A-Share Pre-Market Brief | 2026-07-29 Wednesday

⚠️ Risk warning (up front): This report is a pre-market information review and research record. It does not constitute investment advice and contains no ratings, target prices, position sizes or buy/sell recommendations. The "priority deep-dive / watch closely / watch only / Pass" labels in Section 3 are research-priority labels, not investment ratings, and carry no buy or sell meaning. A-share volatility risk is extremely high, and automatically generated content may contain timeliness gaps or industry-chain mapping errors; it must not be used directly as a basis for trading.

Time window: 2026-07-28 15:00 close → 2026-07-29 09:00 (when this report went to press).

🔴 Major correction (found by risk-auditor QC, fixed before publication)

The first draft of this report took the overnight overseas data from the 2026-07-27 US close throughout, not the 7/28 close. The root cause is that Chinese financial media label US closes by "Beijing date" (an article headlined "close on July 28" actually reports the 7/27 US session). The "WTI −7.5%, Brent −8.7%, SOX −2.23%, Nvidia −4.99%, ASML −5.8%" the first draft wrote on that basis are all 7/27 figures. Worse, in the basis disclosure the first draft spotted a US$4 conflict between Sina's "Brent about US$84" and "88.36", yet closed the discrepancy with an unverified "contract versus spot basis difference" — when those US$4 were exactly one day's price change. Everything has been replaced after re-checking against investing.com historical data and MarketScreener settlement prices.

The correct 2026-07-28 US / commodity closes (the basis this report now uses):

Metric	07-24	07-27	07-28 (the true overnight values)
Philadelphia Semiconductor Index	11,818.9 (−4.25%)	11,554.9 (−2.23%)	11,035.7 (−4.49%)
WTI September contract	89.31 (−3.1%)	82.61 (−7.5%)	79.26 (−4.1%)
Brent September contract	—	88.36	83.90 (−5.1%)

Table scrolls horizontally

This correction directly overturns the first draft's headline argument. The first draft said "the −7.5% single-day oil move happened after the A-share close, so A-shares have not priced it" — in fact that −7.5% settled in the early hours of 7/28 Beijing time, visible to A-shares all through the 7/28 session, and they did not react. That is not an opportunity of "the market has not seen it" but evidence that "the market is unwilling to price it". The genuinely new overnight decline is only WTI −4.1% / Brent −5.1%, roughly half of what the first draft claimed. The airline branch is therefore cut from 1st to 3rd, and Spring Airlines is downgraded from "priority deep-dive" to "watch closely".

At the same time, the SOX's true decline of −4.49% is double the −2.23% the first draft wrote, and it was the 3rd consecutive down day (−4.25% → −2.23% → −4.49%). Avoidance of the AI hardware chain is strengthened accordingly.

Data basis disclosed up front (iron rule: if it cannot be found, write "no reliable data"):

1. Stock quotes, PE/PB, volume ratios and turnover ratios are as at the 2026-07-28 close (Tencent quotes qt.gtimg.cn); the limit-up / limit-down pools are Eastmoney push2ex (date=20260728, 61 limit-ups, 49 limit-downs; the source of the discrepancy with the media's "68" is unverified, see item 3); the top-trader list is Eastmoney datacenter-web (disclosed 2026-07-28); financial data were independently pulled with akshare by the fundamentals-analyst sub-agent on 2026-07-29 and cross-verified against the original filing PDFs.
2. FTSE China A50 futures night-session close: no reliable data. The Sina quotes interface returns 403, and the secondary reports found contradict each other ("closed −0.01%" and "down more than 2%" coexist and cannot be matched to a trading day), so this report does not cite it.
3. There is a discrepancy between 61 limit-ups (Eastmoney push2ex) and 68 (media), and the source of the discrepancy is unverified. The first draft explained it as "whether the BSE / ChiNext / STAR Market are included", but this report's own limit-up pool contains ChiNext names such as 波长光电 (Bochang Optoelectronics, 301421) and 科蓝软件 (Kelan Software, 300663), so that explanation has been falsified by the report's own data and is withdrawn. This report uses 61 as its internal basis throughout and makes no cross-source comparison of counts.
4. All top-trader figures are on the "top five disclosed seats" basis and do not equal whole-market flows. Taking Innolight as an example, buys and sells on the list total RMB 18.365 billion while the day's total turnover was RMB 51.6 billion — the disclosed seats cover only about 35% of turnover, and the direction of the other 65% is unknown.
5. Every PE in the tables is PE(TTM) (the Tencent quotes qt.gtimg.cn basis, independently recomputed and checked by the sub-agent with akshare), not forward PE. The two can have opposite signs for companies with volatile earnings (e.g. 南天信息 (Nantian Information) at TTM +127.95 versus forward −57.57).
6. The July Politburo meeting: as of press time, no authoritative report of a 2026 July meeting having convened was found (the "convened on July 28" text that search returned was verified to be a 2022 article containing "the epidemic must be contained" and has been discarded). By convention it should be held at the end of July and could land at any time.
7. This report has removed 3 items of 2025 old news wrongly recalled by search engines: the childcare subsidy implementation plan, Shanghai's RMB 600 million computing-power vouchers, and the Guangdong paper industry's "anti-involution" initiative. All three are still circulating in several pre-market round-ups today; please do not treat them as new news.
8. Dimensions not covered (honest disclosure): liquidity and funding conditions (month-end effects, PBoC open-market operations) are entirely uncovered by this report, and they happen to be the most direct supporting or falsifying variable for the "banks / baijiu defensive" branch; this report has no current data whatsoever on "AI hardware capex slowdown" (fab capex guidance, PCB utilisation, Avary Holding's construction in progress), so all related discussion in the text is at the narrative level.

---

0. Today in One Sentence

Today's strongest new development is negative: SK Hynix's 2Q26 results landed before the A-share open and missed on both lines — revenue of KRW 79.3187 trillion (vs consensus KRW 84 trillion, 5.5% below) and operating profit of KRW 60.5426 trillion (vs consensus KRW 64.1–64.68 trillion, about 5.4%–6.4% below), but the operating margin of 76.3% is an all-time high, with DRAM ASP up about 30% QoQ and NAND ASP up mid-50% QoQ. Stacked on top is the overnight Philadelphia Semiconductor Index at −4.49% to 11,035.7, the 3rd consecutive down day (−4.25%→−2.23%→−4.49%), with Micron −nearly 9%, AMD and ARM −more than 8%, SanDisk −more than 14% (down more than 51% cumulatively since the start of July) and SK Hynix −nearly 9% — this is no longer "overnight sentiment" but data handed over by the industry participants themselves: prices are still rising, yet revenue and profit missed. The whole memory / AI hardware chain is treated as "avoid" today, and the unabsorbed limit-down bid queues from 7/28 (GigaDevice RMB 2.630 billion, Dongshan Precision RMB 1.346 billion) will most likely keep weighing on it. On the positive side, the only company-level positive cross-verified by the sub-agent is 圣晖集成 (Shenghui Systems Integration, 603163)'s RMB 858 million contract win (equal to 28.7% of its 2025 revenue, with the customer a wholly-owned subsidiary of Avary Holding, and this the second such order this year), but the company itself states "no formal contract has yet been signed", so it is not given the highest priority. The first draft was wrong to make oil the headline catalyst — the day WTI fell −7.5% settled in the early hours of 7/28 Beijing time, visible to A-shares throughout 7/28, and they did not react (the airline sector ranged from −0.27% to +1.48%), which is low willingness to price, not unpriced; the genuinely new overnight move is only WTI −4.1% / Brent −5.1%. And for the hottest theme (domestic DUV lithography), the stock-level mapping falls apart entirely on inspection: all 6 names that hit limit-up on 7/28 are low-turnover concept shells, while the companies that genuinely have lithography businesses — Chipmore, Focuslight, Fortune Precision and Tenjing Sci-Tech — all fell 7~11% that day; and 奥普光电 (Aopu Optoelectronics, 002338) twice stated explicitly in a 2026-05-19 exchange disclosure document that it "has no business dealings with Shanghai Micro Electronics Equipment for the time being". Yesterday's second-hottest line, software / IT localisation, was also falsified across the board after the sub-agent's checks — 4 of the 5 main-board limit-up names posted single-quarter 2026Q1 losses, and 天融信 (Topsec, 002212) sealed limit-up on 7/28 even though it had already filed on 7/15 that its H1 loss widened 309.6%, while Kingsoft Office, which genuinely had earnings that day, rose only 0.18%. Capital direction will most likely follow the path already verified on 7/28: three defensive additions in banks (large state-owned banks with large-order net inflows of RMB 1.201 billion and volume ratios of 1.44~1.51), baijiu and telecom operators. Driver types: overseas industry (negative, and upgraded from "sentiment" to "earnings") + company filings + overseas commodities, with nothing on the policy side. Pre-market state: structural, low-elasticity, defensive, and the negative increment is stronger than the positive. Today is the last full trading day before the FOMC decision (02:00 Beijing time on 7/30); CME FedWatch probabilities have been deleted from this report because there is no verifiable time stamp for the reading (the readings found were scattered across a "hold 63.5%~68.5%" range and drifted quickly within a week, so they do not meet citation standards).

