A-Share Pre-Market Brief | 2026-08-10 (Monday)

Data scope disclosed up front (please read this section first)

1. News window: this issue covers the weekend window from the 15:00 close on 2026-08-07 (Friday) → 07:00 on 2026-08-10 (Monday). Every item carries its publication time; anything whose publication time cannot be confirmed is down-weighted and flagged.
2. Market-data basis: throughout this issue, the price moves, turnover rates, net main-force inflows and 5/20/60-day gains of A-share individual stocks are all as of the 2026-08-07 close, taken from the Tencent Finance single-stock snapshot (5,539 entries market-wide). The limit-up pool / failed-limit-up pool are taken from East Money (74 names). Industry price moves and industry net inflows are taken from the Tonghuashun industry taxonomy (90 industries). Industry net inflow and single-stock net main-force inflow belong to two different measurement systems; they cannot be added together and cannot substitute for each other.
3. Price limits are tiered by board, and every stock in this issue is tagged with its board: Shanghai/Shenzhen main board ±10%, ChiNext (30/301) ±20%, STAR Market (688) ±20%, Beijing Stock Exchange ±30%.
4. Two data items not obtained / left blank in this issue:
   • "Two consecutive days of net main-force inflow" was not obtained in this issue and is left blank. The rule laid down in last Friday's recap was: "net main-force inflow must be read over two consecutive days; a single-day ranking will not do." This issue was unable to obtain per-day fund-flow data for individual stocks, so every "net main-force inflow" figure in this issue is a single-day 2026-08-07 value. Please do not treat it as evidence of sustained inflow. The sole exception is 沪电股份 (WUS Printed Circuit, 002463), whose two-consecutive-day inflow was already confirmed in last Friday's recap. This is the item in this issue that readers most need to verify for themselves during the call auction and the session.
   • Two fund-flow measurement systems had opposite signs on two stocks; one of them has now been adjudicated with exchange data: 中国稀土 (China Rare Earth Resources and Technology, 000831) — Tencent basis +RMB 1.004 billion, Tonghuashun basis −RMB 692 million. Friday's Dragon-Tiger List (exchange-disclosed, already published as this issue went to press) shows net buying of +RMB 632 million in the stock, the largest single net buy of the whole market that day, agreeing in sign with the Tencent basis. Under the source-priority rule "exchange disclosure > authoritative media," this issue adopts "net inflow," and the Tonghuashun "real-time" tier is judged unreliable. 凯莱英 (Asymchem, 002821) (Tencent +RMB 329 million / Tonghuashun −RMB 52 million) did not appear on the Dragon-Tiger List, so it cannot be adjudicated and its fund-flow conclusion remains down-weighted.
5. One verification conclusion that has to come first: within this window a search engine returned a Cailianshe piece, "Weekend News Roundup: Brokerage Sector Gets a Major Positive as Wu Qing Says Leverage Caps Will Be Moderately Widened; A-Shares to See Over One Hundred Billion Yuan of Incremental Funds," which reads exactly like a roundup of this weekend. Tracing it item by item confirmed the article was published on 2025-12-07, 8 months ago, and has nothing to do with this window. Not a single item from it is used in this issue. See section 1.3.

⚠️ Data-collection and implementation notes for this run (internal only):

• The per-day single-stock fund-flow interface failed across the board: stock_individual_fund_flow returned ConnectionError / RemoteDisconnected (push2his.eastmoney.com) for all 12 candidate stocks across 3 consecutive retry rounds. Yet the stock_zt_pool_em family and stock_zh_a_spot_tx on the same domain all worked normally — once again an interface-level rather than a domain-level failure, the same shape as the 8/7 incident. → The consequence is very concrete: the new rule written down in last Friday's recap — "net main-force inflow must be two consecutive days" — could not be executed the very first time it was applied. I did not pass single-day data off as consecutive data; instead I wrote "cannot be done" explicitly in the scope disclosure. I suggest splitting per-day fund flow into an independent fetch step with exponential backoff and caching the previous day's result locally, otherwise this new rule will stay a dead letter indefinitely — it depends on precisely the least stable interface.
• Friday's Dragon-Tiger List was obtained this time, and it directly rescued a conclusion: stock_lhb_detail_em(20260807) returned normally on Monday morning. The failure to obtain it at 15:40 in last Friday's recap was a scheduling problem (the exchange only discloses after 18:00), not an interface failure — that memory item took effect for the third time today, this time validated in the form of "successful next-day back-fill." The more practical value: it adjudicated the fund-flow discrepancy on 中国稀土(000831) (Dragon-Tiger net buy +RMB 632 million, the largest of the day, same sign as Tencent, opposite sign to Tonghuashun). → This suggests a cheap improvement: the pre-market flow could routinely back-fill "the previous trading day's Dragon-Tiger List" — extremely low cost, yet it gives exchange-grade adjudication of fund-flow cases left hanging by the previous evening's recap. I suggest adding it to the pre-market fetch list, not just to the post-close 18:30 re-run.
• Tonghuashun's stock_fund_flow_individual works but only has the "real-time" tier: both 3日排行 and 5日排行 returned empty, so it cannot serve as a substitute check for sustained inflow. And the semantics of the "real-time" tier at 07:00 on a Monday are themselves ambiguous (is it a Friday-close snapshot or something else?). That is the main reason I down-weighted this basis as a whole and did not use it to overturn the Tencent basis — not because its numbers looked unflattering.
• The zllr_d5 field is a trap; I did not fall for it this time, but only just: in the Tencent snapshot zljlr is net main-force inflow (in RMB 10k), but zllr_d5 is not 5-day net inflow — it is 5-day cumulative inflow (gross). Using it directly to compute a "5-day net" produces absurd numbers like RMB 72.7 billion for 兆易创新 (GigaDevice, 603986). I caught it via a ratio test (single-day / 5-day = 0.02), not by remembering what the field meant. I suggest hard-coding the practice of "run a plausibility ratio test on any multi-day field first."
• I parsed the 涨停统计 field of the limit-up ladder wrongly the first time: the format is x/y = x limit-ups within y days; I initially read y as the consecutive-limit-up count, which produced "6 consecutive boards for 汇绿生态 (Huilv Ecology, 001267), 5 for 宝鼎科技 (Baoding Technology, 002552)" — contradicting the "4 boards at most" written in last Friday's recap. It was that contradiction itself that sent me back to re-check — the real ladder tops out at 3 names with 4 consecutive limit-ups (百花医药 (Baihua Medical, 600721), 沃格光电 (Woge Optoelectronics, 603773), 云南锗业 (Yunnan Germanium, 002428)); 汇绿生态 is actually 6 limit-ups in 10 days, and 宝鼎科技 5 in 6 days. Lesson: when a number does not line up with the previous report, suspect this issue's parsing first, rather than rushing to rewrite the conclusion.
• The search engine's failure mode this time was "an article from last year masquerading as weekend news": the headline, genre and column of Cailianshe cls.cn/detail/2221983 all fit "this weekend's news roundup"; only the body date says 2025-12-07. This is the fifth appearance of the same family as the memory item "AI attribution pieces stitching in old announcements," except the vehicle changed from an ETF marketing piece to a search engine's broken time ordering. Had I taken it at face value, the first branch of this issue would have become "brokerages get more leverage + hundreds of billions of incremental funds" — a completely fabricated pivot. Process assertion: for every "news roundup"-type aggregation article in a weekend window, check the article's own publication date before citing it, not just the dates of the events it cites.
• The AI summary returned by the first search was almost entirely broken: that summary offered "a 药明康德(603259) ban, a SK Hynix RMB 38.4 billion capacity expansion, Unitree's subscription tomorrow, mutual funds piling into computing-power ETFs, and a shockingly weak US payrolls print ending Fed rate hikes." Checked item by item: the Unitree subscription is right; 药明康德 is directionally right but stated backwards (药明康德 won the injunction; it was not "banned"); payrolls had the right number but the wrong conclusion (CME still shows a 44.4% probability of a 25bp hike in September, not "hikes are over"); SK Hynix's "38.4 billion" is actually a USD 38 billion new memory fab (a capacity expansion that is bearish for memory prices, written up in the summary as bullish). Only 1 of the 5 items was directly usable. The danger of this kind of summary is that every item is "adjacent to true" — skip the item-by-item check and you copy the whole paragraph.
• ⚠️ The single thing most worth remembering from this issue is that a supposed "self-refutation" was itself a misreading, and risk-auditor caught it. Here is what happened: I was about to put memory in the first branch, then went to check overseas pricing and saw a Citi research summary saying "We see both DRAM and NAND prices decelerating Q/Q in the next four quarters, with prices peaking in 2Q of next year." I read "decelerating" as "falling", and so wrote "DRAM/NAND prices decline Q/Q over the next four quarters," on which basis I demoted memory to the 4th branch — and then congratulated myself here on a "self-refutation that saved the piece." Citi's actual numbers are DRAM +23%/+9%/+2%/0% Q/Q and NAND +29%/+7%/0%/−1%, with the turn down only in 2027H2, and they maintain a "Buy" rating. In other words prices still rise 23% next quarter; what is decelerating is the acceleration, not the price. And Citi explicitly attributes the deceleration to capacity expansion at CXMT and YMTC — which positively supports this branch's "share shift" logic, and I failed to cite it at all. → Two lessons; the second matters more: ① decelerate ≠ decline, and "peaking in 2Q next year" means it keeps rising until then. Directional verbs in English research (decelerate / moderate / soften / normalize) must be pinned to the specific number series before drawing a conclusion; reading only the summary sentence turns the second derivative into the first. This is the same root cause as [[english-summary-billion-yi-tenfold-error]]: the magnitude/direction of an English summary getting distorted in Chinese translation, and the fix is identical — find the number series and back it out. ② The narrative "I refuted myself" manufactures false confidence. Because I had completed a "self-correction," I never went back to check that report's original numbers — the act of self-refutation itself became a free pass from inspection. That is more dangerous than getting it wrong from the start, because it is disguised as rigor. Process assertion: any conclusion of the form "I was going to write A, then after checking changed it to B" must have B run past a primary source too; being the product of a correction does not exempt it.
• The tense of CPI computers +17.4%: it is July data reflecting price increases that have already happened; Citi's figures are forward-looking. After the correction the two are actually pointing the same way (both rising) and are not in opposition; the body text has been rewritten as "consistent, but with the pace narrowing." One more order-of-magnitude note: the statistics bureau also explains that these three items together lifted headline CPI by only about 0.14 percentage points year on year; the body text's claim that it is "harder evidence than any supply-chain field check" refers to evidence grade, not macro weight, and this is now marked in the body.
• I got a basic fact wrong in the memory section, and a sub-agent caught it: 长鑫科技 listed on 2026-07-27 (688825), whereas my first draft said "CXMT is unlisted, so you can only look for proxy names." This is not a wording issue — it dropped the best name in the entire branch: the direct beneficiary is itself tradable on-exchange, with a market cap of RMB 3.5 trillion, the largest A-share IPO of the year. Rewritten after confirmation from three sources: stock_info_sh_name_code (the SSE official name registry) + the Tencent snapshot + an independent search. → A very concrete lesson: for status-type facts such as "is company X listed," I was using memory rather than a query. Such facts have the property that they are almost always right until they go stale, and then they are fatal, and unlike numerical errors they do not expose themselves. Process assertion: any claim of the form "X is unlisted / has no corresponding listed name" must have the registry checked on the spot; writing from impression is not allowed.
• Another systemic problem corrected by a sub-agent: every PE I cited was on a "TTM as of end-Q1" basis, while the interim pre-announcements had all been disclosed between July 3 and 16. The 29.2/30.1/101.8/22.3 given by the Tencent snapshot are entirely correct on recomputation, but the E they use is the Q1 basis; recomputed with the H1 pre-announcements they become 13.9/14.2/36.7/13.3. This is not the snapshot's fault; I simply failed to realize E had rolled over. The body text now lists both bases side by side. → What is more worth remembering is the criterion the sub-agent supplied: these four names sit at the 3%–7% historical percentile on PE and the 70%–82% percentile on PB — a systematic divergence. That is the textbook shape of peaking cyclical earnings; at this position a "low PE" is a trap, not a discount. 德明利 (Demingli Electronics, 001309) is the extreme case — the lowest PE in the whole list, and simultaneously the only one of the four whose Q2 net profit attributable to parent turned negative quarter on quarter.

