broadly, giving a positive sentiment precedent; and the capital replenishment genuinely does open room for balance-sheet growth.
• But the answer to "which one to buy" is not clean. PICC's non-dilution hangs on a 3.9% margin; ICBC and ABC are definitively diluted with dividends under pressure; the names that are genuinely undiluted, carry higher yields and have already digested dilution are 交通银行 (601328) and 邮储银行 (601658) — whose problem is that they also got no money.
• The exchange ratio in one line: ICBC spends 0.9% of book value per share to buy back "less than one half-year" of natural capital consumption (0.36pp consumed per half-year against 0.33pp replenished here); ABC buys back "exactly one year" of consumption (0.62pp consumed year on year against 0.61pp replenished).
• A new constraint (absent from the 2025 plans): ICBC's plan states explicitly that "following an upgrade in its G-SIB bucket, the Bank will face stricter additional capital requirements"; ABC's plan states that it "applies a 1.5% systemically-important-bank additional capital requirement… and faces future G-SIB bucket-upgrade pressure." If ICBC moves from bucket two to bucket three (additional capital 1.5%→2.0%), this round's +0.33pp is not enough to cover the new regulatory requirement, which implies further capital replenishment down the road.

2.2 Branch 2 in detail: the "timing gap" between A-share semis and the overnight SOX

Facts (all hard data):

• 9/4 A-shares: the Tonghuashun semiconductor sector was −2.82%, 13 up / 172 down; on the East Money basis, net main-force outflow from semiconductors was RMB 16.038 billion, from electronics at the top level RMB 27.224 billion, from digital chip design RMB 7.067 billion and from IC packaging & testing RMB 2.620 billion. It was the weakest line on the board.
• 9/4 U.S. (which opens after the A-share close): the SOX was +3.38%, equipment and memory +4% to +7%, while NVDA / AVGO barely moved.

The sequence has been checked item by item, and it is cleaner than the initial read:

✅ This sequence corrects an error in the first draft and strengthens the conclusion: the first draft wrote "ETNEWS broke it on 9/3, so in theory it was available before the A-share open on 9/4." After checking the article number, ETNews's publication time is confirmed as 9/4 20:00 KST = 9/4 19:00 Beijing time, four hours after the A-share close. So the A-share decline on 9/4 had nothing to do with this news; it genuinely is new information inside the window.

But there are three limits that must be stated honestly, or this branch will be overrated:

1. The driver is not hard — "new" does not mean "hard." The ETNews original reads "According to industry sources on September 3" and cites "an industry source familiar with the matter"; several reposts state explicitly that the target figure has not been officially confirmed. Micron's only public statement on record is the FQ3 earnings call in June 2026, which contains no "100,000 wafers per month" capacity figure. This is media citing anonymous supply-chain people — not a filing, not an earnings call.
2. The attribution cannot be made exclusive. If the SOX's +3.38% were driven purely by "Micron's expansion," Micron itself should have led; in fact the equipment names outran Micron (KLAC +7.32% > MU +6.10%). Competing explanations that day include the WFE capex narrative. ⚠️ Note: Treasury yields actually rose on 9/4 (10Y +2.2bp to 4.784%), so "falling rates lift growth stocks" cannot serve as a competing explanation and has been removed here. This brief tags the attribution of the 9/4 SOX +3.38% as "to be verified · undetermined" and makes no exclusive claim.
3. "Equipment is up" speaks to supply, not demand — and for memory you must separate the clocks.
   • Near term (2026 Q3–Q4): neutral to mildly positive for memory. HBM consumes roughly 3 times the wafer area per bit that standard DRAM does, so if the expansion comes from converting existing DRAM lines, conventional DDR supply actually tightens.
   • Far term (from 2027): clearly negative for memory. ETNews states the expansion sites are "large-scale facility investments" in Taiwan / Singapore / Hiroshima, i.e. skewed toward net new capacity, and the incremental bit supply will weigh on ASPs.
   • Mapping the equipment rally straight onto A-share memory stocks as a positive is taking a supply-side input as demand-side evidence — the direction is inverted.

⚠️⚠️ The fourth point, and the hardest finding in this branch — A-share equipment makers are simply not on Micron's chain, and are institutionally on the opposing side:

• Micron is lobbying the U.S. Congress to cut off A-share equipment makers' customers. Reuters / TrendForce reported in April 2026 that Micron is pushing the MATCH Act, aimed precisely at restricting exports of fabrication equipment to 长鑫 (CXMT), 长江存储 (YMTC) and 中芯国际 (SMIC).
• 北方华创 (Naura) is itself on the U.S. Entity List (the 2024 BIS list).
• Micron's expansion sites (Taiwan / Singapore / Hiroshima, Japan) all sit in U.S.-aligned export-control jurisdictions, so the probability of Chinese equipment entering is very low.
• A-share equipment makers' real customers are domestic fabs: 拓荆科技 (Piotech)'s IPO inquiry response discloses top-five customer concentration of 84%–100%, with 中芯国际 (SMIC) the largest customer.
• The only tangentially related overseas lead (Reuters, 2026-08-05, reporting that Samsung / SK Hynix were evaluating AMEC equipment) has been publicly denied by Samsung, and is in any case unrelated to Micron.

The precise characterisation of the transmission channel (read this sentence as written; do not read it as "equipment names are going to win Micron orders"): Micron expands → global WFE outlook revised up → overseas equipment stocks re-rate → A-share equipment stocks re-rate in sympathy. This is a valuation / sentiment channel, not an order channel. The transmission clock: sentiment the same day, orders never. The true order-side independent variable for A-share equipment makers is capex at 中芯 (SMIC) / 长存 (YMTC) / 长鑫 (CXMT) — and incidentally, 长鑫科技 (CXMT) is already listed on the STAR Market (688825), closing 9/4 at RMB 54.80, −1.35%; that is the demand-side observation point the A-share equipment chain should be watching.

Valuation (9/4 close, Tencent basis) — this line has no cushion:

The memory chain (note the shape of these numbers):

⚠️⚠️ 江波龙 PE 12.44 / PB 6.70, 佰维 PE 12.20 / PB 7.92, 德明利 PE 13.01 / PB 9.08 — "the lowest PEs on the board paired with extremely high PBs" is the textbook shape of a cyclical top, not cheapness. The low PE comes from cycle-peak earnings, and the high PB says the market is already paying a full premium on the assets. This brief gives no lead recommendation slot to the memory tier, only a "watch only."

2.3 Branch 3 in detail: agriculture is "loss-driven," not "price-driven"

⚠️ This section has been substantially rewritten against the measured series. The first draft of this brief wrote "the logic of this line is loss-driven, not price-driven," and that sentence is wrong — hog prices really are rising on a sequential basis.

First, reconcile two statements that look contradictory but are both true:

This is not a contradiction; it is two dimensions of the same thing: the absolute price is climbing out of a deep pit, but last year's base is too high. What the market traded on 9/4 was the former. That day's recap explicitly cited "the national quoted price of external three-way crossbred market hogs rising to RMB 11.09/kg" as the driver for the hog-farming sector. So this round is driven by "a price rebound," not by "losses so deep you bet on stockpiling."