---

1. News Overview

Impact level: S = changes an industry trend or delivers direct order elasticity; A = clearly positive for a branch with multiple beneficiaries; B = positive but with a long chain that needs verification; C = sentiment stimulus with weak persistence.

---

2. Strongest Positive Branches, in Descending Order

Branch notes (why positive / short-term or medium-term / any earnings-delivery path / already fully expected / one-day-wonder risk / any main-board name available)

• Branch 3 (oil) — this item has been overturned and rewritten by risk-auditor; please note the difference from the first draft: why positive — jet fuel is airlines' largest cost item (this report did not go back through each company's annual report for the exact share of operating costs, so there is no reliable data; "low-cost carriers have a higher share" is likewise an experiential judgement, tagged as to-be-verified), and falling costs directly improve the next quarter's margin. Already fully expected — this is the key correction here: the first draft's claim that "the largest single-day decline happened after the A-share close and is therefore unpriced" is wrong. WTI −7.5% settled in the early hours of 07-28 Beijing time and was visible to A-shares all through 7/28, and the airline sector's reaction was China Eastern −0.27%, Air China −0.16%, China Southern −0.19%, Spring +0.55%, Juneyao +1.48% — near-zero reaction means the market saw it and was unwilling to price it, which is a negative signal rather than an opportunity. The genuinely new overnight move is only WTI −4.1% / Brent −5.1%, roughly half the magnitude the first draft claimed. The "expectation gap" dimension is cut from "medium-high" to "low–medium" for every name in this branch. One-day-wonder risk — medium-high. Main-board name available — yes.
• Branch 1 (Shenghui Systems Integration): why positive — RMB 858 million equals 28.7% of 2025 revenue (RMB 2.9885 billion) and 29.3% of the order backlog at end-2026Q2 (RMB 2.930 billion), the largest single order by size in the whole market today. The inference that "Avary hit limit-down yet is expanding capacity" has been downgraded per risk-auditor's view: the customer Qingding Precision Electronics is a wholly-owned subsidiary of 鹏鼎控股 (Avary Holding, 002938), and Avary hit limit-down on 7/28 — this is a single counter-directional data point, not enough to refute "AI hardware capex is slowing". Three limits: ① plant fit-out works reflect capacity decisions made several quarters ago, a lagging indicator unsuitable as forward-looking counter-evidence; ② whether Qingding (Huai'an) makes consumer-electronics boards or AI server boards was not verified by this report; ③ this report has no current capex data on Avary at all (construction in progress, 2026 capex guidance, H1 pre-announcement). Cross-verification is needed before it holds; currently tagged "to be verified". ⚠️ But one double standard in this report's own basis must be pointed out (raised by risk-auditor and accepted): for China State Construction, this report rigorously spread RMB 22.4 billion over a 5-year term to get "0.215% of annual revenue on average"; for Shenghui, RMB 858 million is divided straight into a single year's revenue to get 28.7%, with no spreading over the term. Plant fit-out and fixed-asset works usually span multiple accounting periods, so measured with the same ruler the annualised share would be materially below 28.7%. Therefore 28.7% should be read as "an upper bound on the whole-project revenue share, not an annualised figure"; the company has not disclosed the term, so it is tagged "to be verified". Separately, at a 10.25% gross margin, RMB 858 million implies whole-project gross profit of about RMB 88 million against 2025 net profit attributable to parent of RMB 154.5 million — spread over several years, the profit elasticity is far smaller than the revenue elasticity suggests. Earnings-delivery path — clear but with three thresholds (no formal contract signed, unknown term, whether the gross margin can hold 10%). Already fully expected — no; the stock was −7.36% on 7/28. Main-board name available — yes.
• Branch 3 (buybacks and stake increases): why positive — more than ten filings on the same night is a self-rescue consensus after an extreme fall. Short term. Earnings-delivery path — none, purely capital and signalling. Already fully expected — no. One-day-wonder risk — high. Main-board name available — yes. Key reminder: Innolight and Eoptolink are on ChiNext and are not primary recommendations here, but they are the real focus of this branch.
• Branch 4 (defensive): short term (1–2 days), the highest certainty and the lowest elasticity. Main-board name available — yes.
• Branch 5 (lithography): why positive — domestic equipment going "0 → 1", changing semiconductor self-sufficiency over the medium and long term. Medium-term logic, short-term speculation. Earnings-delivery path — essentially none: 5 units a year has no visible impact on any A-share listed company's revenue line; the sub-agent's conclusion is that none of the three main-board names has ever disclosed an order, customer or revenue from domestic DUV (see Section 7). One-day-wonder risk — high: all 6 are first boards, 0 consecutive, and the bid queue is concentrated in Zhangjiang Hi-Tech alone. Main-board name available — yes, but see the mapping-quality tiers.
• Branch 6 (low-valuation defensive): China State Construction's value is not in the RMB 22.4 billion order (the sub-agent's estimate: spread over the 5-year construction period, the annual average is only 0.215% of revenue, and at the 2.81% historical overseas gross margin the annual gross-profit contribution is about RMB 130 million, a rounding error against RMB 39 billion of net profit), but in the three numbers PE 5.05 / PB 0.40 / dividend yield 5.86%.
• Branch 7 (software / IT localisation): the weakest structure, and fundamentally falsified across the board after the sub-agent's checks. 0 consecutive boards means no mainline has formed. The strongest circumstantial evidence is that all 5 main-board software stocks that hit limit-up on 7/28 have no earnings, while 金山办公 (Kingsoft Office, 688111), which genuinely had earnings that day, rose only 0.18% — money bought the ones with no earnings and went around the one with earnings. This is not industry-driven; it is money looking for somewhere to hide. This report downgrades the whole branch to "not recommended", retaining it for observation only.
• Branch 8 (AI hardware): explicitly avoid. Reasons in Section 9 ③.

---

3. Overall Stock-Level Catalyst-Strength Ranking (main board first, in descending order)

The labels are "research priority", not investment ratings: priority deep-dive / watch closely / watch only / Pass, carrying no buy or sell meaning. Catalyst level: S / A+ / A / B+ / B / C.

