A-Share Pre-Market Brief | 2026-09-16 Wednesday

News window: 2026-09-15 15:00 → 2026-09-16 07:00 (from the previous session's close to now). Publication timestamps for all market-wide filings are set by the East Money eiTime field. Overnight offshore read timestamp: after the 2026-09-15 close (CNBC, with last_time checked name by name as 2026-09-15T16:5x~17:1x-0400). Single-stock quotes are 9/15 closing settlement values (Tencent Finance, field 30 timestamp 202609151614). "Volume ratio" always uses Tencent's standard volume-ratio field (f49); the "turnover / trailing 9-day average" measure noted separately in this report is an independent caliber, and the two must not be mixed. Financial data are uniformly for the reporting period 2026-06-30 (2026 interim report), re-checked by the fundamentals-analyst sub-agent + local sampling back-calculation. ⚠️ The interim report is carry-over information that has been public for 2–6 weeks (statutory disclosure deadline 8/31) and is not itself a same-day catalyst. This report uses it only to rank "position size and holding period," not to manufacture direction — direction is always given by fund flows and the limit-up pool. (This constraint is the direct product of the empirical work in §2①.) The limit-up / limit-down / seal-break pools are the official exchange-caliber pools, validated as tc == len(pool) (ZT 32=32, DT 27=27, ZB 24=24). Price limits: Shanghai/Shenzhen main board ±10% | ChiNext (30x) / STAR Market (68x) ±20% | Beijing Stock Exchange ±30%. Every stock in this report is labeled with its board.

---

0. Today in one sentence

1. Today is a "no new catalyst day." Filings inside the window were dominated by the "abnormal share-trading fluctuation" category (20-odd of them), and that category was empirically proven just yesterday to have no predictive power over next-day direction (6 names filed warnings on the evening of 9/14, and 4 of them hit limit-up on 9/15). The absence of any sector-level new catalyst is itself the single most important piece of information today.

2. The most important conclusion this session is that I overturned my own first draft: gross margin cannot be used to rank same-day direction. I originally intended to rank the copper-clad laminate (CCL) chain by "2026 interim gross margin expansion." After QC I tested it on 14 peers: Spearman(YoY gross margin, trailing 40-session return) = −0.415, trailing 20 sessions = −0.156, 9/15 same day = −0.042 (n=14). The correlation is negative or zero. 满坤科技 (Mankun Technology, 301132), with the worst gross margin (−2.56pp), is +110.7% over the trailing 40 sessions, while 沪电股份 (WUS Printed Circuit, 002463), with expanding gross margin (+3.57pp), is only +2.6%. This is exactly the same mistake I made again today, one day after writing "fundamentals are not a short-term ranker" in yesterday's recap.

3. So the basis for ranking branches reverts to fund flows and the limit-up pool — the rule I set for myself in yesterday's recap. 9/15 net main-force inflows: 华正新材 (Huazheng New Material, 603186) +RMB 707 million, 生益科技 (Shengyi Technology, 600183) +RMB 693 million, 北方华创 (Naura Technology, 002371) +RMB 338 million; in the limit-up pool CCL has 华正新材, general-purpose board has 0. The CCL chain still ranks first today, but what supports it is fund flows and limit-ups, not gross margin.

4. ⚠️ A fact unfavorable to this list must be stated at the same time: 9/15 was the first time in 40 days that the CCL chain outperformed the general-purpose board. Mean returns, CCL group vs general-purpose board group: 9/15 single day +5.0% vs +1.1% (CCL wins), but trailing 5 sessions +17.1% vs +19.7%, trailing 10 sessions +16.9% vs +39.2%, trailing 20 sessions +13.0% vs +35.2%, trailing 40 sessions +23.8% vs +70.5% — the general-purpose board leads in all four windows, with the 40-session gap reaching 46.7pp. I am using one day of data against a month-long trend. This may be the first day of a style switch, or it may be noise — today is the day of judgment (falsification conditions in §8).

5. The biggest trap today is memory. "Apple accepts Samsung's 2027Q1 memory quotes, up 30%–40% from 26Q3" (Jiemian News exclusive, 9/15) looks like an S-grade positive, but the purest DRAM name, 长鑫科技 (CXMT, 688825, STAR Market ±20%), closed at −0.50% that day, and RMB 9.84 billion of turnover bought only +RMB 24 million of net main-force inflow, while 兆易创新 (GigaDevice, 603986) got only +RMB 25 million on RMB 10.69 billion of turnover. The +RMB 2.881 billion of net inflow into the memory-chip concept did not land on the three largest names.