---

0. Today in One Sentence

The only primary-verified catalyst over the weekend that "changed a constraint" is judicial, not policy. On 2026-08-07 US time, the US District Court for the District of Columbia granted 药明康德 (WuXi AppTec, 603259; SSE main board) a preliminary injunction against the US Department of Defense's 1260H "Chinese military company" designation, barring the DoD from enforcing that designation for the duration of the litigation, and requiring the company to post security of only USD 1 — the court also found all three of the DoD's listing rationales seriously defective. This is the item with the highest logical hardness in this window, because it is not an opinion, not a research note, not a price-hike rumor, but a verifiable, already-effective change of legal status.

The second item is macro, and A-shares had not priced it as of Friday's close. US July nonfarm payrolls unexpectedly fell by 23,000 (consensus +80,000 to +100,000), with May/June revised down by a combined 103,000 — the data were released at 20:30 Beijing time on 8-07, after the A-share close. The result: spot gold rose 7.44% on the week to USD 4341.91/oz, the largest weekly gain since the week of January 23. The A-share precious-metals sector closed up 4.11% on Friday, but that gain does not contain the payrolls print. This is the cleanest expectation gap this issue can identify.

The third and badly underrated item is official data: in July CPI, computer prices rose 17.4% year on year, tablets 17.2% and mobile handsets 8.5%, with all three gains widening, and the National Bureau of Statistics explicitly attributes this to "artificial intelligence driving the iteration and upgrading of consumer electronics." This is the first time AI-driven shortage-led price increases have shown up in China's official inflation statistics — harder evidence than any supply-chain field check.

The fourth is a fact that requires updating one's common sense: 长鑫科技 (ChangXin Technology, 688825; STAR Market) listed on 2026-07-27, closed at RMB 52.48 on 08-07, with a total market cap of RMB 3.51 trillion and a free float of only 6.7%. So for the news that "Apple is testing CXMT memory chips" (WSJ 08-09), the direct beneficiary is tradable on-exchange and there is no need to buy proxy names — while 兆易创新(603986), widely treated as a "CXMT concept" name, itself clarified on May 26 that its stake is only 1.80% and does not affect net profit. But the direction of this chain must be written precisely: although Citi cut its Micron price target (USD 1400 → 1150) it maintained "Buy," and its forecast is DRAM Q/Q of +23%/+9%/+2%/0% over the next four quarters — the price-increase cycle is not over, what is ending is the acceleration; and Citi attributes the deceleration directly to capacity expansion at CXMT and YMTC, which amounts to a sell-side confirmation of the share shift. A-share memory names have fallen 46%–58% since June 29, following the overseas valuation de-rating and deceleration expectations, not a turn down in prices.

Likely direction of funds: innovative drugs / CXO (a primary-verified catalyst, but the position is already high), precious metals (a hard price move, not yet priced), AI computing power and PCB (a cluster of interim reports delivered on the evening of 08-07). Driver types: event (judicial) + overseas macro + official data + earnings — the one thing absent is new domestic policy. No incremental policy at the NDRC / MIIT / State Council level was found within this window.

Pre-market state call: a higher open is likely, but this is a "easy to gap up, hard to grind higher" combination. Of the three main catalysts, the innovative-drug sector already surged last Friday (medical services +8.34%, the strongest in the market) and precious metals are up 22%–31% over the past 20 days; a "good news realized" gap-up-then-fade is today's single biggest risk. The genuinely low-positioned names (memory, oil & gas) happen to be the two where the catalyst is questionable or points the other way. This issue therefore puts its observational weight on the core names of the main lines rather than on ladder promotion — 4 consecutive limit-ups was already the market high last Friday, the 10-board run has broken, and the attrition rate at height stays extremely high this week.

---

1. News Overview

1.1 Core News Table

1.2 Notes on Impact Grades

• S (2 items): the 药明康德(603259) judicial injunction and the negative US July payrolls print. The former changes a leading company's substantive operating constraints; the latter changes the global pricing environment and A-shares have not priced it.
• A (4 items): gold's 7.44% weekly gain, the July CPI consumer-electronics price increases, the evening of 08-07 cluster of interim results, and Apple testing CXMT (upgraded to A because the direct name 长鑫科技(688825) is already listed). There is also 1 reverse-direction A-grade item: the slump in overseas memory stocks plus Citi's bearish view on the memory price cycle.
• B / C: the rest have longer causal chains, need verification, or are sentiment impulses with weak persistence.

1.3 News Falsified / Down-weighted Within This Window (Important)

---

2. Strongest Positive Branches, Descending

2.1 Branch-by-Branch Notes

① Innovative drugs / CXO — the only "hard catalyst" in this window, and also the most extended branch