⚠️ On the scores (risk-auditor's view, accepted): the first draft's two-digit totals cannot be rebuilt dimension by dimension from the Section 4 model, making them false precision reverse-engineered from the ordering; and the "valuation / dividend" factor that actually influenced the ranking (China State Construction's 58 pts above Agricultural Bank's 56 pts) has no corresponding dimension in the model at all. This report therefore does two things: ① totals become banded (high 60+ / medium-high 50–59 / medium 40–49 / low <40), dropping single-digit precision; ② where the evidence grades of different names are not comparable, this is stated explicitly — for example rank 1 (a company filing cross-verified from three sources) and rank 2 (a macro read-across with no financial verification) do not form a comparable ordering.

⚠️ There is no "priority deep-dive" name today. The first draft listed Shenghui Systems Integration and Spring Airlines as "priority deep-dive"; after QC both were cut to "watch closely": the former's own filing says "no formal contract has yet been signed", and the latter's core argument (the expectation gap) has been falsified by the timeline. Allowing "no top-priority name today" is a rule this report set for itself in the 7/28 recap, and today is its first application. Everything on this list is Shanghai/Shenzhen main board. ChiNext / STAR Market / BSE names never enter this list; see Section 3.1 "Observation zone". Important premise: this report holds that no name today combines all four of "an S-grade catalyst + main board + high elasticity + solid fundamentals". The top two each have one clear shortfall (Shenghui: no signed contract; Spring: no company-level catalyst).

3.1 Observation zone (not main board, no primary recommendation)

---

4. Stock Scoring Model (100 points total)

Three hard constraints added today (the first two come from the lessons of the 2026-07-28 recap, the third is new today): ① An overnight SOX decline >2% → the corresponding A-share branch ordering is cut one level directly. Today the SOX was −4.49% (a 3rd straight down day), and stacked with SK Hynix's miss, the whole AI hardware chain is cut to "avoid", at double the intensity of the first draft (which wrongly worked from −2.23%). ①-b (added today, arising from today's error): when citing overnight overseas closing data, the "trading day the data correspond to" must be recorded and displayed, and inferring it from the report generation date is prohibited; when two sources conflict, check the trading day first, and closing the discrepancy with an unverified "basis difference" is prohibited. ② Anything more than 30% off its 120-day high without a volume-backed stabilisation candle is capped at 3 pts on the "technical sentiment" dimension; "the deepest oversold" may not be used as a positive rationale. ③ Every earnings positive must have its "increment source" broken down: if the main source of the net profit increment is a non-operating item such as finance costs, investment income or fair-value changes, the "earnings elasticity" dimension is capped at 6 pts. 环旭电子 (601231) and 奥普光电 (002338) were cut on this basis today.

---

5. Detailed Analysis of the Top 10

1圣晖集成603163watch closely (pending formal contract signature) · Shenghui Systems Integration

• Related news: on the evening of 2026-07-28 the company filed that it had received notice from Qingding Precision Electronics (Huai'an) Co., Ltd. of an intention to commission the "HD01HD02 plant phase-one interior fit-out primary and secondary distribution works & fixed-asset project", with total engineering and fixed-asset value expected at RMB 858 million (excluding tax). The company also states explicitly: no formal contract has yet been signed with the customer, and the signing and terms remain uncertain (Jiemian⁷).
• Positive logic: a direct order positive, and the highest-elasticity one in the whole market today. The sub-agent's calculation: RMB 858 million ÷ 2025 revenue of RMB 2.9885 billion = 28.7%; ÷ the end-2026Q2 backlog of RMB 2.930 billion = 29.3%. If signed, the backlog could jump from RMB 2.930 billion to about RMB 3.79 billion.
• Industry branch stage: pre-launch. The stock was −7.36% on 7/28, sold off along with the semiconductor / PCB engineering chain, and has not yet reflected the order.
• Fundamental verification (sub-agent results):
  • Core business confirmed: by product in 2025 — cleanroom engineering RMB 1.954 billion (65.39%), other electromechanical installation RMB 967 million (32.37%); by industry — IC semiconductors RMB 1.784 billion (59.68%), precision manufacturing RMB 1.013 billion (33.91%); by region — domestic 57.96%, overseas 41.95% (Vietnam 21.9%, Thailand 15.5%), making it a direct beneficiary of electronics manufacturing capacity moving offshore.
  • Financials: 2025 revenue RMB 2.989 billion (+48.85%), net profit attributable to parent RMB 154.5 million (+35.09%), ROE 13.66%; 2026Q1 net profit attributable to parent RMB 38.7 million (+33.66%). The balance sheet is extremely clean: cash of RMB 1.086 billion with essentially zero interest-bearing debt, net cash of about RMB 1.09 billion equal to 15.4% of market cap. Receivables of RMB 400 million + contract assets of RMB 462 million = RMB 862 million, only 28.8% of revenue (versus China State Construction's 48.2%). 2026Q1 operating cash flow of +RMB 349 million is the best single quarter in its history.
  • The customer is verifiable: Qingding Precision is a wholly-owned subsidiary of 鹏鼎控股 (Avary Holding, 002938); in May 2026 Shenghui had already won an RMB 336 million project from the Avary group — this is the second order of the year, so it is a repeat customer, not a one-off.
• Industry position: the honest conclusion is that it is not the niche leader but second tier. Comparable companies' 2025 annual reports: 亚翔集成 (Yaxiang Integration, 603929) revenue RMB 4.907 billion, net profit attributable to parent RMB 892 million, gross margin 24.57%, ROE 43.55% (⚠️ no DuPont decomposition was done; a 40%+ ROE at an engineering company depends heavily on a low net-asset base and high leverage driven by advances and contract liabilities, and may also include one-off projects, so it is not used as a conclusion about operating quality); 柏诚股份 (Bocheng, 601133) revenue RMB 4.133 billion, net profit attributable to parent RMB 205 million; 太极实业 (Taiji Industry, 600667) revenue RMB 30.682 billion. Shenghui has the smallest revenue scale and its 10.25% gross margin is far below Yaxiang's; its advantages are profitability efficiency (ROE 13.66% versus Bocheng's 6.97% and Taiji's 5.22%), a clean balance sheet and Southeast Asian exposure.
• Technical sentiment: main board; 7/28 −7.36%, volume ratio 0.75, turnover ratio 2.74%, falling on shrinking volume, with no high-level risk. Its market cap is only RMB 7.065 billion, giving the highest order-to-market-cap ratio on the list.
• Final judgement: this is the only positive today that is company-level, quantifiable, high-elasticity and cross-verified by a repeat customer. But four deductions must be accepted at the same time: ① a winning notice ≠ a formal contract, and the company wrote the uncertainty itself; ② gross margin fell from 13.44% to 10.25% over three years, so the +48.85% revenue growth in 2025 was bought with concessions, and 28.7% of revenue elasticity need not translate into proportional profit elasticity; ③ RMB 550 million of convertible bonds is pending, with theoretical maximum dilution of about 7.8% (the actual figure depends on the conversion price — no reliable data); ④ downstream IC semiconductors are 59.68%, the same source as the semiconductor capex now being de-rated — if capex really slows, this order flow will dry up later.
• One piece of counter-information worth stating separately: Avary Holding hit limit-down on 7/28 while its wholly-owned subsidiary was expanding capacity that evening. The market is pricing Avary on "AI hardware capex slowing", while the industrial behaviour is the opposite. This is both support for Shenghui and a challenge to the whole bear case on the PCB chain — but note that a single plant order is not enough to overturn an industry judgement; it is only a counter-example to track.