6. The direction with the hottest sentiment is precisely the one with the worst quality: wind power and electric utilities together account for 7 limit-ups, 21.9% of the limit-up pool, the largest theme cluster in the market — and the Dragon-Tiger List exposed it. 大金重工 (Dajin Heavy Industry, 002487) institutional seats net sold RMB 47 million, 天顺风能 (Titan Wind Energy, 002531) institutions net sold RMB 33 million, while the same hot-money seat (Guotai Haitong Chengdu Tianfu Avenue) simultaneously bought RMB 100 million of 大金重工 + RMB 101 million of 天顺风能.

7. Pre-market state: low expectations, weak open. Overnight S&P −0.45%, Nasdaq −0.78%, Dow −0.63%, the 10-year Treasury yield sits at 5.006%, and China ADRs weakened across the board. The Philadelphia Semiconductor Index +0.40% looks counter-trend, but after excluding the M&A-driven SWKS/QRVO, the median of the 44-name semiconductor sample is −0.03% (−0.01% with the two included) — the median barely moves; overnight is neutral for A-share semiconductors.

---

1. News overview

⚠️ Three pieces of "old news / pseudo-news" were intercepted:

1. The YMTC IPO is not new news — accepted 8/21, under inquiry as of 9/3, 9 trading sessions ago. Several 9/15 articles listed it as the primary catalyst for the equipment rally; it can explain 9/15, but it will not provide another increment today.
2. The CLS article "Price hikes reappear in upstream copper-clad laminate for PCB" was published on 2026-04-12, not in September. The article ranks top in search and its headline carries no date; it contains three specific figures — Resonac +30%, Mitsubishi Gas Chemical +30%, Kingboard +10% — and citing it directly would build the entire branch on a price-increase letter from half a year ago. This report has adopted none of those figures.
3. "Nan Ya Plastics raising CCL prices 20%–25% from 9/1" is unconfirmed (found only on industry sites with no reliable publication date) and is not cited as fact.

A note on the caliber of the in-window filing count: locally, paginated scraping with ann_type=A returned 1,197 items (596 companies), while the QC agent independently rescanned with the same API and got 2,044, a 41% difference, suspected to be incomplete local pagination (there is prior form for "one-sided bias in East Money pagination"). This report therefore uses no statement of the form "the category with the most filings is X" that depends on a complete count; instead it uses a bounded statement: every directionally positive filing hit by keyword scanning inside the window was stock-level with no sector spillover (the two largest verifiable ones are the first share purchase by 东宏股份 (Donghong)'s controlling shareholder and the actual controller of 先惠技术 (Xianhui Technology) crossing the 1% mark, both small in scale). The substantive judgment "there is no sector-level new catalyst today" holds after independent QC re-verification.

---

2. Strongest positive branches, descending

① First, the most important self-refutation of this session: gross margin cannot rank direction

My first draft used "2026 interim gross margin expansion" as the ranking basis for the CCL chain. After QC I tested it empirically on 14 peers, and the conclusion is that this ruler does not hold:

Spearman(YoY gross margin, trailing 40 sessions) = −0.415 | trailing 20 sessions = −0.156 | 9/15 same day = −0.042 (n=14). All three windows are negative or zero. 满坤, with the worst gross margin, is first at +110.7%, while gross-margin expanders 沪电 at +2.6% and 鹏鼎 at −7.1% are at the bottom.

My recap yesterday recorded a fix: "fundamentals are used only to set the position-size cap and holding period; they do not participate in same-day direction ranking; direction ranking must start from the fund-flow set." Today's first draft violated it again. This time I nail it down with data: direction ranking for all branches and stocks in this report is decided by fund flows and the limit-up pool, and gross margin enters only the "position size and holding period" column.

② Along the way, the second claim in my first draft that gets overturned: "high-end PCB has pricing power, hence expansion"

The first draft wrote "gross margin expansion at 沪电/深南/鹏鼎/生益电子 ⇒ high-end PCB has pricing power." This induction skipped two counterexamples in its own table:

• 胜宏科技 (Victory Giant Technology, 300476) — the purest AI-server PCB name in the A-share market, with gross margin −3.00pp (among the worst in the table), directly refuting "high-end, therefore pricing power." It is simultaneously the cheapest in the whole chain (PE 44.21 / PB 5.86), with cash content 1.02x and the lowest debt ratio at 39.85%, and yet it is −5.9% over the trailing 40 sessions and −16.1% over the trailing 20.
• 兴森科技 (Fastprint Circuit Tech, 002436) — substrates / small-batch boards, which "should" have been squeezed, actually posted +2.50pp.
• In addition, 鹏鼎控股 is Apple FPC with revenue up only +5.1%, and its gross margin change has very little to do with CCL price increases; this is an attribution mismatch and has been removed from the evidence set.