• Why it is positive: a 1260H designation materially affects overseas clients' willingness to cooperate and their compliance costs. The court not only granted the injunction but rejected all three DoD rationales one by one — the shareholding ratio mistook a 5.32% share of a fund's net assets for an equity stake (the actual holding is 0.001%); it misread 药明康德(603259)'s status as a third-party laboratory as a university partnership; and the applicant in the military-hospital collaboration was in fact Novo Nordisk, not 药明康德. And the court required security of only USD 1, which in US judicial practice usually signals a high assessment of the plaintiff's likelihood of success.
• Short-term or medium-term: medium term. The injunction holds for the duration of the litigation, and the substantive constraint has been lifted.
• Path to earnings delivery: yes. On 08-03 药明康德(603259) already raised full-year 2026 guidance (total revenue RMB 51.3–53.0 billion → RMB 58.5–60.5 billion) — note this is 7-day-old news and is not counted as a weekend catalyst in this issue, but it shows the earnings channel itself is open.
• Is it already fully expected — this has to be answered on three levels, or the answer will be wrong:
  1. Information level: the injunction itself has not been priced by any market. The A-share 8-07 15:00 close = 03:00 US ET, and the Hong Kong 16:08 close = 04:08 US ET; when the ruling was issued during US business hours both markets had already closed. 药明康德(603259)'s Voluntary Announcement (Interim 2026-033) was pushed out at 15:30 on 08-09. So today, 08-10, is the first tradable moment for this news — strictly speaking it is brand new.
  2. Discount level: but the discount it is meant to repair was never fully applied. The 1260H listing happened on 2026-06-08, and the next day (6/9) 药明康德(603259) fell only −2.62%, 凯莱英(002821) −3.23%, 康龙 −4.20%, 昭衍 −3.79%. The market never believed the list would destroy the fundamentals at the time. A discount that was never priced in cannot be released by removing it. This is the single most easily over-estimated link in the whole issue.
  3. Attribution level: last Friday's surge has a stronger independent explanation. 药明康德(603259) is +20.5% over the last 4 trading days (8/4 +10.00%, 8/5 +4.35%, 8/6 −3.25%, 8/7 +8.49%), which corresponds to the 8-03 interim report + guidance upgrade + employee share ownership plan, and at the sector level to a recovery in global healthcare investment/financing and outbound BD deals for innovative drugs. These are real drivers, but they also mean a good chunk of the good news has already been eaten, and the injunction is only the Nth reason in the queue.
• A falsifiable observation point that can be checked today: if what this news re-rates is really "the geopolitical risk premium on Chinese CXO," then Hong Kong shares (foreign-priced, directly exposed to geopolitical risk) should repair more than the A-shares. Measured on a share-price basis (A-share price ÷ H-share price × FX rate 0.8656 − 1, recomputed name by name): 药明康德(603259) trades at a 7.0% discount, 凯莱英(002821) at a 67.5% premium, 康龙化成(300759) at a 90.4% premium and 昭衍新药(603127) at a 113.8% premium — the pricing foreign investors give the three unlisted-on-1260H small caps is only about half of the A-share level. Today's relative A/H performance is a natural experiment on "substance vs sentiment." (This issue uses only the share-price-based premium and does not cite cross-market PE comparisons: the denominators of the two markets' PEs are inconsistent and do not reconcile with the share-price premium.)
• Risk of a one-day wonder at a high level: moderately high. CXO names all surged on Friday with elevated turnover (药石科技 (Pharmablock, 300725) turnover 23.62%, 博腾股份 (Porton Pharma, 300363) 14.77%).
• Board distribution: Shanghai/Shenzhen main board (药明康德(603259), 凯莱英(002821), 昭衍新药(603127), 九洲药业 (Apeloa Pharmaceutical, 603456)), ChiNext (康龙化成(300759), 泰格医药 (Tigermed, 300347), 博腾股份(300363), 药石科技(300725), ±20%), STAR Market (皓元医药 (Haoyuan Chemexpress, 688131), 美迪西 (Medicilon, 688202), ±20%) — all three boards are represented; note the different price limits.

⚠️ The tiering rule must be enforced strictly here: legally this catalyst reaches 药明康德(603259) and no one else. The 2026-06-08 updated 1260H list (expanded to 188 entities) has been checked: within the pharma/life-sciences scope the listed names include 药明康德(603259) and 诺禾致源 (Novogene, 688315), among others (report on the list update⁹). In publicly searchable lists and authoritative summaries, no listing was found for 凯莱英(002821), 康龙化成(300759), 昭衍新药(603127) or 泰格医药(300347) — this issue accordingly treats them as "not listed." Note this is "not found" rather than official item-by-item confirmation; readers making decisions on this basis should rely on the original US DoD list PDF. The list also contains BGI-affiliated entities; the exact correspondence of each entity has not been verified one by one in this issue and is not expanded on here.

Tier	Company	Nature	Basis
L1 substantive positive	药明康德(603259)	Legal status actually changed	The only listed party and the only plaintiff
L2 valuation repair	康龙化成(300759)	Not listed; no legal cost to remove	North America revenue 61.82%, the greatest narrative elasticity
L2 valuation repair	凯莱英(002821)	Same as above	Overseas revenue 73.77%; the North America figure alone is not disclosed
L3 weakly related	昭衍新药(603127)	Not listed; overseas revenue only 27.35%	Mainly domestic safety evaluation; the longest and thinnest transmission chain

Table scrolls horizontally

For the three companies not on the list, this news does not deliver "cost removal" but "a downward revision in investors' assessment of the probability that Chinese CXO gets hit by administrative lists" — valuation repair at the Bayesian level, not a substantive positive at the order level. Any account that writes it up as three companies gaining "improved order-winning ability" mistakes inference for fact. One more reverse-direction note (not covered by the mainstream narrative): 1260H's compliance bite comes in two tiers — the "no direct contracting with listed entities" tier already took effect on 2026-06-30; from 2027-06-30 it extends to "procuring goods or services produced by listed entities," at which point the bite widens considerably. And this injunction blocks only the single administrative act that is 1260H; it does not block the legislative route of the BIOSECURE Act. The two are independent risk sources.

② Precious metals — the cleanest expectation gap, but the level of the gold price has to be placed correctly first

⚠️ A positional fact that changes the whole section's conclusion: gold is not at a high right now; it is rebounding after a deep drawdown. Spot gold broke USD 5300 for the first time ever on 2026-01-28 and touched USD 5500 the next day for a record high; it is now USD 4341.91, still about 20% below the late-January peak, and only about +0.35% year to date. So the correct reading of this week's +7.44% is: a rebound after 6 months and roughly a 24% drawdown, not "gold making a new high." Most reports only cite the "+7.44% weekly gain" without this positional context, which makes it easy to read as acceleration at a high — same direction, but a completely different risk calibration. One self-checkable corroboration: the original report says "the largest weekly gain since the week of January 23" — that sentence itself shows there was a bigger move in late January, which is exactly where the top was.

• Why it is positive: the −23,000 payrolls print was released at 20:30 Beijing time on 8-07, after the 15:00 A-share close. COMEX gold closed Thursday at 4242.00 → Friday at 4340.70 (+2.33%) → Sunday electronic session at 4405.00. The three A-share gold names rose only 1.88%/3.94%/3.50% on Friday, far less than gold's +7.2% on the week, which shows Friday did not digest the payrolls print at all.
• How big is the unpriced portion: measured from the A-share Friday close, roughly another +3.8% of the gold move is not yet reflected (this is an estimate: daily data cannot be aligned precisely to the gold price at the A-share 15:00 mark). This is the cleanest expectation gap in this issue.
• Short-term or medium-term: medium term. Central-bank gold buying (Q2 net purchases 288.9 tonnes, +62% YoY) is structural demand, not an impulse.
• Path to earnings delivery: direct, and already delivered in the pre-announcements — 紫金矿业(601899) H1 pre-announced net profit attributable to parent of about RMB 39.1 billion (+68%), 中金黄金(600489) RMB 4.1–4.6 billion (+52% to +71%), 山金国际(000975) RMB 2.35–2.48 billion (+47% to +55%).
• Is it fully expected — on two horizons the answer is opposite:
  • 20-day horizon: broadly in step, not obviously front-run. Gold is +8.6% over 20 days while the three stocks are +26.80%/+28.32%/+31.55%, an implied beta of about 3.1–3.7×. Gold miners' beta to gold is usually 2–3×, so this is somewhat high but supported by H1 pre-announcements of +47% to +68%.
  • 60-day horizon: this is "resilience," not "chasing." Over the past 60 days gold is −7.6% while the three stocks are +6.07%/+0.31%/−2.41% — they did not fall through during gold's decline, and that in itself has consumed part of the upside elasticity.
  • The gap versus the January peak: all three gold names' 52-week highest closes were on 2026-01-29, the same day gold peaked. Zijin −18.4% (exactly in step with gold's −18.4%; zero gap), 中金黄金(600489) −36.3%, 山金国际(000975) −31.1%. China National Gold and Shanjin lag gold by 17.9pct and 12.7pct, but most of that gap is a fundamental discount rather than a sentiment discount (see the next point), so it cannot be read directly as upside.
• The item most deserving of caution: all three gold names saw Q2 single-quarter profit fall quarter on quarter. 紫金矿业(601899) −5.3%, 中金黄金(600489) −17.3%, 山金国际(000975) −26.8%. (Over the same period 中国稀土(000831) was −29.0% Q/Q, but it belongs to branch 7, rare earths, and is not counted here.) In Q2 the marginal direction of the fundamentals (down) diverged from the marginal direction of the share prices (up), and that divergence has to be absorbed by the average gold price in Q3.
• Risk of a one-day wonder at a high level: moderate. Precious-metal stocks track the gold price rather than being independently hyped, so the one-day-wonder probability is lower than for theme stocks.
• Board distribution: all Shanghai/Shenzhen main board (±10%).