2春秋航空601021watch closely · Spring Airlines

• Related news (corrected): in the 2026-07-28 US session the WTI September contract settled at US$79.26 (−4.1%) and the Brent September contract at US$83.90 (−5.1%), about −16% over three days, the largest three-day decline since April 2020, triggered by Middle East de-escalation with both the US and Iran avoiding strikes (MarketScreener⁵). ⚠️ The first draft wrongly used the 07-27 settlements here (82.61 / 88.36) and drew from them the wrong conclusion that "A-shares have not priced it" — WTI −7.5% settled in the early hours of 07-28 Beijing time and was visible to A-shares all through 7/28.
• Positive logic: direct, and it hits the income statement rather than just sentiment. Jet fuel is airlines' largest cost item, and falling oil directly widens the margin per seat. Fuel is a higher share of costs at low-cost carriers than at full-service carriers, making the elasticity purer. Note: this report did not go back through each company's annual report for jet fuel's share of costs, so there is no reliable data on the exact figure.
• Industry branch stage: pre-launch — the A-share airline sector barely reacted on 7/28 (Spring +0.55%), with no limit-up ladder and no core leader, i.e. "the news has happened and the price has not reacted".
• Fundamental verification: PE 17.20, PB 2.42, market cap RMB 45.121 billion, one of the lowest valuations among the six listed carriers. This report's 3 sets of fundamentals-analyst verification did not cover the airline sector, so Spring Airlines' revenue / net profit / margin / cash flow / gearing trends are "not independently verified by the sub-agent" — this is the evidentiary disadvantage of this call relative to ①, and must be said plainly.
• Industry position: China's low-cost airline leader.
• Technical sentiment: main board; 7/28 volume ratio 0.95, turnover ratio 0.69%, no unusual moves, no limit-up, no speculative footprints.
• Final judgement: the first draft made it the representative of "new information + unpriced"; that argument has been overturned by the timeline — A-shares could see WTI −7.5% all through 7/28 and barely reacted. It is still a macro read-across straight to the cost side, but current data on jet fuel's share of costs are missing, and this report makes no prediction whatsoever about its price performance or return.

3中国国航601111watch closely · Air China

• Related news: same as ②.
• Positive logic: the highest share of international routes and the largest absolute fuel consumption, so the profit elasticity to falling oil is the greatest among the six listed carriers.
• Fundamental verification: PE 62.79, PB 1.95, market cap RMB 124.8 billion. A high PE means current earnings are at a low level — which is both the source of the elasticity and the source of the risk: when the profit base is low the percentage amplification from cost improvement is strongest, but losses are amplified equally when oil rebounds.
• Technical sentiment: 7/28 −0.16%, volume ratio 0.93, turnover ratio 0.67%.
• Final judgement: the greatest elasticity and the worst fundamentals. Within the same branch, Air China corresponds to elasticity and Spring to quality, with different risk characteristics — this report only distinguishes their character and makes no portfolio or selection recommendation.

4吉祥航空603885watch closely · Juneyao Airlines

• Related news: same as ②. Industry branch stage: it has already moved first — it rose 1.48% on 7/28 with a volume ratio of 1.07, the only clearly green name in the entire airline sector, showing some capital reacted early.
• Fundamental verification: PE 21.00, PB 2.40, market cap RMB 23.837 billion.
• Final judgement: its character sits between Spring and Air China, and it was already +1.48% on 7/28, the earliest reaction in this branch — meaning it has the smallest "unreacted" component.

5中国建筑601668watch closely (low-valuation defensive, not order-driven) · China State Construction Engineering

• Related news: on the evening of 2026-07-28 it filed a contract signed with Kuwait's Ministry of Public Works for the North Kabd wastewater treatment plant and supporting projects, worth KWD 999.85 million ≈ RMB 22.4 billion, comprising 5 years of design and construction + 5 years of maintenance and operation, about 1.1% of audited 2025 revenue.

• Positive logic and elasticity calculation (sub-agent results; this is the most important correction in this item):

  Basis	Share
  RMB 22.4 billion / full-year 2025 revenue of RMB 2,082.1 billion	1.08%
  Spread over the 5-year construction period, about RMB 4.5 billion a year / annual revenue	0.215%
  RMB 22.4 billion / full-year 2025 overseas revenue of RMB 141.7 billion	15.8%
  RMB 22.4 billion / total new contracts in 2026H1 of RMB 2,464.7 billion	0.91%

  Table scrolls horizontally

  At the 2.81% historical overseas gross margin, the average annual gross-profit contribution is about RMB 130 million — a rounding error against RMB 39.07 billion of net profit attributable to parent. This filing is "true in fact, zero in elasticity".

• Fundamental verification (sub-agent results): revenue RMB 2,265.5 → 2,187.3 billion (−3.45%) → 2,082.1 billion (−4.81%), with a further 7.85% decline in 2026Q1; net profit attributable to parent RMB 54.26 → 46.19 → 39.07 billion, two straight years of double-digit decline; ROE has slid from 13.36% to 8.15%; OCF / net profit attributable to parent is only 0.53x; receivables + contract assets swelled from RMB 592.7 billion in 2023 to RMB 1,004.5 billion in 2025 (48.2% of revenue), rising further to RMB 1,064.8 billion in 2026Q1; gearing 76.90%. Overseas revenue is 6.81% of the total but at a gross margin of only 2.81% (versus 10.41% domestically) — "scale without profit".

• Where the real case lies: PE 5.05, PB 0.40, dividend yield 5.86% (2025's RMB 2.718 per 10 shares ÷ RMB 4.64). New overseas contracts in 2026H1 were RMB 165.5 billion, +44.2% YoY, and that trend is real.

• Technical sentiment: 7/28 +0.22%, volume ratio 0.91, turnover ratio 0.66%, no unusual moves.

• Final judgement: what is worth discussing is the 5.86% dividend yield and PB 0.40, not the RMB 22.4 billion order — the order elasticity verifies at just 0.215%/year. The biggest hidden risk is the RMB 1.06 trillion of receivables + contract assets, a direct read-across of collection pressure from local finances and the property chain. If you want elasticity, skip it.

6农业银行601288⑦ 工商银行 (ICBC, 601398) · watch closely · Agricultural Bank of China

• Related news: no stock-specific news. Measured flows on 2026-07-28: the large state-owned banks III sector was +2.44% with large-order net inflows of RMB 1.201 billion, the largest of any industry sector in the market. ABC +3.57% with a volume ratio of 1.47; ICBC +3.10% with a volume ratio of 1.44; CCB +2.70% with a volume ratio of 1.51 — against a whole-market volume ratio of just 0.85, all three had volume ratios above 1.4, which is genuine incremental money rather than a low-volume mark-up.
• Positive logic: capital rotation. The natural choice on the eve of the FOMC: low valuation, high dividend yield, and decoupled from overseas tech pricing.
• Fundamental verification: ABC PE 8.28 / PB 0.88; ICBC PE 7.65 / PB 0.73.
• Final judgement: the least volatility and also the least elasticity. One widely overlooked two-way risk: if the FOMC clearly does not hike and risk appetite recovers, the defensive money that flowed in on 7/28 could flow out just as fast. It must also be said honestly that this report does not cover funding conditions at all (month-end effects, PBoC open-market operations), and those are the most direct supporting or falsifying variable for this defensive logic.