Corrected statement: within high-end PCB, 3 expanded and 1 (胜宏) contracted — the direction is not consistent. The "does it have pricing power" explanation cannot cover the full sample, and I no longer treat it as a pillar of the branch. At the factual level only one thing is retained: the gross margins and profits of 沪电 and 深南 really are improving, but the market has withheld any re-rating for 40 straight sessions (+2.6% / +11.3%).

③ Corrected branch ranking (basis: fund flows + limit-up pool; fundamentals set position size only)

④ A correction of one measurement bias: semiconductors having "fund flow but no sentiment" is a statistical illusion

Yesterday I inferred "fund flow but no sentiment" from "semiconductors had only 2 limit-ups." That is measuring ±20% stocks with a ±10% ruler:

After normalization, semiconductors had 2 limit-ups + 2 STAR Market quasi limit-ups + 3 mid-strength gainers on the day; sentiment was not weak at all. This is the error expressly forbidden by iron rule 3 in CLAUDE.md; I committed it yesterday and correct it this session.

---

3. Master ranking of stock-level positive-news strength (20 names, descending by stock)

Caliber: financials are the 2026 interim report (public for 2–6 weeks, not a same-day catalyst). "Net main-force inflow" is East Money caliber (includes large hot-money orders); "net institutional buying" is the Dragon-Tiger List institutional-dedicated-seat caliber (buy and sell lists de-duplicated) — the two are not the same thing. Volume ratio is Tencent's standard volume ratio. Any item not re-checked is recorded as "not confirmable for now," with the count of unverified items noted next to the total score.

---

4. Stock scoring model (100 points total)

⚠️ Two self-imposed constraints on the model (products of QC):

1. The original model said "risk deductions 0 to −15," but the worked example reached −19, out of range. The cap has been explicitly changed to −20 this session, rather than forcing the score back inside the range.
2. Items marked "not confirmable for now" are scored 0 and are no longer converted into a specific value (the first draft gave the unverified 长鑫科技 "earnings elasticity 6 points, industry standing 10 points," which converted "I couldn't find it" into a score). The 7 affected names are flagged next to their totals as [incl. N unverified items].

Worked example — 生益科技 at 80 points: news source 14 + fund flow 19 (+RMB 693 million with RMB 12.4 billion of turnover) + earnings elasticity 14 + industry standing 13 + position 8 (trailing 40 sessions +10.7%, low within the chain) + persistence 8 + trading attributes 8 − risk 4 (PB at the 99.5% all-history percentile) = 80.

---

5. Detailed analysis of the Top 10

1生益科技600183the only name this session that clears fund flow, position and fundamentals all three · Shengyi Technology · Shanghai main board ±10%

• Direction basis (fund flow): 9/15 net main-force inflow +RMB 693 million, turnover of RMB 12.41 billion among the largest in the market, open 146.00 / close 150.25, open→close +2.91% (rising one-way all day, no gap-up-then-fade), standard volume ratio 1.27, turnover rate only 3.37% — the RMB 12.4 billion of turnover comes from a large market cap, not a high turnover rate.
• Position: trailing 40 sessions +10.7%, trailing 20 sessions +7.8%, on the low side among the five CCL names (华正 +70.0% and 南亚 +35.0% are both far higher). This is its biggest advantage relative to 华正新材.
• Fundamentals (position-size basis): revenue RMB 19.026 billion +50.0%, net profit attributable to parent RMB 3.287 billion +130.4%, ex-non-recurring +111.0%; gross margin 30.69% (+4.84pp), 26Q2 single quarter 32.63%; ROE 17.77% (a clear first among the four). Segments: CCL 64.95% of revenue (gross margin 29.16%, +5.47pp), PCB 29.01% (gross margin 31.20%, +3.35pp).
• The negatives that must be stated alongside: (a) operating cash flow / net profit 0.91x (1.36x a year earlier) — ⚠️ correcting an error in the first draft: this is not the lowest of the four; 金安国纪 at 0.469x and 南亚新材 at −0.051x (negative) are both worse; (b) debt-to-asset ratio 50.33% — ⚠️ likewise corrected: not the highest of the four; 华正新材 at 65.09% and 南亚新材 at 58.70% are higher; (c) receivables RMB 13.224 billion (+50.2%, in step with revenue), 31.2% of total assets.
• Valuation risk (the most important item here): PB 19.73, the 99.5% all-history percentile. The company's 8/7 risk-warning filing itself disclosed that "the CSI 'printed circuit board' industry price-to-book is 9.85, while the company's is 17.24" — and it has since risen from 17.24 to 19.73.
• Final judgment: priority deep-dive. The only name in the branch that can absorb large money, and its position within the chain is on the low side. But PB is at an all-history extreme, so the position must be handled on a medium-to-short holding period.