⚠️⚠️ And the "stockpiling expectation" logic can essentially be falsified — three independent pieces of evidence:

1. The most recent state-reserve action was a release, not a purchase — the opposite direction. The 2026 sequence on China's reserve platform: 3/4 purchase of 10,000 tonnes, 4/3 purchase of 10,000 tonnes, 5/14 rotational purchase of 20,000 tonnes, 5/26 rotational release of 30,000 tonnes, 7/2-3 rotational release plus purchase of 5,700 tonnes, 8/18 rotational release of 12,000 tonnes of domestic frozen pork (supply added). The action closest to 9/4 was a release, not a purchase.
2. A hog-to-grain ratio below 5 is not new information. The level-1 alert threshold is a hog-to-grain ratio below 5:1 (per the NDRC's 2021-06-28 announcement). But computed from the Ministry of Agriculture's data for the 2nd week of August, the ratio was already 4.57:1, lower than 4.78 on 9/5 — and that was already three weeks ago; public data show the ratio has been below 6:1 for the better part of a year. This reading constitutes no incremental marginal information for 9/4.
3. Historically, "level-1 alert → stockpiling" has been negative for share prices. Backtesting the two historical level-1 alerts on forward-adjusted daily bars:

(※ 新五丰's 2023 move was a standalone Hunan state-asset restructuring event, not stockpiling-driven.)

Across the two rounds there are 10 observations, and 9 are negative after 20 days. The reason is not complicated: a single stockpiling batch of 10,000–20,000 tonnes, against national pork output of roughly 4.7 million tonnes per month, is only 0.2%–0.4% of one month's output — a price signal, not a supply-demand variable. Historically the level-1 alert is a coincident indicator of the hog-price bottom, not a positive catalyst for share prices — it appears exactly when the farming end is in the most pain.

⚠️ One common caliber error also needs correcting: the break-even point for hog farming is a hog-to-grain ratio of about 7:1, not 6:1. The 7:1 figure is the revised basis in the NDRC's 2021 regulation plan; 6:1 is the older 2015 basis. The first draft carried the old one.

So what actually drove 9/4? Two legs, and the stronger one is grain prices:

• The grain leg (verifiable): CBOT wheat closed 2026-09-02 at 798.75 cents, its highest close in 3 years; CBOT corn closed 9/2 at 538.0, also a 3-year high. Over the past 3 months wheat is +16.59%, corn +12.20%, soybeans +16.73%.

  ⚠️ But this leg has two counter-facts that must be stated: ① CBOT wheat actually fell on 9/4, −2.52% (782.25→762.50); ② domestic DCE corn's front-month contract closed 9/4 at RMB 2,303/tonne, essentially flat over the past 3 months and still −20.5% from its 3-year high. The 3-year high in international grain prices has not yet passed through domestically. And 亚盛 and 敦煌种业 sell domestic goods — this is the thinnest link in the whole logic chain.

• The capacity leg (a genuine policy margin, and right at the edge of the window): at 15:06 on 9/4 there were reports that the Ministry of Agriculture has brought forward the deadline for this round of hog capacity reduction to the end of Q3; the National Bureau of Statistics puts the end-Q2 breeding-sow herd at 37.80 million head, down 2.63 million year on year, leaving only 0.8% of room above the Ministry of Agriculture's normal-holding target of 37.50 million head, revised down in May 2026.

⚠️⚠️ One data point that must be withdrawn: the first draft cited "the August breeding-sow herd fell 2.94% month on month," and this brief could not verify any primary source for it. The only comparable data available are: the Statistics Bureau's end-Q2 figure of 37.80 million head, and July year-on-year readings of −3.54% / −5.64% / −6.64% from Yongyi / Mysteel / Zhuochuang (sample-agency basis). A single-month sequential decline of 2.94% would be a rate of destocking rarely seen since African swine fever; if true it is major information, and if it is a sample-agency or year-on-year figure being cited as an official sequential number, then the whole "destocking is accelerating" leg has no foundation. This brief treats it as "to be verified · not used as an argument" and awards no points on it below.

Hard fundamental evidence (PE at the 9/4 close, Tencent basis): the agriculture limit-ups are broadly loss-making

Two signals pointing in opposite directions, both of which must be given:

• ✅ The core names followed: 牧原 (002714) +4.49% and 温氏 (300498) +6.15% did not hit the limit but clearly followed on rising volume. The theme has core names, it is not purely small-cap froth, which raises its durability.
• ⚠️ But the leaders' seal orders are terrible: 新希望's seal order / turnover is only 0.003, 敦煌种业's only 0.008 (with 27.93% share turnover), and 平潭发展's 0.062 (17.80% share turnover). Extremely high turnover + an extremely thin seal = very easy to break tomorrow. By contrast 海欣食品 (0.449), 播恩集团 (0.341) and 中水渔业 (0.321) trade small but seal solidly. ⚠️ But note: all three of these are Pass in this brief's final conclusions (播恩 for "inverted industry logic," 中水 for "unrelated to the theme," 海欣 for current-period losses and no standalone catalyst — see §6 / §7.3) — a solid seal is only a trading fact and does not constitute a reason to recommend.

⚠️⚠️ Interim-report measurements (2026H1, pulled by the sub-agent, cross-checked name by name against both Tonghuashun and Sina) — 7 of the 11 post losses attributable to parent:

This table contains four findings that change conclusions:

① ⚠️⚠️ 亚盛集团 (600108) itself filed a negative three weeks ago. The 2026-08-15 "Notice on the Impact of Drought on the Company" (eiTime 2026-08-14 16:33:28) states: "the Company's planting bases in Zhangye, Jiuquan and other areas have experienced drought, with severe drought in a few areas… this drought is expected to have a certain adverse impact on the Company's 2026 operating results." The market treats it as an "agricultural-price beneficiary," but it is a producer, not a trader — a yield loss is a direct loss of revenue, and it only benefits on net when "the size of the yield loss < the size of the price rise," a condition that does not currently hold given domestic corn prices have been essentially flat for 3 months. This is the single most directionally clear piece of counter-evidence in this brief's agriculture branch.

② ⚠️⚠️ 天邦食品 (002124) is in judicial pre-restructuring. It has been in pre-restructuring at the Ningbo Intermediate People's Court since 2024-08-09, extended six times, most recently to 2026-11-09; on 9/3 it filed a further update on a major arbitration. Net assets attributable to parent are only RMB 823 million against a half-year loss of RMB 1.543 billion; cash of RMB 155 million against short-term borrowings of RMB 3.059 billion; a 91.2% liability ratio. This is the most risk-concentrated name in the group and does not belong on any watchlist.