8京东方A000725watch closely · BOE Technology

• Related news: on the evening of 2026-07-28 the controlling shareholder Beijing Electronics Holding proposed to increase its stake by RMB 500 million – 1 billion via centralised bidding within 6 months, with no price range. The stock was −6.53% that day on turnover of RMB 9.805 billion and a 4.86% turnover ratio (JRJ¹⁷).
• The two sides of the signal's strength (sub-agent results): as at the filing date Beijing Electronics Holding directly held 274 million shares, only 0.74% of total share capital. RMB 500 million – 1 billion at RMB 5.58 is about 90–179 million shares — equivalent to more than doubling its own direct holding, a heavy move for it; but against a total market cap of RMB 206.7 billion it is only 0.24%–0.48%, and spread over 6 months that is roughly RMB 4–8 million a day, negligible for secondary-market absorption. It is also the first stake-increase plan in the past 12 months.
• Fundamental verification (sub-agent results): revenue growth of +13.66%→+3.13%→2026Q1 +0.80% and net profit growth of +109%→+10.03%→+5.78%, so growth has stalled; 2026Q1 gross margin 15.60%, down 0.17pct YoY; capex over three years RMB 24.8 → 34.0 → 40.1 billion, with FCF falling from RMB 13.7 billion to RMB 8.73 billion; ROE attributable to parent is only 4.39%; dividend yield 1.00%; minority interests of RMB 74.37 billion are 35.3% of total equity (a great deal of production-line profit does not belong to listed-company shareholders).
• Cycle position: mainstream TV panel prices rose across the board from January to April 2026 and stabilised in May; entering July, terminal restocking slowed, supply and demand loosened, and price declines for large-size TV panels widened markedly, with the industry having already cut utilisation in May–June (Yicai²⁸). In other words, the cycle position is in the early stage of the downturn after the first-half peak.
• Final judgement: the earnings-delivery path of a stake-increase filing = none. It changes no operating variable. Treat it as a sentiment hedge, not confirmation of a bottom. The real verification point is whether it can close green on volume today and hold RMB 5.58.

9张江高科600895watch only (band cut from medium-high to low) · Zhangjiang Hi-Tech

• Related news: The Information (7/27) reported that Chinese state-backed enterprises have begun mass production of immersion DUV lithography machines. The market's speculative logic is that the company holds about 10.779% of Shanghai Micro Electronics Equipment through its wholly-owned subsidiary Zhangjiang Haocheng Venture Capital.
• ⚠️ The sub-agent's three key corrections to that core logic (this report regards this as today's single most important finding):
  1. The "10.779%" comes from filing 2016-003 (an investment of RMB 223.45 million). SMEE has since gone through multiple rounds of capital increase; at the 2025 interim results briefing investors asked three times whether the stake had been diluted, and three times the company avoided the specific ratio, uniformly replying that "the company's subsidiary Zhangjiang Haocheng still holds equity in Shanghai Micro Electronics Equipment; please refer to its official releases for developments". That is: the company only confirms "still holds", and no longer confirms "10.779%". ⚠️ The direction must be written as "unknown" rather than "possibly overstated" — if Zhangjiang Haocheng participated in later capital increases, the ratio could be flat or even higher. This report asserts only that any linear valuation of the stake at 10.779% lacks company endorsement. It should also be noted that "SMEE has since gone through multiple rounds of capital increase" is the sub-agent's inference, with no date or amount given for any round, so it is tagged "to be verified".
  2. The full 2025 annual report (195 pages) contains the four characters "microelectronics equipment" 0 times. The "Shanghai Zhangjiang Microelectronics Port Co., Ltd." that does appear in the annual report is a different, property company (60% held, equity-accounted) — an easily misread same-name trap: Microelectronics Port ≠ Shanghai Micro Electronics Equipment.
  3. The SMEE stake is carried under other non-current financial assets — equity instrument investments (RMB 5.311 billion at year-end, mixed in one pool with other equity investments), measured at fair value, not consolidated, not equity-accounted, and the annual report discloses neither its carrying value nor its fair value separately. The company's 2025 fair-value change gain was −RMB 335 million (negative), but one self-correction is required here: since that line item is a mixed pool and SMEE is not separately disclosed, that −RMB 335 million cannot be attributed to SMEE — it is neither evidence that "SMEE did not appreciate" nor evidence to the contrary. This report withdraws the implication made in the first draft.
• Fundamental verification: 99.47% of 2025 core revenue was "industrial space carriers" (park property), with a lithography revenue share of 0; investment property of RMB 26.576 billion + inventories of RMB 16.719 billion = 68.5% of total assets. 2026Q1 revenue RMB 838 million (−31.28%), net profit attributable to parent RMB 83.51 million (−71.64%), gross margin collapsing from the full-year 48.65% to 25.70%, operating cash flow −RMB 691 million, quick ratio 0.19, gearing 68.67%; RMB 2.35 billion of on-balance-sheet perpetual bonds are counted as equity, and restating them as debt gives about 72.4%. 2026Q1 investment income −RMB 56 million + fair-value changes +RMB 184 million = +RMB 128 million, against net profit attributable to parent of only RMB 83.5 million — i.e. all of the profit comes from investment and fair-value items, with the core business actually a negative contributor. PE 62.01, PB 3.57.
• Technical sentiment: a one-word board on 7/28 with a bid queue of RMB 1.186 billion (about 1.1 million lots), turnover ratio only 1.00% and volume ratio 0.76 — the thickest bid queue and the strongest reluctance to sell in the whole sector, but a one-word board on shrinking volume means the register has not turned over, so selling pressure is concentrated once the seal breaks.
• Final judgement: this is a thematic leader whose "logic can be told but whose earnings can never deliver". The only possible substantive path is a fair-value revaluation or exit realisation after a future SMEE IPO — event-driven option value, with the timing, ratio and amount all unknown. There is no verifiable link between its price and the company's fundamentals; giving it any positive score on the fundamentals dimension would be wrong. First verification point: whether it can post a consecutive board today, and whether any substantive filing about Shanghai Micro Electronics Equipment appears (the 7/28 recap already raised this check item and it still has not appeared).

10永新光学603297watch only · Yongxin Optics

• Related news: included on lithography concept lists; limit-up on 7/28.
• ⚠️ Sub-agent conclusion — this is the only one of the three with an official quantified basis: in its 2025-09-18 unusual share-price movement filing the company stated explicitly: its main business is "optical microscopes and optical components and assemblies", "lithography-related optical components and assemblies account for less than 1% of company revenue", "lithography equipment is not a main application area for the company's products", and there is "no material impact on current earnings"; it also warned investors to treat the lithography concept with caution and guard against concept speculation (National Business Daily²⁹, Yicai³⁰).
• Does it supply Shanghai Micro Electronics Equipment: no reliable data. The company's filings say only "lithography-related" and have never named SMEE; the "supplies projection objectives to SMEE" claims found all come from speculative posts on Xueqiu and similar forums, phrased as "highly likely", which is not evidence and is not accepted by this report. (For contrast: Mloptic has publicly disclosed supplying i-line objective components to SMEE; Yongxin has no disclosure of equivalent grade.)
• Fundamental verification (the healthiest of the three): 2025 revenue RMB 965 million (+8.25%), net profit attributable to parent RMB 220 million (+5.41%); gearing of only 19.58%, current ratio 3.53, OCF / net profit attributable to parent 1.03x, ROE 10.68% (the only double-digit figure among the three). By product: optical components 60.46%, microscopes 37.27%.
• But growth has stalled: 2025 ex-non-recurring was −0.72% (versus +14.25% in 2024); 2026Q1 saw revenue up but profit down — revenue +20.13% while net profit attributable to parent was −17.10%, with gross margin falling from 40.17% to 36.76%, and the public explanation points to USD depreciation causing exchange losses plus lower government grants (that explanation was not verified directly in the company's original filings, tagged "to be verified"); overseas revenue is 46.79%, a large FX exposure. PE (TTM) 53.5 with ex-non-recurring growth ≈ 0.
• Technical sentiment: limit-up, bid queue RMB 121 million, turnover ratio 1.70%; the historical success rate of the top-bid seat on the top-trader list is only 14.01% (among the lowest on the list).
• Final judgement: rated "an indirect supply-chain positive" — it genuinely does have lithography business (this is not fabricated), but the company itself has capped it at "<1%". Even if domestic DUV goes into mass production, that business quintuples and Yongxin wins share, the marginal impact on a RMB 965 million revenue base is still single-digit percentage points. Its financials are the most solid of the three, but that has nothing to do with the lithography theme.