2北方华创002371direction is followable, position size must be compressed · Naura Technology · Shenzhen main board ±10%

In its favor: the absolute leader directly benefiting from the policy; net main-force inflow +RMB 338 million; turnover rate of only 0.92% (the steadiest float); branch breadth of 23 up / 2 down, the cleanest in the market; operating cash flow RMB 3.774 billion / net profit attributable to parent 112%, a sharp improvement from −RMB 3.191 billion in the year-earlier period.

Against it (all hard data):

Item	Reading	Meaning
Revenue vs net profit attributable to parent	Revenue +24.90%, net profit attributable to parent only +5.05%	Negative operating leverage
Consolidation effect	Kingsemi contributed about 5.4pp, organic about +19.5%; Kingsemi H1 net profit was only RMB 8.06 million	The acquisition lifts revenue and contributes almost nothing to net profit attributable to parent
FY2025	Revenue +30.85%, net profit attributable to parent −1.77%, ex-non-recurring −4.20%	The profit collapse has already lasted a year
Gross margin	H1 40.07%, −2.1pp YoY	Pricing under pressure
Contract liabilities	RMB 5.320 billion, only +6.30%	The forward indicator is far slower than revenue

Table scrolls horizontally

⚠️ A valuation illusion that must be exposed: its PE is at only the 47.2% all-history percentile, which looks "cheap." But from 2015 to 2022 its profit base was tiny, and the annual median PE reached as high as 118–243x. Once it entered the era of real profits, the annual median PE was 56.0 in 2023, 38.9 in 2024 and 41.9 in 2025 — the current 81.5x is clearly high relative to the same-caliber norm, and the 89.3% trailing-3-year percentile is the valid reading. PE and PB are both high in the same direction; there is no "low PE, high PB" divergence.

• Final judgment: watch closely, with an explicit position-size cap. Direction comes from policy and fund flows; the fundamentals only support a short holding period, not a heavy position or an overnight hold.

3金安国纪002636the first draft called it "the biggest expectation gap," and QC overturned that framing · Goldenmax International · Shenzhen main board ±10%

• What the first draft got wrong: I cited only the 9/15 single-day +1.81% and concluded from it that "market pricing clearly lags the financials." Once QC added the multi-day position, that sentence does not hold: it is +22.4% over the trailing 5 sessions and +596% over the past year. It is not "not yet started" — it is already catching up.
• But the reason to keep it still holds, just via a different argument: trailing 40 sessions +16.0%, ranking 9th among the 14 names and 4th among the five CCL names — the position really is on the low side (compare 华正 +70.0%, 崇达 +73.0%, 满坤 +110.7%). "Position on the low side" is a fact; "the market hasn't discovered it yet" was something I made up.
• Fundamentals (position-size basis): revenue RMB 3.399 billion +65.7%, net profit attributable to parent RMB 766 million +986.7%, gross margin 33.14%, +21.83pp, first in the chain. The profit is clean: ex-non-recurring RMB 770 million > net profit attributable to parent RMB 766 million, so non-recurring items are negative.
  • ⚠️ But the +986.7% has a low-base component: 25H1 net profit attributable to parent was only RMB 70 million. The §4 scoring has accordingly capped earnings elasticity at 11 points rather than full marks.
  • ⚠️ And the gross margin is not new information: 25Q1 10.8 → 25Q2 11.8 → 25Q4 19.7 → 26Q1 26.4 → 26Q2 37.1, five straight quarters of acceleration, and this slope was already public in the 26Q1 report (around April).
  • ⚠️ Cash quality is poor: operating cash flow / net profit 0.469x, second worst among the four CCL names.
• Technicals: turnover RMB 7.05 billion, turnover rate 11.58%, standard volume ratio 1.13 (flat volume); not on the Dragon-Tiger List, no seat information.
• Final judgment: watch closely (downgraded from the first draft's "priority deep-dive").

4华正新材603186the strongest fund flow, but also the highest position · Huazheng New Material · Shanghai main board ±10%