③ ⚠️ 平潭发展 (000592)'s mismatch is even more complete than the first draft judged — it is a man-made-board manufacturer. Forestry is only 5.05% of revenue, fibreboard 74.86% and real estate 10.89%. Its appearance in the "planting and forestry" limit-up cluster is purely a mapping artefact of the sector classification tag, with no causal link to agricultural conditions. And a market cap of RMB 14.469 billion sits against H1 revenue of RMB 482 million, with a PB of 8.49, the highest in the group, while the company is loss-making.

④ ⚠️ The industry logic for the three feed names (播恩 001366 / 邦基 603151 / 正虹 000702) runs backwards. Breeding-sow destocking = falling feed demand; international corn / soymeal +12% / +17% over the past 3 months = rising feed costs. Both of the theme narratives — "capacity destocking" and "rising grain prices" — are headwinds rather than tailwinds for feed companies. All three hitting the limit on the same day is a sector-tag effect, not their own industry logic. The simultaneous gross-margin declines at all three (邦基 8.73→5.94%, 正虹 7.41→5.27%) are the direct reading of that squeeze. This point overturns the first draft's rationale for putting 播恩集团 (001366) at No. 9 on the master list — its seal-order quality (0.341) is a trading fact, but the industry logic is inverted.

⑤ Both leaders state in official filings that "there is no new information at the company level." 亚盛's 9/5 abnormal-fluctuation filing (eiTime 9/4 17:31) and 敦煌种业's 9/2 filing (eiTime 9/1 18:44) both confirm "there are no material matters that should have been disclosed but were not." The drivers of this rally are entirely at the industry / commodity / policy level, with no company-level fundamental increment at all.

⑥ A tension that must be acknowledged (not papered over): 敦煌种业 (600354) is fundamentally the best name in the group — revenue +20.9%, net profit attributable to parent +166.3%, ex-non-recurring +188.4%, gross margin up to 48.62%, seeds 89.22% of revenue, a 37.5% liability ratio, and a PE of 33.2, the lowest among the group's profitable names — yet its seal-order quality is the worst on the board (0.008, with 27.93% share turnover), it is only −5.0% from its one-year high, it has already risen 40.4% over 20 days, and it hit limit-down as recently as 9/2. Good fundamentals and whether it breaks its limit tomorrow are two orthogonal questions. This brief handles it as follows: place it in the A tier on fundamentals, while still flagging "elevated position + sloppy seal" risk at the trading level, and do not move it out of the risk warnings simply because the fundamentals are good. Three further haircuts are required: minority interests of RMB 144 million are 50% of net profit of RMB 288 million; OCF is only RMB 19.6 million, a net-profit-to-cash conversion of 0.14×; and the price is already near its one-year high.

2.4 Branch 4: the seven-ministry plan — direction, but no names

The "Implementation Plan for Promoting Coordinated Digital and Green Transformation (2026-2030)" was jointly issued by seven ministries — the Cyberspace Administration, the NDRC, the MIIT, the Ministry of Ecology and Environment, the Ministry of Housing and Urban-Rural Development, the Ministry of Agriculture and the Ministry of Commerce — published at 2026-09-04 18:04:53 (China News Service), inside the window. It contains 4 areas and 14 key tasks; on the consumption side it names: expanding the supply of AI terminals, smart wearables, green appliances, green building materials and new-energy vehicles, and supporting consumption of green smart goods; on the supply side it addresses energy efficiency at data centres and 5G base stations, and a 2030 green-power consumption share target.

How this brief handles it: acknowledge it as a B+ grade policy positive, but give it no lead recommendation slot, for three reasons:

1. The document names no companies, putting it in the highest-risk tier of "pure concept mapping";
2. No quantified subsidies and no funding arrangements (the source text has been checked; it contains only directions and share targets);
3. The coverage is too broad (AI terminals / wearables / appliances / building materials / NEVs / data centres), so broad that any name can claim to benefit, which amounts to no screening power at all.

Usable as a bonus for positions already held in related directions, not as a reason to initiate today.

---

3. Master list of single-stock catalyst strength

Sorted in descending order on the single-stock dimension. The total-score basis is in §4. The board column is tagged with the applicable price limit.

⚠️ The master list carries only 16 names and is not padded out to 30. This window is a weekend window, and only three branches carry genuinely new information (capital increases / overnight semiconductors / the agriculture continuation); none of the other limit-up names has any new catalyst inside the window. Per the "evidence first" rule, it is better to be short than to pad with names that have no evidence.

⚠️ The ordering is the output of the §4 scoring model, recomputed row by row and reconciled. This table has been re-ranked against measured interim-report data, with seven substantive changes versus the first draft: ① 中国人保 falls from 1st to 2nd (the margin of safety on non-dilution is only 3.9%, and at the company's own assumed price it is diluted instead); ② 北方华创 rises to 1st (ex-non-recurring share of 99.1%, the cleanest on the board, plus the only net main-force inflow of the 13 names on 9/4); ③ 兆易创新 added at 3rd (the only name in the memory chain to "pass all three rulers"); ④ 中微公司 cut from 71 to 58 and 拓荆 from 65 to 48 (fair-value changes are 72% / 69% of operating profit, ex-non-recurring TTM PE 150.6 / 172.7); ⑤ 亚盛集团 cut from 57 to 38 and reclassified as Pass (the company itself filed the drought negative); ⑥ 播恩集团 / 海欣食品 removed (the feed names' industry logic is inverted), and 天邦食品 added to the Pass tier (judicial pre-restructuring); ⑦ 北方华创 corrected from 76 pts to 64 pts and moved down to 7th — a recomputation before finalisation found the first draft's component sum was wrong (13+11+9+15+8+6+8−6=64, the first draft mistakenly wrote 76), and its board tag was corrected at the same time: the 002 prefix originally belonged to the SME board, which merged into the SZSE main board in 2021, so the price limit is ±10% and not ChiNext's ±20% (measured over 600 trading days, the stock's largest single-day moves are +10.07% / −10.01%). 龙版传媒 was likewise corrected from 15 to 17 pts for the same summation error.

---

4. Single-stock scoring model (100 points total)

Component scores for three representative names (illustrative, so readers can check the basis):

---

5. Detailed analysis of key names (in descending §3 score order, 11 names in total)

The order in this section has been rearranged to follow §3. All financial data are 2026 interim-report values pulled by this brief itself (cross-checked across Tonghuashun and Sina, with revenue breakdowns from East Money stock_zygc_em), read at 2026-09-07 07:08–07:45; quotes are the 9/4 close.

51 中国人保PICC / 601319SSE main board ±10% | watch closely | 72 pts (cut from 83 pts in the first draft)

• Related news: 2026-09-06 16:13:34 (eiTime), "Plan for Issuing A-Shares to Specific Investors." Raising up to RMB 15 billion, with the Ministry of Finance as the sole investor subscribing in cash for the full amount; shares issued not to exceed 10% of pre-issuance total share capital; 5-year lock-up. Plan source text³
• Fundamental verification (2026H1): total operating revenue RMB 355.115 billion (+9.6%), insurance service revenue RMB 286.873 billion (only +2.4%), net profit attributable to parent RMB 36.745 billion (+38.5%), ex-non-recurring RMB 37.190 billion (+40.2%), BVPS RMB 7.69 (+10.1% versus year-end), weighted ROE 11.2% (+1.7pp). PICC P&C's combined ratio was 94.5% (−0.8pp), with motor at 93.5% and liability insurance at 103.1%, the only underwriting loss. Group comprehensive / core solvency ratios were 246.6% / 197.7%.