③ AI computing / PCB / semiconductor equipment — the only branch where earnings have actually landed, but the "four blowouts" framing has to be taken apart

• Why it is positive: what was disclosed on the evening of Friday 08-07 were formal, realized interim reports. But it must be said at the same time that all four had already issued pre-announcements in early July, so the real incremental information is not "zero to one" but "where within the pre-announced range the formal number landed" (see each stock below). The semiconductor sector saw net inflows of RMB 10.229 billion on Friday (the largest in the market), and the components sector was +6.45% with 11 limit-ups.
• The four are entirely different in nature and cannot be framed together (the ex-non-recurring/attributable ratios have been checked name by name against the CNINFO originals):

• 盛美上海(688082) has to be spelled out separately: of its 42% nominal growth, RMB 473 million comes from fair-value changes on "equity investments in listed shares" (booked at the June 30 closing price, generating no cash and swinging both ways; the underlying names are not disclosed). Meanwhile the core business is under pressure — revenue only +13.87% (the lowest of the four), gross margin −3.87pct, FX losses of RMB 146 million, and Q1 net profit attributable to parent once at −57.66%. Recomputed on the ex-non-recurring basis, the PE goes from a headline 82.3 to 124.5. More importantly, two forward indicators are falling at the same time: "goods shipped" fell from RMB 1.551 billion to RMB 1.354 billion, and contract liabilities from RMB 737 million to RMB 704 million — that does not support "core business acceleration in the second half."
• The best forward visibility actually belongs to 东威科技(688700): contract liabilities RMB 694 million → RMB 1.114 billion (+60.61%), equal to 1.65× first-half revenue. Its 53.98% liability ratio is "fake leverage" — short-term and long-term borrowings are both zero, and 90% of liabilities are contract liabilities and accounts payable, i.e. the result of more orders.
• Is it fully expected: it varies. Components/PCB already surged last Friday (生益科技(600183) hit limit-up on RMB 10.73 billion of turnover), so the handoff is hard; 广合科技(001389) is only −3.12% over 20 days and −8.39% over 60 days, the combination of "hardest earnings + position not extended," and belongs to the "earnings disclosed but the share price has not followed" scan pool that last Friday's recap specifically asked for.
• Risk of a one-day wonder at a high level: high for the PCB leg (already up for several sessions), low for the equipment leg.
• Board distribution: STAR Market (寒武纪(688256), 盛美上海(688082), 东威科技(688700), ±20%), SZSE main board (广合科技(001389), 沪电股份(002463)), SSE main board (生益科技(600183)).

⚠️ One variable that cuts across three branches: RMB appreciation is genuinely eating into exporters' profits. USD/CNY is about 6.765–6.77. FX losses explicitly recorded in interim/quarterly reports: 广合科技(001389) RMB 119 million (finance costs swinging from −RMB 5.6 million to +RMB 118.7 million), 盛美上海(688082) RMB 146 million, 凯莱英(002821) a RMB 123 million swing in Q1 finance costs (from −RMB 45 million to +RMB 78 million, wiping out all of its profit growth). The value of this thread is that it changes how the same report should be read: 广合科技(001389) delivered +94% after absorbing RMB 119 million of FX losses, whereas 盛美上海(688082) used mark-to-market equity gains to cover RMB 146 million of FX losses. Both are "earnings growth," with opposite meanings. It also weakens the precious-metals branch: a +7.44% move in the USD gold price is discounted by the exchange rate when it flows through to RMB revenue — the pass-through is not 1:1.

④ Memory — the branch where this issue must get the direction right

The surface evidence for this branch would be enough to support a first-ranked branch, yet this issue ranks it 4th. The reasoning has to be laid out:

• All three bullish pieces of evidence are real: Apple testing CXMT memory chips (WSJ 08-09, primary foreign media); July CPI computers +17.4%, tablets +17.2%, with the National Bureau of Statistics explicitly attributing it to "artificial intelligence driving the iteration and upgrading of consumer electronics" (official statistics, extremely high evidence grade); and the semiconductor sector's RMB 10.229 billion net inflow on Friday, the largest in the market.
• Overseas pricing is decelerating but has not turned down — this must be precise: on 08-07 Citi cut its Micron price target from USD 1400 to 1150, but maintained a "Buy" rating, forecasting Q/Q DRAM +23% / +9% / +2% / 0% and NAND +29% / +7% / 0% / −1% over the next four quarters, with prices peaking in 2027Q2 and only turning to DRAM −3% / NAND −5% in 2027H2. In other words: DRAM still rises 23% Q/Q next quarter; the price-increase cycle has not ended, what has ended is the acceleration. "Q/Q gains narrowing to 0" and "prices falling" are two different things, and conflating them inverts the conclusion.
• The reason Citi gives for the deceleration is precisely this branch's core logic: it explicitly attributes the weakening momentum in DRAM/NAND prices to capacity expansion at CXMT and YMTC. That amounts to sell-side research confirming, in reverse, that Chinese memory makers are materially taking global share — the same coin as "Apple testing CXMT."
• The nature of the US memory-stock slump also has to be stated clearly: at the 08-06 close SanDisk was −14%, Western Digital −19%, SK Hynix −6% and Micron −4%, mainly because SanDisk's next-quarter guidance missed (the results themselves beat) — a valuation de-rating of "expectations that ran too far ahead," not deteriorating fundamentals, and certainly not a cycle top.
• How the two timelines coexist: CPI computers +17.4% is July data reflecting shortage-driven increases that have already happened; Citi's numbers are a forward-looking judgment. The two point the same way (both rising); the latter merely says the gains will narrow quarter by quarter. There is no contradiction here.
• So how should the A-share memory names' −25% to −49% over 20 days be read? All four peaked on the same day, 2026-06-29; from the high: 佰维存储(688525) −51.7%, 江波龙(301308) −46.5%, 兆易创新(603986) −50.4%, 德明利(001309) −58.3%. This decline follows the overseas "valuation de-rating + price deceleration" expectation, not a turn down in prices. So it is neither pure mispricing nor a trend collapse — what really determines each stock's direction is its own leverage and profit quality, not the industry's price direction (see the module makers' idiosyncratic risks below).
• The part of the logic that genuinely holds is the "share shift": Apple testing CXMT is at bottom Chinese memory makers taking share from US and Korean players. That holds for domestic substitution; it does not hold for the memory price cycle — a share shift actually pushes global prices down.

⚠️ A piece of common sense that must be corrected: 长鑫科技(688825) is already listed. 长鑫科技(688825, STAR Market ±20%) listed on the STAR Market on 2026-07-27 at an offer price of RMB 8.66 and closed at RMB 52.48 on 08-07 (+506% versus the offer price), with a total market cap of RMB 3.51 trillion, a free-float market cap of only RMB 236.3 billion (about 6.7% of shares), a PE of 124.48 (note: this uses FY2025 net profit attributable to parent as the denominator and has already been overwritten by the interim report — RMB 3.51 trillion ÷ 124.48 backs out TTM attributable profit of only about RMB 28.2 billion, less than the RMB 50–57 billion of H1 alone, which is logically impossible; annualizing the realized H1 figures puts the forward PE in the 30× range), and Friday turnover of 11.45%. Its H1 2026 pre-announcement is revenue of RMB 110–120 billion (+612.53% to +677.31%) and net profit attributable to parent of RMB 50–57 billion (+2244% to +2544%). This means the direct beneficiary of the "Apple testing CXMT" news is 长鑫科技(688825) itself, not any proxy name. The market still widely uses the now-expired premise that "CXMT is unlisted, so you can only buy proxies" (that premise lapsed after 2026-07-27), and buying 兆易创新(603986) and the like as CXMT concept plays on that basis is aiming at the wrong target. But note its price pattern: across its first 10 trading days it has ranged between RMB 47 and 55 (Friday +1.00%, −2.76% over 5 days, net main-force inflow −RMB 7 million), and turnover value has decayed from RMB 141.2 billion on day one to about RMB 27 billion on Friday. The news is landing on a consolidation platform, not mid-rally — cleaner than the proxies in that respect. The risk is equally plain: a RMB 3.5 trillion market cap, a 6.7% free float and 11.45% daily turnover mean moves in either direction get amplified (see the PE basis note below).

• The relationship between the A-share memory chain and CXMT must be verified name by name, not assumed. Verified results:
  • 兆易创新(603986) is the only one of the four with a publicly disclosed direct link — it holds about 1.80% of CXMT and CXMT is its DRAM foundry. But the company voluntarily clarified on 2026-05-26 that the stake is booked under "other equity instrument investments" and that fair-value changes from CXMT's IPO do not affect the company's net profit. More importantly there is a reverse channel: CXMT focuses on the mainstream bulk markets of servers/PCs/handsets while GigaDevice focuses on niche markets such as set-top boxes and base stations, so if the Apple orders land, GigaDevice's niche DRAM foundry capacity allocation risks being squeezed out. The net impact of "Apple testing CXMT" on GigaDevice is directionally uncertain and could even be negative — the opposite of the market's instinct to treat it as a "CXMT proxy."
  • The substantive procurement relationships between 佰维存储(688525), 江波龙(301308), 德明利(001309) and CXMT: no reliable data available. CXMT's prospectus does not disclose the names of its top five customers, and all three companies' own interim pre-announcements use generic phrasing such as "major global memory wafer foundries" and "mainstream wafer manufacturers" — not one of them names CXMT. Their link to this news is purely at the sentiment level, and there is also a reverse channel in which "once CXMT's capacity is locked up by mega-customers like Apple, the domestic wafer allocation left for small and mid-sized module makers actually shrinks."