• ⚠️ Two corrections first: (a) the sub-agent's initial table recorded this stock's code as 002843 (which is actually 泰嘉股份, Taijia) and on that basis gave "net profit attributable to parent −30.7%." Local re-checking of the real 603186 data gives revenue RMB 2.960 billion (+41.3%), net profit attributable to parent RMB 173 million (+305.3%), ex-non-recurring +385.1%, the exact opposite direction; (b) but its YoY gross margin is only +2.99pp, the weakest of the four CCL names, and below 沪电 (+3.57), 深南 (+4.73) and 鹏鼎 (+4.90) — the first draft wrote "sharp expansion" for it in the table (the only cell in the whole table with no number), which is me using the net-profit caliber to cover a missing gross-margin figure. The net margin improvement (+3.8pp) exceeds the gross margin improvement (+2.99pp), which means part of the improvement is not a price contribution.
• Fund flow and pattern: net main-force inflow +RMB 707 million (highest in the sample); limit-up on shrinking volume — standard volume ratio of only 0.41 (turnover RMB 2.48 billion vs trailing 9-day average RMB 4.46 billion), sealed at 09:31:07, 1 seal break, turnover rate 6.35%. Sealing the limit-up with less volume means lighter selling pressure.
• ⚠️ But the position is the highest in the chain: trailing 40 sessions +70.0%, trailing 20 sessions +36.6%, 4th among the 14 names. This and "limit-up on shrinking volume" are two signals pointing in opposite directions, and I am hiding neither of them.
• Risks: PE 96.92 / PB 16.01; 2nd limit-up within 5 days; not on the Dragon-Tiger List, no seat information.
• Final judgment: watch closely, but explicitly limited to a single day and a capped position. It is the name with the strongest fund flow, and also the one with the greatest chase-high risk.

5长鑫科技CXMT / 688825the first draft overstated the evidence; restated here at its actual strength · STAR Market ±20%

Test item	Reading	Note
Directness of the positive news	The purest DRAM name, the most direct A-share mapping of "Samsung quotes up 30–40%"	Should react first
9/15 change	−0.50%	Did not react
Volume ratio	0.96 (standard caliber)	⚠️ Correction: the first draft wrote 0.81 using a homemade "turnover / trailing 9-day average" caliber and called it "shrinking volume." At the standard volume ratio of 0.96 this is "volume in line with normal," not shrinking.
Net main-force inflow	+RMB 24 million (RMB 9.84 billion of turnover)	Net inflow is close to zero
Recent price action	−5.2% cumulative over the trailing 3 sessions	⚠️ Correction: the first draft's "5th straight day of grinding down" was wrong; 9/10 was +0.58%. It is really 3 down days.
Valuation	PE 44.99 / PB 18.31	PB extremely high
Time since listing	Listed 2026-07-27, only 37 trading sessions	No comparable historical benchmark, and the lockup-expiry schedule has not been checked

Table scrolls horizontally

Strength of the corrected conclusion: downgraded from "a textbook distribution pattern" to "the positive news did not buy any incremental demand." Only two pieces of evidence stand up: (a) the most direct name closed lower on the positive-news day; (b) RMB 9.84 billion of turnover bought only +RMB 24 million of net main-force inflow, and the memory concept's +RMB 2.881 billion did not land on the three largest names (长鑫 +RMB 24 million, 兆易 +RMB 25 million, 江波龙 −RMB 220 million). ⚠️ There is also a gap I did not check: within the memory concept's +RMB 2.881 billion, the three leaders together account for only −RMB 171 million, and I did not trace where the remaining roughly RMB 3 billion went. If it is a second-tier broadening, that would be bullish evidence and would undermine this characterization. This is the largest uncertainty in this section, and §8 retains a falsifiable condition for reinstating it.

⑥ A branch-level warning on semiconductor equipment

The two gaps are almost identical, gross margins are both down 2–3pp, and net profit growth at both is only 1/5~1/3 of revenue growth. The pattern looks like "trading price for volume + a downward mix shift." But three limitations must be stated: (a) contract liabilities are only a proxy for the order book, and neither company discloses order amounts; (b) 北方华创's contract liabilities are +23.95% versus year-end 2025, which is an improvement — the half-year and sequential calibers point in opposite directions; (c) n=2. This is one-sided evidence, not a proven conclusion. Operational implication: direction can be given, a heavy position cannot; exit the moment the fund-flow verification point fails, and do not hold on the grounds of "waiting for the fundamentals to come through" — because the fundamentals themselves are weakening at the margin.

7胜宏科技300476my framework cannot explain it, so I list it separately · Victory Giant Technology · ChiNext ±20%

• It is the purest AI-server PCB name in the A-share market; by "high-end means pricing power" it should have expanded, yet its gross margin is −3.00pp; by "general-purpose boards get squeezed" it should have fallen, but it is not a general-purpose board either.
• At the same time it has the lowest valuation in the chain (PE 44.21 / PB 5.86), cash content 1.02x, the lowest debt ratio at 39.85%, net profit attributable to parent +33.3%, while being −5.9% over the trailing 40 sessions and −16.1% over the trailing 20, the worst in the whole table.
• I have no framework that explains it, and I will not invent one. Treated as "attribution unclear": no recommendation slot, but placed on the tracking list — it is simultaneously the strongest counterexample to two of my frameworks and the cheapest name in the chain, and both of those are worth watching.