  ⚠️ See through the profit mix: within the +38.5% in net profit attributable to parent, insurance service revenue contributes only +2.4% and the COR improvement contributes on the underwriting side, while the main elasticity comes from the investment side (total investment return 2.6%→3.7%). That means the sustainability of the 11.2% ROE depends heavily on equity markets — the most pro-cyclical profit mix of the three.

• ⚠️⚠️ Why the cut from 83 pts to 72 pts — the "only one not diluted" argument was weakened by this brief's own calculations:
  • Issuing at the 9/4 close of RMB 7.99: ΔBVPS +0.159% (accretive), at 4.245% of pre-issuance share capital.
  • But BVPS is RMB 7.69 against a share price of RMB 7.99, a margin of safety of only 3.9%.
  • More to the point: the assumed issue price PICC itself uses in the plan is RMB 7.41 (the floor of the 20-day average), which is below BVPS — on the company's own assumption PICC is diluted by 0.159%, not accreted.
  • And its dividend per share is diluted −4.07%, between ICBC's (−3.34%) and ABC's (−6.16%); the dividend filing states explicitly that it "intends to keep the total distribution unchanged and adjust the per-share distribution ratio accordingly." The dividend yield goes 2.75%→2.64%, the lowest of the six financial names.
  • The +6.15pp solvency uplift is self-computed (RMB 15 billion / minimum capital of RMB 243.943 billion); PICC's plan provides no static estimate table, so please treat it as a self-computed value.
• Technicals and sentiment: 9/4 −0.62%, turnover value RMB 778 million, share turnover only 0.27% — not priced in ahead at all, which remains its greatest advantage.
• Final judgment: watch closely (no longer "priority deep-dive"). The logic still holds (relatively the least disadvantaged of the three + zero front-running + a 5-year MoF lock-up), but its fulcrum is far thinner than the first draft assumed and cannot be used as a hard conclusion.

52 兆易创新603986SSE main board ±10% | watch closely | 70 pts (newly added this round) · GigaDevice

• Related news: the same overnight semiconductor branch as §5.1.
• Why it is singled out inside an otherwise "watch only" verdict on memory: it is the only one of the 13 names that passes three independent rulers at once.
  • Ruler 1 (was the profit actually earned by operations?): 2026H1 revenue RMB 11.566 billion (+178.7%), net profit attributable to parent RMB 6.857 billion, ex-non-recurring RMB 4.883 billion (+796.9%) — even stripping out RMB 2.228 billion of fair-value changes, ex-non-recurring is still close to +800%. Gross margin 63.13%, with Q2 gross margin at 66.57%, +9.49pp sequentially.
  • Ruler 2 (did the profit turn into cash?): OCF +RMB 6.048 billion, a conversion ratio of 0.88 — against the control group: 江波龙 −RMB 3.151 billion, 佰维 −RMB 6.261 billion, 德明利 −RMB 6.736 billion.
  • Ruler 3 (was the inventory stocked with borrowed money?): interest-bearing debt of only RMB 45 million, cash of RMB 16.965 billion, net cash of RMB 16.920 billion; inventory only 0.11× net assets (德明利 2.19×, 佰维 1.43×, 江波龙 1.39×); liability ratio 11.47%.
• ⚠️ Two counter-points that must be given: ① PE (TTM) is 33.02, but the PE historical percentile has already fallen to 9.3% — sitting in a low-valuation zone that corresponds to peak cyclical earnings, which is itself a warning shape for a cyclical rather than cheapness; ② PB of 6.99 looks undemanding at a 34% percentile, but cannot be taken at face value — its net assets include RMB 11.437 billion of "other equity instrument investments," and the company's 2026-05-26 clarification filing states explicitly that it holds only 1.80% of 长鑫科技 (CXMT) and that fair-value changes on that asset flow through OCI only and do not affect net profit, so the PB is diluted by a single non-operating asset.
• Technicals and sentiment: 9/4 −2.95%, turnover value RMB 11.312 billion (the largest in the memory chain), net main-force outflow RMB 1.915 billion, the largest in the memory chain. It is −56.1% from its one-year high.
• Final judgment: watch closely. It is the only name on this line whose financial statements can support its valuation; but note the directional argument in §2.2 — Micron's expansion is a far-term negative for memory, and GigaDevice's inclusion rests on financial quality, not on industry conditions.

53 工商银行ICBC / 601398SSE main board ±10% | watch closely | 69 pts

• Related news: 2026-09-06 15:31:02 (eiTime). Raising up to RMB 100 billion: Ministry of Finance RMB 70 billion + China Tobacco RMB 10 billion + Shanghai Tobacco RMB 5 billion + Yunnan China Tobacco RMB 5 billion + Hunan China Tobacco RMB 5 billion + Hunan Tobacco RMB 5 billion. Issuance cap 10%, 5-year lock-up. It also filed the same day an "Announcement on Introducing Strategic Investors and Signing a Conditional Strategic Cooperation Agreement" — a document that did not exist in the 2025 round, indicating the tobacco system is positioned as a strategic investor rather than a passive financial subscriber.
• Fundamental verification (2026H1): revenue RMB 465.859 billion (+9.1%), net profit attributable to parent RMB 173.682 billion (+3.3%), BVPS RMB 11.12, weighted ROE 8.63% (−0.19pp), NIM 1.29% (1.30% for full-year 2025), NPL ratio 1.29% (1.31% at year-end), provision coverage 217.58%, up 3.98pp from year-end, CET1 13.21% (0.36pp of net consumption over the half-year).

  Profit quality skews positive: revenue +9.1% against net profit attributable to parent of only +3.3%, with the gap absorbed into provision rebuilding (provision coverage up, NPL ratio down) — this is "spending current revenue to buy a future cushion." On this point ICBC and ABC run in opposite directions.

• Measured dilution: 3.451% of pre-issuance share capital, ΔBVPS −0.897%, EPS / dividend per share −3.34%, dividend yield 3.82%→3.69%; CET1 13.21%→13.54% (+0.33pp, per the plan source text).

  The exchange ratio: 0.9% of book value per share buys back "less than one half-year" of natural capital consumption. And the plan source text flags that "following a G-SIB bucket upgrade the Bank will face stricter additional capital requirements" — if it moves from bucket two to bucket three (additional capital 1.5%→2.0%), this round's +0.33pp will not cover it, implying further capital replenishment down the road.

• Technicals and sentiment: 9/4 +0.37%, share turnover 0.09%, extremely quiet.
• Final judgment: watch closely. The best asset-quality direction and the smallest dilution of the three.