⑤ Humanoid robots — the event is certain, but note that it is not "new weekend news"

Unitree's pricing took place on 08-06; today is only the subscription day. This is a known scheduled event that the market has had 4 days to digest, and it should not be priced as a surprise positive. The issue PE of 219.23× compares with an industry 38.56×, and the raise of RMB 6.099 billion exceeds the original RMB 4.202 billion plan by about RMB 1.897 billion. The strategic placement includes the National Social Security Fund, DeepSeek and the 中国石油(601857) group. Subscription days are usually two-way for concept stocks: attention rises, but liquidity is drained and "good news realized" is common. This issue puts no humanoid-robot concept stock on its recommended list, because the overwhelming majority of so-called concept names have no publicly verifiable supply relationship with Unitree — precisely the "concept ≠ beneficiary" of iron rule 2.

⑥ Oil / tanker shipping — against expectations, but the direction reverses very fast

The Hormuz crisis has run since 2026-02-28, and Brent once broke USD 100 on March 8; it is now USD 84.43 — showing the market has been pricing an "eventual reopening." Last Friday most A-share oil & gas names fell over 5 days (中国石油(601857) −2.80%, 中海油服(601808) −4.35%), which corroborates this. The weekend change runs against expectations: Iran installed a hardliner, approved the security framework, and the IRGC stressed "maintaining control until the enemy accepts all conditions," with WTI +1.3% in Monday's early session. That is a reason for a short-term bounce, not for a trend reversal — Omani mediation is still under way and "making progress," and any breakthrough headline flips the direction instantly. Hence a B grade, and it does not enter the top three recommendations.

⑦ Rare earths / minor metals — dual catalysts from earnings and supply, but the fund-flow bases conflict

中国稀土(000831) posted interim net profit of RMB 237 million, +46.53%, and hit limit-up on Friday; the minor-metals sector was +2.85%. The suspension of the Mengkang rare-earth project in Laos by an associate of 厦门钨业(600549) constitutes a supply-side disruption. The flow-of-funds question has been adjudicated as genuine inflow using exchange data: Friday's Dragon-Tiger List shows net buying of +RMB 632 million, the largest single net buy of the day (listing reason: one of the top 5 securities by deviation of daily price move reaching 7%), consistent in sign with Tencent's +RMB 1.004 billion. But valuation is the hard problem: PE 244.9×, and this is a limit-up stock that received large buying on the day its good news was realized, so next-day absorption pressure is naturally elevated.

---

3. Master Ranking of Single-Stock Positive Strength (descending by stock, 22 names)

Quotes are as of the 2026-08-07 close; PE is the Tencent snapshot PE (TTM), and wherever that basis has been overwritten by the interim report, the ex-non-recurring / interim basis is listed separately. Net main-force inflow is a single-day figure (two consecutive days could not be verified; see the scope disclosure).

---

4. Single-Stock Scoring Model (out of 100)

Four deduction rules specific to this issue (aimed at the problems exposed by last Friday's recap):

1. Where ex-non-recurring/attributable < 0.8, earnings elasticity is re-scored entirely on the ex-non-recurring basis (this is why 盛美上海(688082) fell from A to B).
2. For cyclicals whose PE and PB historical percentiles diverge systematically (very low PE but elevated PB), no "undervalued" credit is given — points are deducted instead, because this is the signature of peaking earnings (all four memory names qualify).
3. Single-day net main-force inflow earns no points, because two consecutive days cannot be verified; when Dragon-Tiger institutional seats and net main-force inflow point in opposite directions, the institutional seats take precedence (this is why 光迅科技(002281) is Passed).
4. For names "not on the list / with no publicly disclosed supply relationship," directness is capped at 12 points, no matter how large the price move.

---

5. Detailed Analysis of the Top 10

1药明康德603259Score 81, priority deep-dive · SSE main board ±10%

• Related news: on 2026-08-07 US time, Chief Judge Boasberg of the US District Court for the District of Columbia granted a preliminary injunction, finding that the company "has shown the designation is likely arbitrary and capricious," with security of only USD 1.00 (to be posted by 08-14). The court rejected the DoD's three rationales one by one: the shareholding ratio mistook a 5.32% share of a fund's net assets for an equity stake (the actual holding is 0.001%); it misread the company's status as a third-party laboratory as a university partnership; and the applicant in the military-hospital collaboration was in fact Novo Nordisk. The company's Voluntary Announcement, Interim 2026-033, was pushed out at 15:30 on 08-09. Source¹
• Positive logic: direct, and it affects revenue rather than merely sentiment. A 1260H designation materially raises overseas clients' compliance costs and their reluctance to work with the company. The injunction suspends that constraint for the duration of the litigation.
• Stage of the industry branch: late in the fermentation phase. CXO names already saw a broad limit-up wave last Friday (凯莱英(002821) and 昭衍 at limit-up, 博腾 +20.02%, 药石 +18.39%), with both a core name (药明康德(603259)) and a ladder behind it — but most of the ladder consists of companies that are not on the list.
• Fundamental verification (2026 interim report, disclosed 08-03): revenue RMB 28.897 billion +38.94%; net profit attributable to parent RMB 11.080 billion +29.43%; ex-non-recurring RMB 10.572 billion +89.39%; gross margin 53.91% (+9.46pct); operating cash flow RMB 9.699 billion (OCF/attrib. 0.88); liability ratio 27.85%. Q2 single-quarter revenue was RMB 16.461 billion +47.7% with ex-non-recurring +93.5%, a clear acceleration from Q1.
  • A structure that is easily read backwards: attributable growth (+29.4%) is far below ex-non-recurring (+89.4%) because 2025H1 contained RMB 3.669 billion of investment income (disposal of the WuXi XDC stake) with no equivalent in 2026H1. On the surface "profit growth is mediocre"; in reality the core business nearly doubled.
  • Backlog of RMB 66.43 billion, +25.2% YoY (primary disclosure). Back-calculation check: RMB 66.43 billion ÷ the RMB 59.5 billion midpoint of full-year guidance = 1.12× annualized revenue, a reasonable range for CRDMO.
  • Net cash of about RMB 38.6 billion (8.4% of market cap) and goodwill of only RMB 863 million. Construction in progress is RMB 8.247 billion (+40.35%), directed at the Delaware (US) and Singapore sites — a structural hedge against geopolitical risk.
• Industry position: the global CRDMO leader, the only listed party and the only plaintiff.
• Technicals and sentiment: SSE main board ±10%. Friday +8.49%, 5-day +20.55%, 20-day +25.56%, 60-day +52.64%, only 0.4% below the high for the year. Turnover 3.75% (not high, so not a sentiment-driven crowd). The A-share at RMB 154.82 versus the H-share at HKD 192.30 (≈RMB 166.5) means the A-share trades at about a 7% discount — the only one of the four CXO names where the A discounts to the H.
• Final judgment: priority deep-dive. The highest logical hardness in the market, the most solid fundamentals in the market, the lowest valuation (PE 21.3, or 25.3 ex-non-recurring), and the only one where the A discounts to the H. But set expectations at "the news is real, the room is limited" — the discount it has to repair was only ever marked down 2.62%. The point to watch today is not whether it rises but whether it can hold a new high for the year on volume.

2广合科技001389Score 74, priority deep-dive · SZSE main board ±10%

• Related news: the evening of Friday 08-07, interim report (announcement 2026-063): revenue RMB 4.388 billion +80.98%, net profit attributable to parent RMB 956 million +94.39%. [CNINFO original verified]
  • This issue at one point suspected that RMB 956 million was "revenue miscopied as net profit," and ruled it out along four paths: ① the announcement lists net profit and revenue on separate lines (RMB 955,588,247.43 vs RMB 4,388,260,234.74); ② a 21.8% net margin is structurally consistent with a 39.27% gross margin; ③ EPS of RMB 2.14 × share count backs out correctly; ④ the PE back-calculation is the hardest of all — RMB 80.266 billion ÷ TTM attributable profit of RMB 1.480 billion = 54.2, matching the snapshot exactly; had net profit been only RMB 95.6 million, the PE would be 129.
• Positive logic: direct, affecting revenue. Server PCB is about 90% of revenue, covering 8 of the world's top 10 server manufacturers.
• Stage of the industry branch: fermenting. The components sector was +6.45% on Friday with 11 limit-ups, but Guanghe itself is −3.12% over 20 days and −8.39% over 60 days; it has not kept up with the sector.
• Fundamental verification: ex-non-recurring RMB 939 million +96.71%, ex-non-rec./attrib. 0.98; operating cash flow RMB 937 million, OCF/attrib. 0.98 — profit and cash move together, the highest quality among the four AI hardware names.
  • The revenue mix corroborates the AI export logic: overseas RMB 3.171 billion (72.26% of operating revenue, +97.26%) at a 41.62% gross margin; domestic RMB 916 million at a gross margin of only 13.70%. (The two together are RMB 4.087 billion; the roughly RMB 301 million difference from revenue of RMB 4.388 billion is other business; the percentages use operating revenue of RMB 4.388 billion as the denominator.) Essentially all the profit comes from overseas computing-power customers.
  • A key detail: finance costs swung from −RMB 5.6 million to +RMB 118.7 million, of which net FX losses were RMB 119.3 million. It produced +94% after absorbing that RMB 124 million adverse swing.
  • Interest-bearing debt is only RMB 646 million against cash of RMB 2.642 billion — a net cash position. Construction in progress went from RMB 255 million to RMB 845 million (Guangzhou plus Thailand expansion), and the Thai plant is 100% focused on high-end computing-server motherboard products.
• Industry position: leader of the server PCB niche (not of the whole industry).
• Technicals and sentiment: SZSE main board ±10%. Friday +5.62%, 5-day +19.73%, turnover 7.06%, main-force +RMB 92 million. Dual-listed A+H; the H-share 01989 closed Friday at HKD 125.80 (+10.93%), so the A-share trades at about a 56% premium to the H-share.
• Risks: also disclosed on 08-07 was a RMB 3.6 billion convertible bond plan (revised draft) (Yunqing Phase II RMB 1.85 billion + high-layer-count boards RMB 1 billion + working capital RMB 750 million), the second financing round after roughly RMB 2.85 billion raised via H-shares in March, carrying conversion dilution and front-loaded heavy-asset depreciation risk; the A-share free float is only 32% of total shares and the lock-up expiry structure needs separate verification. If the RMB keeps appreciating, FX losses and price competitiveness both come under pressure.
• Final judgment: priority deep-dive. "Hardest earnings + lowest position + lowest PE" hold simultaneously, making it the name that best fits the "earnings disclosed but the share price has not followed" scan pool required by last Friday's recap. But the RMB 3.6 billion convertible bond is a demerit to watch from today.