5.11 Sub-agent fundamental verification status

All 3 sets of verification completed: ① the lithography chain (Zhangjiang Hi-Tech / Yongxin Optics / Aopu Optoelectronics) ② buyback, stake-increase and order names (China State Construction / USI / BOE / Shenghui Systems Integration) ③ the software / IT-localisation group (Inspur Software / CETC Digital / Nantian Information / Topsec / Geer Software, anchored on Kingsoft Office).

The verification directly rewrote the following rankings and ratings: Shenghui Systems Integration was raised from outside the top ranks to 1st; China State Construction was cut from "priority deep-dive" to "watch closely" (order elasticity of 0.215%/year); USI fell from 6th to 14th (an operating net contribution of −RMB 20 million); CETC Digital was cut from "watch closely" to "watch only" (the SSE inquiry letter); Topsec and Nantian Information were added and listed directly as Pass; the Pass rationale for Inspur Software and Geer Software was upgraded from "negative PE" to company-level evidence; the evidence on Aopu Optoelectronics was upgraded from an old media article to the original exchange disclosure; and Kingsoft Office's "pre-announced +209.98%–263.89%" was corrected to the adjusted basis of +18.03%~+38.55%.

The five data points most in need of further verification (raised by the sub-agent and adopted here as tracking items):

1. Zhangjiang Hi-Tech's current actual stake in SMEE and its carrying / fair value (highest priority). This requires checking the "other non-current financial assets — equity instrument investments" breakdown in the late-August interim report, and Zhangjiang Haocheng's latest capital contribution ratio in the National Enterprise Credit Information Publicity System. Until then, 10.779% should be treated as a 2016 historical figure.
2. Whether Aopu Optoelectronics has made any new lithography / SMEE-related disclosure after 2026-05-19 (June–July Q&A on the interactive platform, or an unusual-movement filing after the limit-up). If it denies it again, the "pure concept mapping" rating becomes fixed.
3. Shenghui Systems Integration's formal contract signing announcement and gross-margin terms, plus the issuance progress and conversion price of the RMB 550 million convertible bond (to quantify the dilution).
4. The impairment-test details for Topsec's RMB 4.145 billion of goodwill and the mid-2026 impairment risk — it is 43.5% of net assets attributable to parent and 59% of market cap, while the company sits in a combination of "revenue −33%, H1 loss widening 310%", precisely the trigger condition for impairment. If an impairment is recognised, the "PB 0.74, below book" narrative collapses instantly.
5. The five software companies' actual 2026H1 IT-localisation / AI order values or revenue shares — this is the biggest blank in this round of checking: not one of them has disclosed a verifiable IT-localisation revenue breakdown. Until the interim reports arrive in late August, "these companies are supported by IT-localisation orders" is an unverifiable inference.

---

6. Pass List

A self-imposed constraint that must be stated (from the 7/28 recap): the Pass list asserts only "should not be bought" (a fundamental judgement) and does not assert "will fall" (a price judgement). Of the 20 names on the 7/28 Pass list, 14 hit, yet the price direction was wrong on 6 including 创新医疗, 南京熊猫 and 海南海药 — their fundamentals were indeed poor, but they hit limit-up that day. The same applies here: the names above may still rise today, and this report states only that they have no verifiable fundamental basis for purchase.

---

7. Intra-Branch Ranking

Branch 1 · Beneficiaries of falling oil (A-shares barely reacted on 7/28)

• Hardest stock: 春秋航空 (601021). Best suited to main-board trading: 中国国航 (601111) (liquidity) or 春秋航空 (601021) (quality). Who is following: 南方航空, 新凤鸣. Who is a Pass: 中国东航 (600115), 恒力石化 (600346).

Branch 2 · Cleanroom / electronics plant capex

• Hardest stock: 圣晖集成 (603163) (the only one that won the order). Best suited to main-board trading: 圣晖集成. Who is a Pass: 鹏鼎控股 (its limit-down bid queue is unabsorbed).

Branch 3 · Lithography (mapping-quality tiers — the most important section of this report)

Single-day clue one (not "decisive evidence"; the wording has been toned down per QC): on 2026-07-28 all 6 names that hit limit-up were the low-turnover concept shells furthest removed from lithography manufacturing, while every company that genuinely has lithography components / equipment business fell heavily —

Company-level evidence two (this one is hard, unlike the single-day clue above): none of the three main-board names has ever disclosed an order, customer or revenue from domestic DUV lithography. And there is an obvious valuation-fundamentals mismatch — the most weakly rated, Aopu Optoelectronics, is the most expensive (PE 106.0 / PB 7.2 / ROE 5.7%), while the relatively most substantive, Yongxin Optics, is the cheapest (PE 53.5 / ROE 10.7%), yet all three hit limit-up together on 7/28. This shows that day's pricing was ordered by "the market's imagination about thematic relatedness", not by verifiable evidence.

⚠️ Two weaknesses in this inference (raised by risk-auditor and accepted): ① the sample is a single day, 7/28, so the conclusion needs 2–3 trading days of confirmation; ② there is board-level beta contamination — all 7 "genuine chain names" above are on the STAR Market / ChiNext (the STAR 50 was −6.33% and the ChiNext Index −7.35% that day) while all 6 limit-ups are on the main board (the SSE Composite was −1.16%), so "genuine chain falls / concept shells rise" is to a considerable extent really "STAR and ChiNext fall / main board rises". There is no reliable source for same-board comparison data, and this report could not rule out that contamination.

The conclusion under those limits: theme strength A, stock-level delivery C. The "stock-level delivery C" is supported by the sub-agent's company-level evidence (Aopu's two denials, Yongxin's self-imposed <1% cap, Zhangjiang Hi-Tech's zero lithography revenue) and is hard; "the market is not pricing the industrial logic" is only a single-day clue and is soft.

• Hardest stock: none. The genuine chain companies are all on the STAR Market / ChiNext and all fell heavily that day, while everything available on the main board is a concept shell. Best suited to main-board trading: 张江高科 (600895) (in terms of trading recognisability only). Who is following: 永新光学, 电科数字. Who is a Pass: 奥普光电, 深纺织Ａ.

Branch 4 · The buyback / stake-increase support wave

• Hardest stock: 中际旭创 (300308), but not main board. Best suited to main-board trading: 京东方 A (000725). Who is following: 科达利, 国城矿业. Who is a Pass: 重庆港.

Branch 5 · Defensive (banks / baijiu / telecom operators)

• Hardest stock: 农业银行 (601288). Best suited to main-board trading: 农业银行 / 工商银行. Who is a Pass: 金种子酒 (600199).

Branch 6 · Software / IT localisation (0 consecutive boards, not yet a mainline)

Independent financial verification of this branch is complete, and the conclusion is: the whole branch is not recommended. The sub-agent checked 5 main-board limit-up names: 4 posted single-quarter 2026Q1 losses (only CETC Digital was marginally profitable), 3 have already filed H1 losses or continued losses, and 1's H1 "swing to profit" rests on winning a lawsuit; not one of the 7/28 limit-ups corresponds to any new positive filing — the relevant filings from 7/14–7/21 are instead two loss pre-announcements, one "still loss-making ex-non-recurring" pre-announcement, and one reply to an exchange inquiry letter.