54 敦煌种业600354SSE main board ±10% | watch only | 67 pts · Dunhuang Seed

• Related news: the agriculture theme; CBOT wheat closed 9/2 at 798.75 cents and corn at 538.0, both 3-year closing highs.
• Fundamental verification — it is the best fundamental profile in the agriculture group: 2026H1 revenue RMB 868 million (+20.9%), net profit attributable to parent RMB 145 million (+166.3%), ex-non-recurring RMB 142 million (+188.4%), gross margin 40.67%→48.62%, seeds 89.22% of revenue (at a 51.95% gross margin), liability ratio 37.5%, cash RMB 805 million, almost no long-term borrowings. PE (TTM) 33.2, the lowest among profitable agriculture names.
• ⚠️ But three haircuts are required, and the third is decisive:
  1. Minority interests of RMB 144 million are 50% of net profit of RMB 288 million — profit attributable to parent comes almost entirely from the controlled subsidiary with large minority holders, and the parent system's contribution needs to be unpacked separately.
  2. OCF is only RMB 19.6 million, a net-profit-to-cash conversion of 0.14×; H1 cash received from sales was RMB 387 million against revenue of RMB 868 million (44.6%). The concentration of seed-industry sales in Q1 and Q4 partly explains this, but 0.14× itself needs watching.
  3. ⚠️⚠️ The trading picture is the worst on the board: seal order / turnover of only 0.008 (worst on the board), share turnover 27.93%, only −5.04% from its one-year high (the high of 9.12 was set on 9/1), already +40.4% over 20 days, and it hit limit-down at −9.98% as recently as 9/2.
• A tension that must be acknowledged: A-tier fundamentals + high-risk trading, both true at once. This brief handles it on the principle that "fundamental quality is not a short-term ranker" — it scores high because of the statements, and it is judged "watch only" because of the seal order and the position. Good fundamentals do not get it removed from the risk list.
• The company's own statement: the 9/2 abnormal-fluctuation filing (eiTime 9/1 18:44) states explicitly that "there are no material matters that should have been disclosed but were not."
• Final judgment: watch only.

55 农业银行ABC / 601288SSE main board ±10% | watch closely | 66 pts

• Related news: 2026-09-06 15:31:02 (eiTime). Raising up to RMB 160 billion: Ministry of Finance RMB 130 billion + China Tobacco RMB 10 billion + Jiangsu Tobacco RMB 5 billion + Zhejiang Tobacco RMB 5 billion + Hubei Tobacco RMB 5 billion + Beijing Tobacco RMB 3 billion + Shuangwei Investment RMB 2 billion. Issuance cap 15%.
• Fundamental verification (2026H1): revenue RMB 410.871 billion (+11.07%, the highest of the six), net profit attributable to parent RMB 146.381 billion (+4.93%), BVPS RMB 8.17, weighted ROE 10.14% (−0.61pp), net interest margin 1.28% (−17bp yoy), NPL ratio 1.25%, provision coverage 290.10% (down from 299.61% yoy), loan provision ratio 3.63% (down), CET1 10.80% (−0.62pp yoy, the fastest consumption of the six).

  ⚠️ The highest revenue growth on the board, but entirely volume-for-price — NIM is −17bp year on year, held up by balance-sheet expansion (loans +6.2%, bonds +9.6%), with the cost written directly into capital. And with both provision coverage and the loan provision ratio falling, the cushion is being consumed rather than accumulated — the opposite direction from ICBC.

• Measured dilution: 6.568% of pre-issuance share capital, ΔBVPS −0.913% (almost identical to ICBC's −0.897%), EPS / dividend per share −6.16% (close to twice ICBC's), dividend yield 3.58%→3.36%; CET1 10.80%→11.41% (+0.61pp, per the plan source text), exactly replacing one year of natural consumption.

  ⚠️ Correcting a misreading in the first draft: "ABC's 15% cap = bigger dilution" is wrong. Triggering the 15% cap would require the share price to fall to RMB 3.048 (−56.2%); the cap is simply not a binding constraint, and issuing at market ABC uses only 43.8% of it. It asked for 15% to leave room for price error.

• Technicals and sentiment: 9/4 +1.02%, the strongest of the big four (the filing had not yet been published at Friday's close, so this is more likely the heavyweight rotation at work).
• Final judgment: watch closely, but not in the top three lead picks. It needs the money most (fastest CET1 consumption, largest capital improvement at +0.61pp), but its dividend per share is also under the most pressure (−6.16%).

56 牧原股份002714SZSE main board ±10% | watch closely | 65 pts · Muyuan Foods

• Related news: hog prices rising sequentially for consecutive sessions (9/4 10.58 → 9/5 10.82 → 9/6 10.97 RMB/kg); reports at 15:06 on 9/4 that the Ministry of Agriculture has brought forward the deadline for this round of capacity reduction to the end of Q3; the end-Q2 breeding-sow herd was 37.80 million head, leaving only 0.8% of room above the 37.50 million head holding target.
• Positive logic: a sequential hog-price rebound (+27.4% from the April low) + an earlier capacity-reduction deadline. ⚠️ Note: this brief has falsified the "stockpiling expectation" chain — the state-reserve action closest to 9/4 was the 8/18 rotational release of 12,000 tonnes (supply added), the opposite direction; and 20 days after the two historical level-1 alerts Muyuan was −28.94% and −1.52% respectively.
• Fundamental verification: best-in-industry cost control and the acknowledged core name of the hog-farming sector; but still loss-making currently (PE −226.48), PB 3.21.
• Technicals and sentiment: 9/4 +4.49% on rising volume without a limit-up — which is precisely the advantage: most of the 16 agriculture names that did limit up are loss-making small caps with sloppy seals, whereas a core name rising moderately on volume is a mark of a quality theme.
• Final judgment: watch closely. If agriculture continues on Monday, the core name catching up is more certain than the small caps already on consecutive limit-ups; if the theme fades, its drawdown will also be smaller than theirs.

57 北方华创002371SZSE main board ±10% | watch closely | 64 pts · Naura

• Related news: at the 2026-09-04 U.S. close, the Philadelphia Semiconductor Index was +3.38% (CNBC, last_time=2026-09-04), with KLAC +7.32%, LRCX +5.12%, AMAT +4.31%, ASML +4.17%, while NVDA managed only +0.84% and AVGO +0.21%. The attribution points to Micron's HBM expansion (ETNews, published 9/4 20:00 KST = 9/4 19:00 Beijing time⁷), four hours after the A-share close on 9/4.
• Positive logic (the channel must be stated precisely): Micron expands → global WFE outlook revised up → overseas equipment stocks re-rate → A-share equipment stocks re-rate in sympathy. This is a valuation / sentiment channel, not an order channel.