3紫金矿业601899Score 72, priority deep-dive · SSE main board ±10%

• Related news: US July payrolls −23,000 (20:30 Beijing time on 8-07, after the A-share close) → spot gold +7.44% on the week to USD 4341.91; Q2 global central-bank net gold purchases 288.9 tonnes, +62%.
• Positive logic: direct, affecting revenue. A higher gold price directly expands mined-gold gross profit.
• Stage of the industry branch: mid-rebound. Note that gold itself is still about 20% below the late-January high; this is not acceleration at a high.
• Fundamental verification: Q1 revenue RMB 98.498 billion +24.8%, net profit attributable to parent RMB 20.079 billion +97.5%, gross margin 36.33% (+13.4pct), OCF/attrib. 1.39×, liability ratio 51.37% (falling steadily). H1 pre-announced attributable profit of about RMB 39.1 billion, +68%. 2025 ROE 33.04% (a cyclical peak ROE in a record-high gold-price year, not a steady-state level).
  • Is it a gold name or a copper name: the 2026Q1 gross-profit mix — mined gold RMB 17.318 billion, copper in total RMB 12.670 billion (54.4% and 39.8% respectively of mining-segment gross profit after excluding "other/eliminations"; the denominator is mining-basis gross profit, not company-wide gross profit). Gold has clearly surpassed copper for the first time. Equal-magnitude elasticity: gold +10% → attributable profit +RMB 6.02 billion (8.9% of TTM); copper +10% → +RMB 5.32 billion (7.9%). Conclusion: a genuine "gold-copper twin engine," with lithium as a third leg from 2026 (lithium carbonate equivalent planned to go from 25,500 tonnes to 120,000 tonnes).
  • 2025 mined gold was 89.58 tonnes (+22.81%), with 105 tonnes planned for 2026; Q1 weighted all-in cash cost was about RMB 275.9 per gram (gold concentrate gross margin 80.89%).
  • Every +USD 100/oz in the gold price ≈ +RMB 21.69 per gram (at USD/CNY 6.7472) → on 105 tonnes, after-tax attributable profit rises by about RMB 1.385 billion, or 2.0% of TTM attributable profit. (A static estimate assuming rigid costs and sales = output.)
• Industry position: a global mining leader and the core resource name in A-shares.
• Technicals and sentiment: Friday only +1.88% on turnover of 1.33% (the lowest in the whole list, no sentiment crowd); 20-day +26.80%, 60-day +6.07%; −18.4% from the January 29 peak, exactly in step with gold's −18.4%, so the catch-up gap is zero.
• Risks: Kamoa-Kakula Q1 attributable copper output was 27,361 tonnes, −53.8% YoY, casting doubt on delivery of the 1.2 million tonne copper plan for 2026 — copper's unit price elasticity is 2.6× gold's, and copper volume matters more than gold price for full-year profit; the 51.37% liability ratio is the highest of the four, with assets spread across the DRC, Serbia, South America and elsewhere; on 07-30 it announced the termination of the Allied Gold acquisition, switching to subscribing a private placement for a 9.2% stake (the original acquisition did not complete).
• Final judgment: priority deep-dive. The lowest PE in the whole list (15.2) + the most solid cash flow + the lowest turnover + almost no move on Friday (the payrolls print undigested). Its problem is not that it has run too far; it is copper output.

4东威科技688700Score 66, watch closely · STAR Market ±20%

• Related news: interim report on the evening of 08-07 — revenue RMB 674 million +51.94%, net profit attributable to parent RMB 99 million +133.21%, proposed dividend of RMB 1.70 per 10 shares (RMB 50.7068 million in total, 51.16% of half-year attributable profit).
• Positive logic: indirect but with a clear chain — AI server PCB capacity expansion → PCB makers' capex → electroplating equipment orders. "Selling shovels to the shovel sellers."
• Fundamental verification: ex-non-recurring RMB 95.8 million +133.93%, ex-non-rec./attrib. 0.97; operating cash flow RMB 311 million (+924.83%), OCF/attrib. 3.14×, the best in the whole list.
  • The best forward visibility in the list: contract liabilities RMB 694 million → RMB 1.114 billion (+60.61%), equal to 1.65× first-half revenue, which the company explains as "prepayments rising with order growth."
  • The 53.98% liability ratio is "fake leverage": short-term and long-term borrowings are both zero; of RMB 2.274 billion of total liabilities, contract liabilities of RMB 1.114 billion + accounts payable of RMB 669 million + notes payable of RMB 268 million = RMB 2.051 billion, or 90%, all operating liabilities. Looking only at "a 54% liability ratio and a 0.72 quick ratio" would produce exactly the opposite conclusion.
  • The overseas gross margin of 47.33% is well above the domestic 33.64%.
• Industry position: over 50% domestic share in VCP (vertical continuous plating) equipment, and the only pure precision electroplating equipment name on the STAR Market — a niche leader.
• Technicals and sentiment: STAR Market ±20%. Friday +8.22%, 5-day +22.37% (the strongest pre-market week of the four AI hardware names), 20-day −13.06%, turnover 6.85%.
• Risks: market cap of only RMB 18.7 billion on a TTM profit base of RMB 178 million, so the timing of a single project acceptance can rewrite a quarter; receivables turnover 192 days, inventory turnover 537 days; new-energy electroplating (composite copper foil) is still at the "customer validation" stage with no disclosed revenue scale (no reliable data), so earnings stand on the single leg of PCB electroplating.
• Final judgment: watch closely. Contract liabilities at 1.65× revenue are the best forward indicator this issue could find. But a PE of 105.5 has already priced in a fair amount, and the stock is up 22% in the pre-market week.

5寒武纪688256Score 65, watch closely · STAR Market ±20%

• Related news: interim report on the evening of 08-07 — revenue RMB 5.996 billion +108.13%, net profit attributable to parent RMB 2.311 billion +122.61%.
• Fundamental verification: ex-non-recurring RMB 2.166 billion +137.30%, ex-non-rec./attrib. 0.94 (non-recurring items are only RMB 145 million, or 6.3%). The cloud product line at RMB 5.994 billion = 99.98% of revenue.
  • Profit outran revenue because of operating leverage: revenue doubled while selling expenses actually fell (RMB 28 million → RMB 26 million) and administrative expenses fell (RMB 98 million → RMB 83 million), with R&D up only 30%, so the operating expense ratio dropped from 23.5% to 13.6%.
  • The profit also absorbed RMB 397 million of asset impairments (of which RMB 396 million was raw-material write-downs) — stripping these out, pre-tax profit could have been roughly RMB 400 million higher, which is "conservative" but also a risk signal.
  • Why is cash flow only RMB 311 million (OCF/attrib. 0.13): it is not a collection problem — cash received from sales of goods was RMB 7.452 billion, 1.24× revenue, and receivables actually fell from RMB 671 million to RMB 222 million. The cash is tied up in inventory of RMB 9.015 billion (raw materials RMB 5.749 billion + outsourced processing RMB 2.537 billion, versus only RMB 5.333 billion at the start of the year) and prepayments of RMB 745 million → RMB 2.914 billion, i.e. large-scale wafer stockpiling and prepaid capacity. This is an offensive move, and success depends on whether that material can be sold at the expected price (the RMB 396 million write-down shows some of it already cannot).
  • A positive signal: contract liabilities went from RMB 0.6 million to RMB 188 million — customers are starting to prepay.
• Technicals and sentiment: Friday +2.72%, turnover only 2.54%, 20-day −14.29% (good earnings but the share price has not followed).
• Risks: PE 226.3 (249.4 ex-non-recurring), PB 55.62; 95.08% of receivables and contract assets come from the top five customers; the effective tax rate is near zero (deferred tax / loss carryforwards), and normalization will directly compress net profit.
• Final judgment: watch closely. (Valuation sits in the most expensive band in the list, so its observational value exceeds its allocation value.) Earnings quality ranks second of the four (behind Guanghe only), but valuation is among the most expensive in the list. It suits use as an indicator of branch strength, not as a value-for-money holding.