8超声电子000823three lines of evidence point to the same conclusion · Shantou Ultrasonic Electronics · Shenzhen main board ±10%

Dimension	Reading
Company's own statement (9/13 15:32 filing)	"The company currently supplies no products to NVIDIA"; PCB supporting 800G/1.6T optical modules "has generated no corresponding revenue to date"
Business mix	PCB is only 46.27% of revenue (another 25.30% is LCD monitors); the PCB segment's gross margin is 13.78%, −4.41pp, the largest single-segment decline in the full sample; the CCL segment's gross margin is only 11.92% (versus 生益 29.16% and 金安 33.14%), i.e. low-end CCL that has not captured this round of price increases
Trading pattern	Day 5 after 4 consecutive limit-ups, +5.61%, turnover rate 25.93%→30.75%→40.42%, accelerating daily; position: trailing 40 sessions +72.9%; net institutional buying RMB 273 million, but net main-force outflow of RMB 416 million for the day (hot-money net selling RMB 241 million)

Table scrolls horizontally

"Net institutional buying" and "net main-force outflow" are not contradictory; they are two calibers measuring different things: the Dragon-Tiger List counts only the top 5 buy/sell seats on the listing day. The conclusion is "institutions are taking, hot money is leaving." In its own denial filing the company itself named the 9/10 turnover rate of 25.93% and the 9/11 turnover rate of 30.75% as evidence of "a risk of overheated trading" — and 9/15 already reached 40.42%.

• Final judgment: Pass (high position).

9大金重工002487the dissection sample for the wind-power line · Dajin Heavy Industry · Shenzhen main board ±10%

• On the surface: limit-up, net main-force inflow +RMB 602 million, total Dragon-Tiger List net buying +RMB 119 million.
• It reverses once the seats are broken out (buy and sell lists de-duplicated): institutional-dedicated seats net sold −RMB 47 million, Shenzhen-HK Stock Connect +RMB 59 million, hot money +RMB 107 million, of which the single Guotai Haitong Chengdu Tianfu Avenue seat bought RMB 100 million.
• The same hot-money seat simultaneously bought RMB 101 million of 天顺风能 (where institutions also net sold −RMB 33 million). One hot-money seat pushed two stocks at once, and institutions were selling in both.
• This answers the "wind-power attribution unclear" item left over from yesterday: the attribution is still unclear, but the nature of the money is now clear — a single hot-money line driving it, with institutions distributing into the move.
• Final judgment: Pass.

10意华股份002897and 双星新材 (Shuangxing New Material, 002585) — two mislabeled names · Yihua

意华股份: the market treats it as a "server and switch connector" name. The interim report mentions "server," "switch" and "data center" 0 times each.

• Real mix: PV mounting is 42.73% of revenue and −30.6% YoY (the largest single business, and shrinking); communications connectors are only 28.72%. The interim report does contain substantive disclosure on high-speed optoelectronic connectors (QSFP-DD 400G and OSFP112 800G already shipping in volume, customers including Huawei, ZTE and Foxconn), but how much of the RMB 822 million the high-speed portion accounts for is not disclosed and is not confirmable for now.
• Earnings: net profit attributable to parent −37.87%; gross margin 17.78%, lowest of the 5 periods; contract liabilities −35.5%.
• Dragon-Tiger List caliber correction: the listing reason is "cumulative price deviation of 20% over three consecutive trading days," so the net institutional buying of RMB 92 million is the 9/11 + 9/14 + 9/15 three-day cumulative total (the Dragon-Tiger List's "total market turnover" of RMB 5.204 billion ≈ the three-day sum; 9/15 alone was only RMB 2.294 billion). And over the same period another institutional seat net sold RMB 26 million.
• Final judgment: Pass.

双星新材: the company's 9/14 17:45 filing directly rejects the MLCC label in its own words:

"The MLCC release film project is still in the cultivation and construction stage... the product currently has negative gross margin, and the project accounted for about 0.3% of the company's total revenue in 2025"; "the company is not involved in AI computing-power-related application areas"

• 0.3% × 2025 revenue of RMB 5.108 billion ≈ RMB 15.30 million, with negative gross margin; the interim report's product-level disclosure contains no line item called "MLCC release film." H1 ex-non-recurring profit was only RMB 5.86 million, so 87% of the profit comes from non-recurring items; FY2025 net profit attributable to parent was −RMB 527 million. PB 1.64 looks cheap, but it corresponds to the 97.1% trailing-3-year percentile (the 2024/2025 median PB was only 0.73/0.71).
• Final judgment: Pass. Note that the filing was published after the close on 9/14 while 9/15 still produced a locked limit-up for the 3rd consecutive board — again confirming that "a company's self-initiated risk warning has no predictive power over next-day direction"; it constrains position size, not direction. But as a matter of business fact, the MLCC label does not hold.