  ⚠️⚠️ This brief has verified: A-share equipment makers are not on Micron's chain, and are institutionally on the opposing side. Reuters / TrendForce reported in April 2026 that Micron is pushing the MATCH Act, aimed at restricting exports of fabrication equipment to 长鑫 (CXMT), 长江存储 (YMTC) and 中芯国际 (SMIC); 北方华创 is itself on the U.S. Entity List (the 2024 BIS list); and Micron's expansion sites (Taiwan / Singapore / Hiroshima, Japan) all sit in U.S.-aligned export-control jurisdictions. The reading "equipment names will win Micron orders" is wrong — the transmission clock is: sentiment the same day, orders never. The true order-side independent variable for A-share equipment makers is capex at 中芯 (SMIC) / 长存 (YMTC) / 长鑫 (CXMT); and incidentally, 长鑫科技 (CXMT) is already listed on the STAR Market (688825), closing 9/4 at RMB 54.80, −1.35% — that is the demand-side observation point to watch.

• Fundamental verification (the real reason it ranks first): 2026H1 revenue RMB 20.161 billion (+24.9%), net profit attributable to parent RMB 3.370 billion (+5.0%), ex-non-recurring RMB 3.341 billion.
  • Ex-non-recurring / net profit attributable to parent = 99.1%, the cleanest on the board — against the control group, 72% of AMEC's and 69% of Piotech's operating profit comes from fair-value changes and investment income.
  • Operating cash flow +RMB 3.774 billion, a net-profit-to-cash conversion of 1.12×.
  • Inventory −2.0% year on year, the only decline among the 13 names — meaning goods shipped are converting into revenue rather than piling up.
  • Gross margin 40.07%, liability ratio 50.26%. PE (TTM) 81.48, PB 11.42, total market cap RMB 463.1 billion.
• ⚠️ But three counter-points must be given at the same time: ① net profit attributable to parent grew only 5.0% while revenue grew 24.9% — that is negative operating leverage; ② Q2 revenue was −4.7% sequentially, the only decline among the equipment names; ③ gross margin was −1.44pp sequentially. "The cleanest statements" and "the best growth" are not the same thing.
• Technicals and sentiment: SZSE main board ±10% (the 002 prefix originally belonged to the SME board, which merged into the SZSE main board in 2021 — the first draft of this brief mistagged it as ChiNext ±20%, now corrected). 9/4 −2.69% (a smaller fall than the sector's −2.82%), while net main-force inflow was +RMB 101 million, the only positive among the 14 semiconductor names listed in §7.2. It is −34.0% from its one-year high (set 2026-07-01).
• Final judgment: watch closely. The single hard verification point is net main-force flow in the first 30 minutes — an overnight positive of this handoff type becomes a one-day wonder if the net figure turns negative after an initial spike.

58 中微公司AMEC / 688012STAR Market ±20% | watch only | 58 pts (cut from 71 pts in the first draft)

• ⚠️⚠️ The reason for the downgrade is profit quality, which the first draft missed entirely: 2026H1 revenue RMB 6.691 billion (+34.9%), net profit attributable to parent RMB 2.825 billion (+300.2%), but ex-non-recurring is only RMB 1.123 billion — fair-value changes of RMB 992 million + investment income of RMB 1.086 billion = RMB 2.078 billion, 72% of operating profit of RMB 2.872 billion.
• On an ex-non-recurring basis, AMEC's TTM PE is 150.6×, not 75.94×.
• ⚠️ There is also a timing problem: the measurement date for these paper gains is 2026-06-30, and the one-year highs of every semiconductor name covered in this brief fall between 2026-06-26 and 07-14. The interim-report paper gains are marked at the sector's peak day; the current share price is −32.9% from that high, so this fair-value line will most likely reverse in Q3. This portion of "profit" is reflexive — it exists only when semiconductor stocks rise, and has to be given back when they fall.
• Technicals and sentiment: 9/4 −1.64%, the smallest decline among equipment names; but net main-force outflow of RMB 117 million (北方华创 was the only net inflow). STAR Market ±20%, with an RMB 500,000 + 2-year eligibility threshold, so some readers cannot buy it — that is information, not a reason for exclusion.
• Final judgment: watch only. The gap between headline valuation and real earnings power is too wide.

59 楚天龙003040SZSE main board ±10% | watch only | 57 pts · Chutian Dragon

• Related news: no new news of any kind inside the window — this has to be written first.
• Flow facts (hard): limit-up on 9/4, with a seal order of RMB 543 million and a seal order / turnover ratio of 0.522 (second strongest on the board); Dragon-Tiger net buying of RMB 671 million, first across the whole market that day (listing reason: top 5 by daily price deviation of 7%).
• Fundamentals: PE −640.55 (loss-making), PB 7.03. The purported catalyst, the "PBoC digital RMB action plan," was published 2025-12-29 and took effect 2026-01-01 — old news, not a new catalyst inside this window.
• Final judgment: watch only. The flow intensity is top-tier on the board, but the news side is empty and the fundamentals are loss-making. It is listed here because the "flow fact" is worth watching, not because it has any logic. When the seal order disappears, it is over.

510 温氏股份300498盛美上海 (ACM Research Shanghai, 688082) — in brief · Wens Foodstuff

• 温氏股份 (300498, ChiNext ±20%, 56 pts, watch closely): the second-largest hog farmer, currently loss-making (PE −39.70). 9/4 +6.15% without a limit-up, which together with Muyuan is evidence that "the core names followed." On ChiNext's ±20%, +6.15% is still a long way from the limit.
• 盛美上海 (688082, STAR Market ±20%, 52 pts, watch only): PB 8.93 is the lowest among the equipment names, and 9/4 −4.05% was oversold. ⚠️ But behind the headline +42.1% in net profit attributable to parent, ex-non-recurring is −15.3% year on year — fair-value changes of RMB 473 million are 41% of operating profit. Ex-non-recurring has already turned to negative growth, the first equipment name to show this signal.

6. Pass list

---

7. Within-branch rankings

7.1 The central financial enterprise capital-increase branch

Hardest single stock: 中国人保 (601319) — but the word "hardest" needs a discount this round: its non-dilution hangs on a 3.9% margin, and the moment the final issue price falls below BVPS of RMB 7.69 it flips to dilution. Highest trading recognisability: 中国人保 again (the only one of the six not reflected in the price at all yet). Who is just tagging along: 中国人寿 (601628) — its name is on the RMB 360bn list, but the money goes into the unlisted group parent. ⚠️ A counter-intuitive addendum: the ones genuinely unharmed on the dividend dimension are 交行 (4.38%) and 邮储 (4.03%) — undiluted and with the highest yields, having already digested the 2025 round's dilution. They have exactly one problem: they got no money either, so there is no event driver. In a sector priced off dividends (PB and CET1 across the six are almost uncorrelated), this deserves a separate note.

7.2 The semiconductor equipment and memory branch

Hardest single stock: 北方华创 (002371). Best quality but direction in doubt: 兆易创新 (603986). Who is just tagging along: 芯源微 / 精测电子 / 中科飞测 (pure sector beta, with no earnings support). ⚠️ Two things must be said separately about memory: ① at the far end (from 2027) Micron's expansion adds supply and depresses ASPs, which is a negative direction for memory; at the near end (26Q3-Q4), if the expansion comes from converting existing DRAM lines, conventional DDR supply actually tightens, which is neutral to mildly positive. The two clocks must not be mixed. ② Mapping the equipment rally straight onto memory stocks as a positive is taking a supply-side input as demand-side evidence — the direction is inverted. ⚠️ One methodological reminder: this section ranks by fundamental quality, but fundamental quality is not a short-term ranker — same-day moves are led by fund flows. This section answers "can the fundamentals hold it up," not "will it go up."