6中金黄金600489Score 63, watch closely · SSE main board ±10%

• Related news: same as Zijin. H1 pre-announced attributable profit of RMB 4.1–4.6 billion (+52.15% to +70.7%).
• Fundamental verification: Q1 revenue RMB 22.588 billion +52.0%, attributable profit RMB 2.381 billion +129.4%, gross margin 20.47%. Q1 operating cash flow of −RMB 2.168 billion is seasonal (2025 full-year OCF/attrib. was 1.70×, with Q4 alone contributing about RMB 6.9 billion).
  • The greatest unit gold-price elasticity of the four: every +USD 100/oz → attributable profit of about +RMB 245 million = 3.7% of TTM (Zijin only 2.0%, because its profit is diluted by copper and lithium).
  • ⚠️ The place most easily miscalculated by a factor of 3: 2025 total gold output was 56.79 tonnes, but only mined gold of 18.40 tonnes (32%) enjoys price elasticity; the 38.39 tonnes of smelted gold is a tolling model on purchased feed, and the smelting segment's gross margin is only 3.91%, so a higher gold price is nearly neutral for it. Computing elasticity on 56.79 tonnes overstates it by roughly 3×.
  • Cash cost per gram: not disclosed by the company (no reliable data) — the 2025 annual report (201 pages) and the 2026 Q1 report have been searched page by page, and the only "RMB per gram" figure in them is the SHFE gold contract price. This issue makes no estimate.
• Technicals and sentiment: Friday +3.94%, 20-day +28.32%, turnover 2.34%; −36.3% from the January 29 peak, the largest gap of the three gold names.
• Risks: no volume growth — it is a pure price name: mined gold rose only 0.05 tonnes in 2025 with a further 0.32 tonnes planned for 2026, and mine copper was −19.25% in 2025; Q2 attributable profit −17.3% Q/Q is already a preview. Related-party concentration is extremely high: the top five customers are 91.53% of total sales, of which related parties are 60.66%. In 2025 ex-non-recurring profit (RMB 5.772 billion) exceeded attributable profit (RMB 4.934 billion), a RMB 838 million gap of net non-recurring losses whose breakdown is still to be checked.
• Final judgment: watch closely. The largest elasticity and the largest gap, but the elasticity is "brittle" (no volume growth and 68% of output does not enjoy the price). Of the three gold names it is the best for a short-term gold beta and the worst for a medium-term hold.

7长鑫科技688825Score 63, watch closely · STAR Market ±20%

• Related news: The Wall Street Journal reported on 08-09 that Apple is testing CXMT memory chips in its iPhone and MacBook lines to ease the AI memory shortage; HP and Acer already use them in devices for non-US markets.
• Positive logic: the most direct possible — it is the company being tested.
• Fundamental verification: listed 2026-07-27 at an offer price of RMB 8.66, raising RMB 57.919 billion at an offer-based market cap of RMB 579.188 billion — surpassing SMIC's RMB 53.2 billion as the largest IPO in STAR Market history (if the over-allotment option is fully exercised, total proceeds reach RMB 66.607 billion); the issue PE was 308.92×. H1 pre-announced revenue of RMB 110–120 billion (+612.53% to +677.31%) and attributable profit of RMB 50–57 billion (+2244% to +2544%). DDR4/DDR5/LPDDR4X/LPDDR5/5X are all in mass production. The prospectus discloses that the top five customers total 59.99% to 74.12%, but does not disclose their names; end customers are listed as Alibaba Cloud, ByteDance, Tencent, Lenovo, Xiaomi, Transsion, Honor, OPPO and vivo.
• Technicals and sentiment: it has ranged between RMB 47 and 55 throughout its first 10 trading days (day-one close RMB 49.00, Friday close RMB 52.48), with turnover value decaying from RMB 141.2 billion on day one to about RMB 27 billion on Friday. Friday +1.00%, 5-day −2.76%, main-force −RMB 7 million, turnover 11.45%. The news lands on a consolidation platform, not mid-rally — cleaner than any proxy name.
• Risks: a total market cap of RMB 3.51 trillion, a free-float market cap of only RMB 236.3 billion (about 6.7%), and 11.45% daily turnover — the three together mean moves in either direction are extremely amplified. Be careful with the valuation basis: the snapshot PE of 124.5 uses the FY2025 denominator, while the forward PE on annualized H1 is in the 30× range; the two differ by an order of magnitude, so do not use 124.5 to argue "expensive"; only 10 trading days since listing, so there is no historical trading structure to reference; at the industry level this is a "price-increase deceleration" phase, not a "price decline" phase (Citi forecasts Q/Q DRAM +23%/+9%/+2%/0% over the next four quarters, turning down only after the 2027Q2 peak).
• Final judgment: watch closely. It is the only name for which the "Apple testing CXMT" story holds both legally and commercially, and its pattern is platform consolidation rather than a high. But the combination of a RMB 3.5 trillion market cap and a 6.7% free float makes its volatility unsuitable as a lead recommendation, hence watch closely rather than priority deep-dive.

8山金国际000975Score 61, watch closely · SZSE main board ±10%

• Fundamental verification: Q1 revenue RMB 5.938 billion +37.41%, attributable profit RMB 1.394 billion +100.89%, gross margin 40.67%, OCF/attrib. 1.79× (the best of the four), liability ratio 22.64%, minority interests only 9.1% (the lowest of the four, so the least profit leakage). H1 pre-announced attributable profit of RMB 2.35–2.48 billion (+47% to +55%).
  • 2025 mined gold was 7.60 tonnes (−5.47%) and 2026Q1 1.65 tonnes (−6.8%); the post-consolidation-amortization cash cost went from RMB 142.18 to RMB 165.16 per gram (+16.2%), with a gross margin of 84.56%.
  • ⚠️ A measurement trap that has to be called out: in 2025 revenue, metals commodity trading was RMB 9.979 billion = 58.36% of revenue but contributed only RMB 138 million of gross profit = 2.5% (copper trading gross margin 0.63%, tin ingots 0.38%), while mining and processing was 41.6% of revenue yet contributed 97.4% of gross profit. Its revenue growth cannot be used to judge earnings quality.
• Technicals and sentiment: Friday +3.50%, 20-day +31.55% (the strongest in precious metals), turnover 2.47%, −31.1% from the January peak.
• Risks: a two-way squeeze of falling output and rising cash cost, leaving earnings standing on the single leg of the gold price; Q2 attributable profit −26.8% Q/Q, the worst of the four; full-year 2026 output guidance could not be found in either the annual or the quarterly report (no reliable data).
• Final judgment: watch closely. Its statement quality (cash flow, leverage, minority-interest share) is the best of the three gold names, but volume is falling and costs are rising — a substantial part of its discount is a fundamental discount, not a pure sentiment discount.

9沪电股份002463Score 59, watch closely · SZSE main board ±10%

• Related news: no new primary catalyst in this window; it is a continuation of last Friday's main line. The components sector was +6.45%.
• Rationale: last Friday's recap already verified that it is the only confirmed sample under the "two consecutive days of net main-force inflow" rule (8/6 +RMB 160 million → 8/7 +RMB 957 million, two consecutive days and sharply larger). Friday +7.40%, 20-day −2.76%, PE 56.3.
• Final judgment: watch closely. No new catalyst, but the healthiest pattern and fund flow. This issue places it here because it is the only name that satisfies last Friday's new rule — a rule this issue cannot verify for any other stock.

10生益科技600183Score 58, watch closely · SSE main board ±10%

• Related news: no new primary catalyst in this window. The CCL leader, limit-up on Friday with turnover value of RMB 10.73 billion, net main-force inflow of +RMB 1.930 billion, 5-day +32.51%, 20-day −6.30%, PE 86.5, turnover 3.25%.
• Final judgment: watch closely. (Already locked at limit-up, so next-day absorption is an observation point rather than an entry point.) It is the core name of the components branch, with more observational than purchase value — whether it can hold the board on volume today is the single best signal for whether the PCB/CCL main line continues.

---

6. Pass List

⚠️ The semantics of this list have to be declared first (this is the lesson bought with 19 names and 10 wrong calls in last Friday's recap): "Pass" = this issue does not recommend buying; it does not mean predicting a decline. In last Friday's pre-market Pass list, 10 of 19 names were called wrong and 5 of them hit limit-up — but on verification in the recap, not one of the fundamental disqualifications for those companies was overturned; what changed was the environment: turnover across the two markets was RMB 2.66 trillion that day, 2,856 names rose, and the indices all gained more than 1%; on a broad-rally day every extended name found absorption. "This stock's fundamentals do not stack up" was right; "therefore it will fall tomorrow" was wrong. This issue therefore states the environmental conditions explicitly:

• If the indices rally broadly today and there are more than 70 limit-ups, this list may rise along with everything else, in which case please judge on "did it underperform its sector" rather than "did it fall."
• Only if today is a differentiated day (fewer than 50 limit-ups or the indices close down) should absolute price moves be the test. Without this condition, this list will be systematically scored wrong on every broad-rally day.

---

7. Intra-Branch Ranking

7.1 Innovative Drugs / CXO

• The hardest single stock: 药明康德(603259) — the only one whose legal status genuinely changed, the only A-to-H discount, and the most solid fundamentals.
• Highest trading recognizability: still 药明康德(603259) (a clear core name, turnover not high, room for volume expansion).
• Who is chasing: 康龙化成(300759) and 昭衍新药(603127) — the largest gains with the weakest causal link to the catalyst.
• Who is Passed: 昭衍新药(603127) (loss-making core business + unrelated to the catalyst).