---

6. Pass list

---

7. Intra-branch ranking

Branch 1: Copper-clad laminate (CCL) — ranked by fund flow and position; gross margin enters only the position-size column

The most solid name: 生益科技 (fund flow + low position + fundamentals, all three). Strongest fund flow but highest position: 华正新材 — these two signals point in opposite directions, hence the single-day limit. Branch-level risk: (a) underperformed the general-purpose board by 46.7pp over the trailing 40 sessions, with 9/15 only the first day of reversal; (b) valuations high across the board.

Branch 2: Semiconductor equipment

Branch 5: High-end PCB (good fundamentals, the market has refused for 40 straight sessions)

Industries in yesterday's limit-up pool that could not be attributed (standing section)

Re-check: industries with "zero gainers" yesterday (standing action)

---

8. Verification signals at tomorrow's open

Design principles (the direct product of yesterday's reconciliation): ① structural criteria outrank price criteria — yesterday's market-wide pass rate for "can it hold after a gap-up open" was only 17.4%; ② no single list may contain more than two criteria from the same source; ③ every verification point must target the variable that would actually lose me money, and must have an absolute condition, not only a relative one.

① The number-one judge this session: is CCL's one-day reversal a real switch or noise?

② Structural criteria for the other branches

③ Stock-level signals (each with an absolute threshold)

• 生益科技 (600183): turnover holds > RMB 10 billion (9/15 was RMB 12.41 billion); it should not gap up sharply (9/15 was "open at the low, up all day").
• 金安国纪 (002636): it must rise on expanding volume; the absolute condition is a gain > 3% with a volume ratio > 1.3. On 9/15 it had volume without price (turnover rate 11.58%, volume ratio 1.13, up only 1.81%); if volume stays large and price stalls again today, that is distribution, not accumulation.
  • ⚠️ The first draft wrote its verification point as "can it outperform 生益科技" — a purely relative criterion: CCL falling 6% while it falls 5% would also "pass," which does not test the variable that actually loses me money. Replaced with the absolute condition above.
• 华正新材 (603186): the session after a limit-up, it must not break the 9/15 opening price of RMB 239.70 (prior close 228.70, open today 239.70, close 251.57; the three prices have been back-calculated as self-consistent).
  • ⚠️ The first draft wrote "call-auction turnover not exceeding 1/4 of yesterday's full-day total (about RMB 620 million)" — RMB 620 million is only about 1.59% of free-float market cap, which in call-auction practice is almost impossible to reach, i.e. a threshold that passes almost by construction (I have been burned by "a volume threshold that passed with 18x of headroom and still lost money"). Replaced.
• 拓荆科技 (688072): standard volume ratio holds > 1.5, and it must reclaim the 9/7 close of RMB 651.92 to count as a breakout (the 9/15 close of 649.00 is still below it).
• 超声电子 (000823): if volume expands again today while the gain keeps narrowing (a 2nd consecutive day of expanding volume and stalling price), distribution is confirmed.

④ Risk signals

1. The 10-year Treasury yield sits at 5.006%. ⚠️ Caliber note: CNBC's Treasury change_pct field is broken again this session (US2Y is given as 0.00% while last/prev differ by +0.8bp), so this report has discarded that field, and therefore the direction of "up/down versus the prior value" was not obtained this session — only the level is stated, with no assertion about direction. But the absolute level of 5% itself matters for this list: semiconductor equipment names all trade at PE 74–111 and CCL names all at PB 14–25, all long-duration assets.
2. China ADRs weakened across the board overnight (KWEB −1.01%, FXI −1.26%, YINN −3.77%).
3. Do not chase highs in the call auction: on 9/15, 1,707 names gapped up market-wide, and only 297 = 17.4% "held after the gap-up"; 桂林旅游 (Guilin Tourism) open→close −15.48%, 欧亚集团 −17.49%.
4. If wind power / utilities keep broadening: note that their institutional seats were net sellers, that this is a single-line hot-money move, and that seal quality is the worst in the whole pool (this item is an observational statement and contains no trading advice).
5. The base of the limit-up pool keeps shrinking: 55 names on 9/14 → 32 on 9/15 (−42%), seal-break rate 42.9%, first-boards 78.1%, advance rate 13.0%.

⑤ Calendar for the next 5 trading sessions

---

9. Final recommendation conclusions

① The 5 stocks most worth watching today

⚠️ The double standard on admitting 拓荆科技 versus 臻宝科技 must be explained clearly: §6 passes on 臻宝科技 for "industry classification unverified," while 拓荆科技's "financials not re-checked" still got a recommendation slot. These are not the same kind of uncertainty — with 臻宝 it is not even confirmed whether it belongs to this supply chain at all (the qualitative claim is in doubt), whereas 拓荆 has a clear industry standing (domestic leader in thin-film deposition equipment, directly covered by the policy, with fund flow and volume) but I simply did not pull its financials this time (a quantitative gap). Even so, this is still a loosened criterion, so I give it the smallest position in the entire list and flag [incl. 1 unverified item] next to its total. If its volume ratio drops below 1.5 today, it has no fundamentals to fall back on.