7.3 The agriculture branch

Hardest single stock (fundamentals): 敦煌种业 (600354) — but the worst trading picture; both are true at once, and this brief does not paper over that tension. Hardest single stock (holdability): 牧原股份 (002714) — core-name status + a position not run ahead + best-in-industry costs. Highest trading recognisability: 亚盛集团 (600108, 2 limit-ups, the leader the market recognises) — but its seal order is only 0.167, and the company itself filed a drought negative; high recognisability does not mean safe, and this brief has reclassified it as Pass. Who is just tagging along: 新希望 and 平潭发展 — seal orders of 0.003 / 0.062, the classic shape of capital using a theme to distribute stock.

---

8. Verification signals for today's open

⚠️ This section consists entirely of signals to be verified. The A-share market had not opened at the time of writing and there is no auction data; everything below sets criteria for the open rather than reporting measurements.

① Auction signals

• The relative strength of the three capital-increase names is the direct referee for this brief's first branch. If 中国人保 (601319) outperforms ICBC / ABC → the market has understood the dilution differences; if all three rise together by similar amounts → the market is indiscriminately trading the words "capital injection," and this brief's stock-selection conclusion is void for the day (the medium term is unchanged).
• A more important control group: watch how 交通银行 (601328) and 邮储银行 (601658) — the two that got no injection — trade. If they rise in step with ICBC / ABC, that is indiscriminate sector speculation with poor durability; if they clearly underperform, the market is pricing "who got the money" rather than "who got diluted."
• 中国人保: ⚠️ its own gap-up size does not constitute a falsification condition for this brief — the higher the share price, the more likely the final issue price lands above BVPS of RMB 7.69, which actually helps the "no dilution" case; a lower price does the reverse. This brief therefore does not use PICC's own move as a criterion, only its strength relative to ICBC / ABC (see the next line).
• The sloppy agriculture limits: the auction in 新希望 (000876, seal 0.003), 敦煌种业 (600354, seal 0.008) and 平潭发展 (000592, seal 0.062) is the touchstone for the agriculture theme — if all three gap down together, the theme most likely fades for the day.

② Sector signals

• Semiconductors: within the first 30 minutes, if the equipment names (北方华创 / 中微 / 拓荆) produce 3 or more up >5% with positive net main-force flow, the overnight SOX transmission is established; if net main-force flow turns negative after an initial spike, downgrade it immediately to a one-day wonder. ⚠️ A single name rising alone does not establish the sector.
• A second, finer criterion for semiconductors: last Friday, of the 14 semiconductor names listed in §7.2 only 北方华创 had positive net main-force inflow (+RMB 101 million). If Monday's inflow is still concentrated in that one name, it is leader siphoning rather than sector repair.
• Agriculture: we need to see the core names 牧原 (002714) / 温氏 (300498) follow on rising volume. If the core names do not follow and only small caps hit the limit, that is a precursor to the theme fading.
• Media: whether 龙版传媒 (605577) breaks its limit is the signal light for the whole line (still sealing after a regulatory warning = extreme sentiment; opening the limit = confirmation that media is fading).

③ Single-stock signals

• 北方华创 (002371): whether it can break above the 9/4 close of RMB 638.17 on volume. Note it is SZSE main board ±10%, not ChiNext.
• 兆易创新 (603986): last Friday's net main-force outflow of RMB 1.915 billion was the largest in the memory chain. Whether that outflow can turn positive is the dividing line between "unjustly sold off" and "being distributed."
• 楚天龙 (003040): whether the RMB 543 million seal order is still there on Monday. A pure flow name — when the seal disappears, it is over.
• 敦煌种业 (600354): the best fundamentals with the worst seal makes it a natural sample for testing whether fundamentals can rescue a sloppy limit.

④ Risk signals

• Gap up then dump: if the three capital-increase names gap up together and turn negative within half an hour, the market has already priced the dilution, and the first branch should simply be abandoned for the day.
• A lone limit-up: if agriculture is left with only 1–2 limit-ups and no core names, the fade is confirmed.
• Core names not following: if only small caps rise in semiconductors while 北方华创 does not, the overnight positive has not been accepted.
• A weak index: if the Shanghai Composite stalls on volume around 3930 (only 1.78% below 4000), a failed heavyweight push will drag on the whole board.
• Yesterday's strong themes fading: performance of 9/4's prior-day limit-up names was already a mean of −0.57%, a median of −1.78%, and only 15 of 44 (34%) up — the wealth effect was already contracting, and that is the shared background pressure on every theme on Monday.
• Macro headwind: the FOMC meets on September 15-16, with CME-implied odds of a 25bp hike around 58%–60%. This is a week of pricing hike expectations, which is a persistent pressure source for high-multiple growth stocks (semiconductors at PB 9–23×).

---

9. Final recommendations

① The 5 stocks most worth watching today

Selection rules (written out openly so they can be checked): ① draw only from names tagged "watch closely" in §3 ("watch only" and "Pass" are excluded without exception, which is why §3's No. 4 敦煌种业 at 67 pts and No. 9 楚天龙 at 57 pts do not appear here); ② each branch may take at most 2 slots, to keep the variables in the list independent of one another and to enforce slot discipline on short-duration branches; ③ within a branch, rank by descending §3 score. On those rules: the capital-increase branch takes 中国人保 (72) and 工商银行 (69); the semiconductor branch takes 兆易创新 (70) and 北方华创 (64); the agriculture branch takes 牧原股份 (65). 农业银行's 66 pts is higher than Muyuan's, but it is excluded because the capital-increase branch has already filled its 2 slots — that is a consequence of the rule, not a negative judgment on it.

⚠️ One line on position structure (the discipline this brief most wants readers to take away): of the 5 names above, Nos. 1 and 5 (semiconductors / memory) belong to a short-duration branch of the "overnight offshore handoff" type — the positive may be fully realised in the opening jump itself. Last Friday's recap had just recorded exactly this lesson (it judged the nature of the precious-metals "front-running window" correctly and still put 4 of its 5 recommendation slots into it). This brief therefore imposes a self-constraint in building the watchlist: a short-duration branch takes at most 2 slots, and net main-force flow in the first 30 minutes is the criterion for keeping or discarding that branch's conclusion. This brief gives no position advice and no entry or exit timing.

One further note: the three lines in this list are three mutually unrelated variables (a policy event / the overnight offshore market / industry prices and flows). They are not three copies of the same variable — even if one is falsified, the other two remain independently valid. That is a deliberate structural choice.