• Hardest stock: 金山办公 (688111), but it is not main board and barely moved that day. Not one of the 7 main-board names is "hard".
• The relatively most substantive on the main board: 电科数字 (600850) (the only profitable one), but even it only reaches "watch only".
• Who is a Pass: 南天信息, 浪潮软件, 天融信, 格尔软件 — all four have negative evidence filed by the companies themselves.
• The single most important sentence about this branch: money bought the 5 with no earnings and went around the only one with earnings. This is the strongest circumstantial evidence for "risk-averse rotation rather than industry-driven", and the core reason not to participate in this branch today.

---

8. Opening Verification Signals for Today

① Auction signals (09:15–09:25)

• 圣晖集成 (603163): the first verification point for this report's top name. It fell −7.36% on shrinking volume yesterday, and the RMB 858 million order is post-close news. The auction move reads out the market's pricing of that order directly: a clearly higher open = its size is accepted; a flat or lower open = the market has already priced in "no signed contract + unknown term + downstream semiconductor capex just further suppressed by SK Hynix's miss". This report records the reading only and gives no price judgement.
• 张江高科 (600895): yesterday's one-word board had a bid queue of RMB 1.186 billion. Watch two things — (a) whether it is still a one-word board (and if so, whether the queue is >RMB 500 million; halving means divergence); (b) if it opens up less than 5%, the lithography branch is most likely a one-day wonder today. A turnover ratio of just 1.00% means the register has not turned over, so selling pressure is concentrated once the seal breaks.
• The lithography ladder (永新光学 603297, 海立股份 600619, 奥普光电 002338, 深纺织Ａ 000045): the core verification — whether any of them posts a second board. All 6 were first boards, 0 consecutive, with no height; if not one posts a consecutive board today, the branch is confirmed as a standard one-day wonder.
• 兆易创新 (603986): yesterday's limit-down bid queue was RMB 2.630 billion. SK Hynix's results have landed and missed, so the possibility of "a potential upside surprise" is gone — what the auction now verifies is something else: with the negative confirmed, does the RMB 2.630 billion queue keep building (another day of memory-chain selling) or get absorbed (the bad news is out)?
• 中际旭创 (300308): the size of the auction gap determines directly whether the triple support of "an RMB 4–8 billion buyback + RMB 2.839 billion of net buying on the top-trader list + the HK$980 H-share par anchor" is effective. Opening up more than 5% without fading = the bottom-fishing money is genuine; opening up and fading = yesterday's RMB 2.839 billion of net buying was catching a falling knife.
• Airlines (Spring 601021, Air China 601111): the auction move is the direct reading of "whether the market is willing to price the overnight −4.1%": close to flat = the 7/28 state of not pricing continues; a clearly higher open = willingness to price has turned. This report records the reading only and gives no price judgement.

② Sector signals (09:30–10:30)

• Lithography: do 3 or more names hit limit-up quickly? Is there a core leader moving on volume? The key is whether a second-board ladder appears — yesterday's 0 consecutive boards is this branch's biggest structural defect.
• Software / IT localisation: this report has downgraded the whole branch to "not recommended" (all 5 main-board limit-up names are fundamentally falsified). It is observed only today: 2–3 second boards must appear (candidates: 电科数字 600850, 南天信息 000948, 久其软件 002279, 天融信 002212, 科蓝软件 300663). If it is still "all first boards, 0 consecutive", it is confirmed as residual risk-averse warmth and should be deleted from the branch list tomorrow. A cleaner test: watch whether 金山办公 (688111) catches up today — if what rises is still the loss-makers rather than it, this line had nothing to do with the "IT localisation industry" from start to finish.
• Airlines / transport: can the sector rise more than 2% overall? This is the only macro branch where "the news has happened and the price has not reacted"; if there is still no reaction today, the market's willingness to price the oil positive is extremely low and it should be downgraded tomorrow.
• The semiconductor / PCB engineering chain (圣晖集成 603163, 亚翔集成 603929, 柏诚股份 601133): if Shenghui opens higher and its peers follow, the market accepts the counter-narrative that "electronics plant capex is not slowing"; if Shenghui rises alone and peers do not move, this order is only a stock-specific event and does not constitute a branch.
• Banks / baijiu: can they sustain 7/28's volume ratios above 1.4? If volume ratios fall below 1 but they still close green, it is passive catch-up rather than active allocation.
• AI hardware: the speed at which limit-down bid queues are absorbed is the only indicator. GigaDevice RMB 2.630 billion, Dongshan Precision RMB 1.346 billion, Cambridge Technology RMB 422 million, Tongfu Microelectronics RMB 304 million.

③ Stock-level signals

• Top-5 absorption: the volume absorbed in the top 5 levels after Zhangjiang Hi-Tech's seal breaks; whether Innolight sells off on volume within 10 minutes of a higher open.
• Fast sealing: whether the second-board candidates in lithography seal before 10:00 (the later the seal, the weaker).
• Volume without breaking: whether yesterday's broken-seal names (Inspur Software once, Geer Software twice, CETC Digital once, Shenzhen Textile A once) can repair today.
• Large-order net inflows: if the banking sector keeps large-order net inflows above RMB 1 billion, the defensive mainline holds.
• Being crowded out by a stronger name: if lithography posts a second board today while software is still 0 consecutive, software's money will be drained away by lithography.

④ Risk signals (any one of these lowers expectations)

1. A higher open that dives: if Zhangjiang Hi-Tech's one-word board breaks and it falls below a 5% gain within 30 minutes → the lithography branch ends for the day.
2. A lone limit-up: if Zhangjiang Hi-Tech is the only lithography name to seal → no ladder, Pass the whole branch.
3. The back row falls back: if yesterday's 12 software limit-ups close broadly green-to-red today → the risk-averse rotation is over.
4. The core does not follow: if Innolight opens higher and fades → yesterday's RMB 2.839 billion of net buying failed, and the AI hardware chain keeps selling off.
5. A weak index: if the SSE Composite breaks 3780 (it closed at 3813.31 on 7/28) → whole-market risk appetite keeps contracting.
6. Yesterday's strong themes retreat: if the second-board names in yesterday's consecutive ladder (16 consecutive-limit-up names in this report's measured limit-up pool: 爱丽家居 6 boards, 顺钠股份 4 boards, 新亚制程 3 boards, plus 13 on 2 boards) break their seals en masse, the speculative sentiment layer retreats in step — this is the only layer still supporting the market yesterday.
7. SK Hynix results missing expectations (to be observed) → ⚠️ confirmed to have happened: revenue 5.5% below consensus and operating profit 5.4%~6.4% below. This risk trigger condition, set by this report itself, has been hit, so the memory chain should be treated today as "the negative has landed" rather than "the negative is unverified".
8. Shenghui Systems Integration files that "the contract could not be signed" or is delayed → the top name's logic is void immediately.
9. The July Politburo meeting convenes that day with a conservative tone → a two-way variable requiring real-time tracking.

---

9. Summary of the Day's Observation Priorities

This section is a research observation record and contains no investment advice, ratings, target prices or position recommendations.

① The 5 main-board stocks placed under observation today

Statement up front: this report holds that no high-confidence name exists today, and there is no "priority deep-dive" name today. Of the 5 below, only the first has a quantifiable company-level event, and that event's formal contract has not been signed; the other 4 are all low-elasticity observation items. The lesson of 7/28 was that "a fixed output structure forces out recommendations that do not exist" — this report retains the list, but tags each name with its evidence grade and invalidation conditions, and includes no return forecast or price judgement.

Today's most important note is not in this table but in Section 9 ③ item 1: SK Hynix's results have landed and missed, the only major new development confirmed before the open, and its direction is negative.