7.2 AI Computing / PCB / Semiconductor Equipment

• The hardest single stock: 广合科技(001389) (best on earnings, cash flow, valuation and position simultaneously).
• Highest trading recognizability: 生益科技(600183) (clearest core-name status and turnover value), but the highest evidence density and margin of safety belongs to 广合科技(001389).
• Who is Passed: 中京电子(002579) (turnover 31.09%, PE 353.9).

7.3 Precious Metals

• The hardest single stock: 紫金矿业(601899). But remember it is no longer a pure gold name — copper volume matters more than the gold price for its full-year profit.
• Largest pure gold-price beta: 中金黄金(600489) (3.7%), but the elasticity is "brittle" (68% of output does not enjoy the price).
• Who is Passed: no Pass in this branch, but all four names saw Q2 single-quarter profit fall quarter on quarter, so the branch as a whole gets "watch closely" rather than "priority deep-dive" (Zijin excepted).

---

8. Opening Verification Signals for Today

8.1 Call-Auction Signals

• 药明康德(603259): the single most important observation point. If it gaps up 3%–5% with expanding auction volume, the market is treating the injunction as a substantive positive; if it gaps up more than 7% or locks limit-up, be wary of "good news realized" — it is only 0.4% below the high for the year, so a gap-up is an immediate new high and absorption pressure is at its greatest. Whether it can hold on volume after the gap matters more than the size of the gap.
• The three precious-metal names: 紫金矿业(601899) / 中金黄金(600489) / 山金国际(000975). The reference benchmark is "roughly +3.8% of unpriced gold move." A 2%–4% gap up in the auction is a reasonable reflection; a gap of less than 1% would say funds do not accept this expectation gap and the branch's strength needs re-rating; a gap above 6% would already be front-run.
• 长鑫科技(688825): STAR Market ±20%. It is the sole arbiter of the "Apple testing CXMT" story. If it gaps up weakly or opens down, the whole memory chain's sentiment mapping fails, and any follow-through in 佰维 / 江波龙(301308) / 德明利(001309) will be a feint.
• 广合科技(001389): a gap up within 5% on volume is the pattern this issue would most like to see (earnings delivery rather than a sentiment push); note that the RMB 3.6 billion convertible bond plan may be read as dilution; if that causes it to open lower, that is an opportunity to observe.

8.2 Sector Signals

• Whether 3 or more CXO names hit limit-up quickly, and whether the core name 药明康德(603259) leads on volume (rather than being led by unlisted-on-1260H high-beta names such as 康龙化成(300759) and 昭衍新药(603127)). If the leaders are companies not on the list, this is sentiment diffusion rather than logical pricing, and persistence should be discounted.
• Relative A/H performance is today's best natural experiment: if the Hong Kong CXO names (药明康德(603259) 02359, 康龙 03759, 昭衍 06127) gain clearly more than the A-shares, the market is genuinely re-rating the geopolitical risk premium; if the A-shares surge while the H-shares are flat, it is an A-share sentiment crowd.
• Components / PCB sector: whether 生益科技(600183) can hold the board on volume is the best single signal for continuation of the main line. If it gaps up and fades, the PCB branch's handoff is at its end.
• Precious metals: watch whether the leader leads or the back row moves first. If 紫金矿业(601899) stalls while small caps hit limit-up first, that is the late-stage catch-up pattern.

8.3 Single-Stock Signals

• Absorption in the top 5 recommendations: 药明康德(603259), 广合科技(001389), 紫金矿业(601899), 东威科技(688700), 寒武纪(688256). Focus on whether the "gap up → quick fade to flat → late-session push" pattern of disagreement appears, which is common on good-news-realized days.
• Large-order net inflow: because this issue cannot verify two consecutive days of inflow, please verify it yourself intraday — only if a stock shows net inflow today in the same direction as last Friday does it constitute the first credible "two consecutive days" sample. This is the explicit homework this issue leaves to the reader.
• Whether a stronger name crowds others out: if 长鑫科技(688825) surges it will pull funds straight out of every proxy in the memory chain, and the Unitree subscription will pull funds out of the robotics concept.

8.4 Risk Signals

• Gap up then dive: two of the three main catalysts (CXO, precious metals) are already extended, so gap-up-then-fade is the highest-probability negative pattern today.
• A lone limit-up: if 药明康德(603259) is the only CXO name at limit-up while the rest are red, this is stock-specific logic rather than sector logic — which is in fact consistent with this issue's tiering (only one name is L1), so it need not be seen as failure, but expectations should be narrowed.
• Ladder attrition: last Friday's maximum was only 3 names with 4 consecutive limit-ups (百花医药(600721), 沃格光电(603773), 云南锗业(002428)), the 10-board run has broken, and the failed-limit-up rate was 26.0%. The attrition rate at height stays extremely high this week, and this issue does not include names above 4 boards in its watch pool.
• IPO liquidity drain: the Unitree subscription is today, and the frozen funds plus sentiment diversion are unhelpful for small and mid-cap theme stocks.
• Index level: last Friday the Shanghai Composite closed at 3940.04 (+1.02%) on two-market turnover of RMB 2.6644 trillion, above RMB 2 trillion for a 10th consecutive session. If turnover shrinks to below RMB 2.4 trillion today after a gap-up open, that is a clear demerit signal.
• FX: if the RMB keeps appreciating, it is a continuing drag on profits for exporters such as 广合科技(001389), 盛美上海(688082) and 凯莱英(002821), and it also discounts the pass-through from the USD gold price into RMB revenue.

---

9. Final Recommendations

① The 5 Stocks Most Worth Watching Today

② The 3 Strongest Branches Today

③ Directions Not Adopted as Main Branches, and Why

1. Humanoid robots / Unitree concept stocks — Unitree's subscription today is a scheduled event already priced on 08-06, not new weekend news; the overwhelming majority of concept names have no publicly verifiable supply relationship with Unitree; the robotics chain itself barely moved on Friday (Inovance +1.91%, 双环传动(002472) +0.36%, 新时达(002527) +0.94%), which says the market has digested it; and the subscription will drain liquidity.
2. The three memory module makers (佰维存储(688525), 江波龙(301308), 德明利(001309)) — note this is not because of a bearish view on the memory price cycle (Citi forecasts DRAM still +23% Q/Q next quarter, with prices peaking only in 2027Q2), but because of these three companies' own leverage and profit quality: PE historical percentile 3%–7% versus PB percentile 70%–82% is the signature of peaking earnings, not undervaluation; Longsys's net interest-bearing debt of RMB 12.6 billion ≈ net assets of RMB 12.4 billion; Biwin's operating cash flow / attributable profit is −93%; 德明利's Q2 attributable profit has already turned negative Q/Q. And none of the three has a publicly disclosed supply relationship with CXMT (no reliable data), so the Apple story reaches them only as sentiment spillover, with a reverse channel in which "domestic wafer allocation gets crowded out by mega-customers like Apple." To participate in this chain, the direct name 长鑫科技(688825) is cleaner than any proxy, both legally and commercially.
3. The high ladder above 4 boards (云南锗业(002428), 百花医药(600721), 沃格光电(603773) and others) — last Friday's maximum was only 3 names with 4 consecutive limit-ups, the 10-board run has broken, and the failed-limit-up rate rose to 26.0%, so the attrition rate at height stays extremely high this week; most of them also have clarification announcements or extreme valuations (云南锗业 PE 3430.5).
4. The CXO names with the biggest gains that are not on the 1260H list (康龙化成(300759), 昭衍新药(603127)) — they have risen the most, trade at 90%–114% premiums to their H-shares, and have the weakest causal link to the catalyst. 昭衍新药's core business is actually loss-making, with profit coming from fair-value revaluation of cynomolgus monkeys.
5. 盛美上海(688082) — the +42.14% attributable profit is mark-to-market gains, ex-non-recurring is −15.31%, the ex-non-recurring PE is 124.5, and goods shipped and contract liabilities are both falling. Do not list it alongside the three real-earnings names.
6. 光迅科技(002281) — the largest net main-force inflow in the market (+RMB 2.241 billion) but the largest institutional-only-seat net sale in the market (−RMB 274 million): hot money pushing the board while institutions distribute.

④ Final One-Sentence Judgment

Today is a day of "real catalysts, extended positions, and getting the direction right": all three main catalysts (the 药明康德(603259) judicial injunction, gold after the negative payrolls print, and the weekend interim results) are real and tradable for the first time today, but the vehicles for the first two already rose a good deal last Friday, so a good-news-realized gap-up-then-fade is today's highest-probability risk; the names that genuinely combine "earnings already landed + position not extended" are 广合科技(001389) and 紫金矿业(601899), while the memory chain, which looks the most "mispriced and low," is seeing its overseas pricing move the other way, so the deep decline is a trend rather than an expectation gap. This issue puts its observational weight on the core names of the main lines rather than the limit-up ladder, and treats "did it outperform its sector" rather than "did it rise" as today's test of being right.

---

⚠️ Risk warning: this list is pre-market information triage and observation 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.