② The 3 strongest branches today

③ Directions not worth chasing today

1. Precious metals — 9/4 non-farm payrolls of +162,000 far exceeded expectations (+55,000 expected), prior readings revised up by 55,000, and CME-implied odds of a 25bp hike on 9/16 rose to about 58%–60%, with spot gold down as much as 2% that day. A-share precious metals were already −1.78% last Friday (2 up, 12 down). Last week's first branch has become today's avoid list — do not give it a second chance.
2. 龙版传媒 (605577) and the publishing / AI-video momentum names — an SSE regulatory warning + AI-video revenue at <0.01% of the total (RMB 80 in June, RMB 75,000 in July) + H1 net profit −34.46% + 5 consecutive limit-ups totalling 61.14%, four negatives stacked.
3. 亚盛集团 (600108) — ⚠️ the company itself filed on 8/14 that drought is "expected to have a certain adverse impact on 2026 operating results"; it is an agricultural producer, so a yield loss is a direct loss of revenue, while domestic corn has been flat for 3 months. The market's reason for buying it runs opposite to the company's own disclosure.
4. 天邦食品 (002124) — in judicial pre-restructuring (already extended six times, to 2026-11-09), with net assets attributable to parent of RMB 823 million against a half-year loss of RMB 1.543 billion, cash of RMB 155 million against short-term borrowings of RMB 3.059 billion, and a 91.2% liability ratio.
5. The feed names (播恩集团 001366, 邦基科技 603151, 正虹科技 000702) — the industry logic runs backwards: capacity destocking = falling feed demand, rising grain prices = rising costs, so both theme narratives are headwinds for them, and gross margins are already sliding.
6. 平潭发展 (000592) — forestry is only 5.05% of revenue and fibreboard 74.86%; it is a man-made-board manufacturer, with a PB of 8.49, the highest in the agriculture group, while loss-making.
7. The levered-inventory-arbitrage memory names (江波龙 301308, 佰维存储 688525, 德明利 001309, 香农芯创 300475) — OCF deeply negative, inventory +219% to +361% year on year, net interest-bearing debt / net assets 73% to 124%; 德明利's Q2 gross margin collapsed 18.58pp sequentially with net profit −20.2% sequentially; 香农芯创's PE percentile of 0.1% paired with a PB percentile of 90.8% is the most extreme cyclical-topping shape.
8. Equipment names whose profits rest on paper gains (中微公司 688012, 拓荆科技 688072) — fair-value changes are 72% / 69% of operating profit, and the ex-non-recurring TTM PEs are in fact 150.6× / 172.7×; and with the paper gains measured on 6/30, exactly the sector's peak this year, Q3 will most likely reverse.
9. Digital RMB / fintech (四方精创 300468, 恒宝股份 002104, 天娱数科 002354) — no new catalyst of any kind inside the window, and the so-called "action plan" is old news from 2025-12-29; share turnover of 27%–38% is already extremely overheated.
10. 宇邦新材 (301266) — the offer price of RMB 20.08 is a 19.58% discount to the pre-suspension close of RMB 24.97, so the "tender offer" provides no price support here.
11. 中国人寿 (601628) — the RMB 35 billion injection goes into the unlisted group parent; the listed company got none of the money.

④ The final one-line judgment

Today is a three-track day of "an event for the heavyweights, an overnight move for tech, and inertia for the themes," directionally positive but structurally tearing; and there are exactly two real points of contention, both of which will be settled within the first half hour — first, "is the RMB 360 billion injection read as an injection or as dilution" (the former makes all six large banks rise together, the latter makes 中国人保 601319 uniquely strong, with BoCom / PSBC, the two that got no injection, serving as the best control group), and second, "can the overnight SOX +3.38% be converted into positive net main-force flow in A-shares" (last Friday only 北方华创 002371 had a net inflow among the 14 semiconductor names listed in §7.2). These two answers decide whether this brief's first two branches succeed or fail; everything else is noise.

---

Six corrections the sub-agent measurements made to the first draft (merged into the body; kept here for the record)

Open items after finalisation (must-check list for the next post-close recap)

1. ⚠️ Top priority: the primary source for "the August breeding-sow herd fell 2.94% month on month." The sub-agent could not verify any official origin; all that can be found is the Statistics Bureau's end-Q2 figure of 37.80 million head and July sample-agency year-on-year readings of −3.54% / −5.64% / −6.64%. If that number is in fact a year-on-year or sample-basis figure being cited as an official sequential number, then the "destocking is accelerating" leg of the agriculture branch is hollow. This brief treats it as "to be verified · not used as an argument"; the post-close recap must close it out.
2. The pass-through progress of domestic corn / wheat spot prices. CBOT wheat and corn set 3-year highs on 9/2, but DCE corn closed 9/4 at RMB 2,303/tonne, flat over the past 3 months and −20.5% from its 3-year high, and CBOT wheat was −2.52% on 9/4 itself. For international grain prices to become revenue at 敦煌种业 / 亚盛, they must first show up in domestic prices.
3. An exclusive attribution for the 9/4 SOX +3.38% — this brief explicitly cannot attribute it solely to the Micron news (the equipment names outran Micron itself: KLAC +7.32% > MU +6.10%) and tags it "to be verified · attribution undetermined." Competing explanations include the WFE capex narrative ("falling yields" has been removed: U.S. 10Y actually rose +2.2bp on 9/4).
4. A same-day attribution for A-share semis −2.82% and RMB 16.038 billion of net outflow on 9/4 — not obtained for a second consecutive piece; now a carried-forward item.
5. Micron's official statement on the HBM expansion — only the Korean outlet's citation of anonymous supply-chain sources exists; the static link to the FQ3 prepared remarks on investors.micron.com returns a 301 redirect and the source text could not be read. Micron's FQ4 earnings call (expected in late September) is the only upgrade point.
6. Position detail behind the fair-value changes at AMEC / Piotech / ACM Shanghai — only line-item amounts were obtained; the interim-report notes were not opened, so "Q3 will reverse" is currently an inference, not a fact.
7. Inventory write-down provisions in the interim-report notes of the three module makers (江波龙 / 佰维 / 德明利) — RMB 65.3 billion of inventory and RMB 38.9 billion of net interest-bearing debt in aggregate make this the most risk-concentrated line item in the memory chain.
8. The sponsors / lead underwriters for the three capital increases — not obtainable from the body of the issuance plans, which is why "brokers benefit" was not given as a branch in this brief.
9. The record dates for ICBC's and ABC's 2026 interim dividends — these determine whether the −3.34% / −6.16% dividend-per-share dilution happens this year or is pushed into mid-2027. The wording of the three dividend filings differs (PICC states explicitly that it will "keep the total distribution unchanged").
10. The two core falsifiable points in this brief must be reconciled name by name after the close; sector-level moves are not an acceptable substitute:
    • ① "中国人保 outperforms ICBC / ABC" — and at the same time check whether BoCom / PSBC (which got no injection) rise in step.
    • ② "北方华创's net main-force flow turns positive and at least 3 equipment names are simultaneously >5%" — if the inflow is still concentrated in that one name, it is leader siphoning, not sector repair.

---

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