② The 3 strongest branches today

Two notes:

1. The oil branch falls from 1st to 3rd — the first draft made it the headline catalyst on the basis that "the largest single-day decline happened after the A-share close". QC proved that timeline wrong: WTI −7.5% settled in the early hours of 07-28 Beijing time, visible to A-shares all through 7/28 with almost no reaction. The expectation gap therefore does not hold, and the genuinely new overnight move is only about half the magnitude.
2. Domestic DUV lithography is excluded from this table — theme strength A, stock-level delivery C: after the sub-agent's checks on the three main-board names, one denied it in an exchange document, one was capped by the company at "<1% of revenue", and one has a lithography revenue share of 0 with its core shareholding data out of date. It is still the branch with the highest trading recognisability today, but it is not the branch with the strongest positive — these two things must be said separately.

This table lists what is "relatively strongest", not what is "strong". All three positive branches today are clearly weaker than the single negative development (SK Hynix's miss + a 3rd consecutive down day for the SOX).

③ Directions not worth chasing today

1. The whole AI hardware chain (CPO / optical modules / memory / CCL / PCB / semiconductor equipment) — today's reason for avoidance has been upgraded from "overnight sentiment" to "the participants' own results": SK Hynix's 2Q26 revenue was 5.5% below consensus and operating profit 5.4%~6.4% below (landing before the open); the overnight SOX was −4.49%, a 3rd straight down day, with Micron −nearly 9%, AMD and ARM −more than 8% and SanDisk −more than 14%; unabsorbed limit-down bid queues from 7/28: GigaDevice RMB 2.630 billion, Dongshan Precision RMB 1.346 billion, Cambridge Technology RMB 422 million, Tongfu Microelectronics RMB 304 million. The opposing facts must also be stated, and today they are actually stronger: SK Hynix's DRAM ASP was +about 30% QoQ, NAND ASP +mid-50% QoQ, and the 76.3% operating margin is an all-time high, and the company also said "phone and PC sales are temporarily being adjusted because memory is hard to obtain" — the price cycle is not over; what missed was volume and market expectations, not price. So this avoidance is at the trading level (selling pressure unabsorbed, the negative just landed), not an industry-level judgement.
2. The "genuine chain" companies within the lithography chain — Focuslight, Chipmore, Fortune Precision, Tenjing Sci-Tech, King Lai, Saiwei Electronics. They belong simultaneously to "lithography beneficiaries" and "semiconductor equipment being de-rated", and 7/28 proved the latter won.
3. 奥普光电 (002338) and 深纺织Ａ (000045) — the former twice stated explicitly in an exchange disclosure document that it "has no business dealings with Shanghai Micro Electronics Equipment for the time being", and the latter's main business is polarising film, unrelated to lithography.
4. Treating 环旭电子 (601231)'s earnings flash as evidence that "the memory chain has bottomed" — on breakdown, its RMB 206 million increase in total profit came entirely from RMB 226 million of finance-cost savings, and the net contribution from operating businesses was −RMB 20 million; moreover, memory price increases are a cost item for it, not a price item, so the logic may be directionally backwards.
5. The entire software / IT-localisation branch — the sub-agent checked 5 main-board limit-up names: 4 posted single-quarter 2026Q1 losses, 3 have already filed H1 losses or continued losses, and 1's "swing to profit" rests on winning a lawsuit; not one limit-up corresponds to a new positive filing, and the most recent relevant filings are instead two loss pre-announcements, one "still loss-making ex-non-recurring" and one reply to an exchange inquiry letter. The most glaring item: Topsec filed in black and white on 7/15 that its H1 loss widened 309.6%, and still sealed limit-up on 7/28. Meanwhile Kingsoft Office, which genuinely had earnings that day, rose only 0.18% — money bought the ones with no earnings and went around the one with earnings.
6. Thematic limit-up names with negative TTM earnings — 浪潮软件 (−20.34), 格尔软件 (−55.74), 金种子酒 (−28.10), 爱丽家居 (−207.05), 创新医疗 (−218.02). And note: both of Topsec's current figures, "PE 169" and "PB 0.74", will become invalid once the interim report lands (TTM net profit attributable to parent turns negative; stripping out RMB 4.145 billion of goodwill gives a tangible PB of about 1.63).
7. High-level lone names with no volume — 爱丽家居 (603221) has 6 boards but turnover of only RMB 13 million and a 0.33% turnover ratio; 艾艾精工 (603580)'s seal broke 5 times; 顺钠股份 (000533)'s seal broke 12 times with a 28.08% turnover ratio.
8. 中国石油 (601857) and 中海油服 (601808) — PetroChina rose +1.83% against the trend on 7/28 while the overnight 7.5% single-day oil collapse is not yet reflected, so they are on the pressured side today.
9. 长鑫科技 (688825) — on its 3rd day of listing it still traded RMB 44.428 billion on 7/28 with a 20.66% turnover ratio, the largest single drain on liquidity in the whole market.

④ The final one-sentence judgement

The only major new development confirmed before today's open is negative — SK Hynix's 2Q26 revenue and operating profit both missed, and stacked with the overnight SOX at −4.49% (a 3rd consecutive down day), the driver of the decline in the memory / AI hardware chain has been upgraded from "overnight sentiment" to "the industry participants' own results"; meanwhile this report's first-draft headline argument (that A-shares had not priced oil) was overturned by the timeline in QC — A-shares could see WTI −7.5% all through 7/28 and barely reacted, which is low willingness to price, not an opportunity. On the positive side only one quantifiable company event remains (Shenghui Systems Integration's RMB 858 million contract win), but its formal contract is unsigned, its term is unknown, and its downstream is precisely the semiconductor capex just further suppressed by the miss. The two hottest themes — lithography and software / IT localisation — both fail entirely on company-level evidence.

This report's conclusion today is therefore: no top-priority name, the negative development outweighs the positive ones, and every item on the list is an observation item rather than a recommendation.

---

⚠️ Risk warning: this list is only a pre-market information review and observation and does not constitute investment advice. The "priority deep-dive / watch closely / watch only / Pass" labels in the text are research-priority labels, not investment ratings, and carry no buy or sell meaning; the text contains no target prices, entry levels, return forecasts or position recommendations. A-share volatility risk is extremely high, and automatically generated content may contain timeliness gaps or industry-chain mapping errors; it must not be used directly as a basis for trading.

The first draft of this report took the overnight overseas data from the wrong trading day throughout; risk-auditor QC found this and it was corrected before publication (see the "Major correction" at the top). That error directly overturned the first draft's headline argument. This disclosure is retained because readers have a right to know what a report's conclusions have been changed from, and why.

---

Data sources: Tencent quotes qt.gtimg.cn (stock and index closing prices, turnover, turnover ratios, volume ratios, PE, PB, market cap, 2026-07-28), Eastmoney push2ex (getTopicZTPool/getTopicDTPool, date=20260728, 61 limit-ups / 49 limit-downs), Eastmoney datacenter-web (RPT_DAILYBILLBOARD_DETAILSNEW, the 2026-07-28 top-trader list), and the Eastmoney push2 sector flow interface. Financial statement and filing checks were pulled with akshare by the fundamentals-analyst sub-agent on 2026-07-29, which parsed the original filing PDFs (the full 195-page Zhangjiang Hi-Tech 2025 annual report, the Zhangjiang Hi-Tech 2025 interim results briefing record, Aopu Optoelectronics investor-relations record 2026-05-001, Yongxin Optics' 2025-09-18 unusual-movement filing, USI's H1 earnings flash, Shenghui Systems Integration's contract-win filing and 2026Q2 operating report, CETC Digital's reply to SSE Official Letter [2026] No. 1059, and the 2026 interim pre-announcements of Topsec / Nantian Information / Geer Software / Inspur Software, plus Kingsoft Office's voluntary 2026 interim pre-announcement filing). News sources are linked in each item of Section 1. Reconciliation baselines: reports/a-share/2026-07-28.md, reports/a-share/2026-07-28-recap.md.