Position discipline (the direct product of this session's empirical work): §2① has proven that the Spearman correlation between gross margin and same-day direction is −0.415 ~ −0.042. The "reason for recommendation" column above therefore contains only fund flow and position, and "fundamentals" has been moved wholesale into the position-size column. Anyone who uses this table to size positions by fundamental strength is repeating exactly the error the data refuted for me today.

⚠️ Concentration: 3 of the 5 slots are in CCL (slots 1, 3 and 4). The first draft claimed the three verification points were independent variables; QC pointed out that this is wrong — all three presuppose that "the CCL framework holds," so they are same-source. And §8① shows that this premise is being attacked by 40 sessions of relative-strength data. So the treatment this session is not to hide behind diversification, but: (a) make §8① the number-one judge for the whole list; (b) state explicitly that "if the primary criterion fails, all three recommendation slots lapse at once"; (c) set the three CCL positions at medium / medium-low / smallest respectively, with no overnight hold. This is a list that writes its own greatest weakness right at the front.

② The 3 strongest branches today

③ Directions not recommended for chasing today + reasons

1. Memory (DRAM / NAND / modules) — the most direct name, 长鑫科技 (688825), closed down 0.50% on the very day of "prices up 30–40%," and RMB 9.84 billion of turnover bought only +RMB 24 million of net inflow; the concept-level +RMB 2.881 billion did not land on the three largest names. ⚠️ But note that I did not trace where the remaining roughly RMB 3 billion went; if it is a second-tier broadening, this characterization would be undermined (§8② leaves a reinstatement condition).
2. General-purpose / small-and-medium-batch PCB — the interim reports show "revenue growth with profit decline" (崇达 +22.6%/−31.3%, 满坤 +24.0%/−30.5%), and 崇达 and 满坤 both self-stated cost squeeze and overheated sentiment in their 9/15 filings. ⚠️ But in fairness it must be said: this group is +70.5% over the trailing 40 sessions, far ahead of the CCL group I am recommending at +23.8%. My Pass on them rests on fundamentals and the companies' own statements, and fundamentals have been a negatively correlated ranker over the past 40 days. This is the highest-risk judgment in this list.
3. Wind power and utilities — institutions net sold in both leaders, and the same hot-money seat bought both for about RMB 100 million each; 闽东电力's seal order / turnover is 0.006 with 28 seal breaks; attribution remains unconfirmed to this day.
4. Mislabeled theme stocks — 双星新材 (002585) and MLCC (the company itself states negative gross margin and 0.3% of revenue), 意华股份 (002897) and "server connectors" (the interim report mentions server/switch/data center 0 times), 超声电子 (000823) and NVIDIA (the company itself states there is no supply). All three labels are denied by the companies' own disclosures.
5. 通鼎互联 (002491) — northbound net buying of RMB 437 million was the largest in the market, but its 2026H1 optical fiber and cable segment gross profit of RMB 272 million equals 134% of company-wide net profit attributable to parent, with no explanation anywhere in the interim report. Until the source of profit can be verified, no amount of money constitutes a reason to buy.
6. Agricultural planting, retail, travel and hotels — the three together make up 44.4% of the limit-down pool, and "zero gainers" is harder evidence of a receding tide than the size of the decline.
7. All chase-high behavior in the call auction — 1,707 names gapped up on 9/15, and only 17.4% "held after the gap-up."

④ Final one-sentence judgment

There is no new news today, so what is being tested is whether I corrected yesterday's homework — and what I most need to admit today is that I once again tried to rank short-term names by gross margin, and the correlation coefficient of −0.415 across a 14-name sample snapped that ruler on the spot; the direction in this list now comes entirely from fund flows and the limit-up pool, fundamentals decide only position size, and its most fragile point is written right at the front — the CCL chain has now lost to the general-purpose board for 40 days and won only one day on 9/15, and today is the day of judgment.

---

⚠️ Data-retrieval failures, QC revisions and open items this session (not for clients)

1. East Money's push2.eastmoney.com and push2his were unavailable across the board: stock_individual_fund_flow and stock_zh_a_hist both returned ConnectionError: RemoteDisconnected; on the sub-agent side, all three of clist/push2/hist refused connections and CNINFO returned 403. Switched paths: fund flow via push2delay's fflow/daykline; daily bars via Tencent web.ifzq.gtimg.cn/appstock/app/fqkline/get (a channel newly adopted this session, success rate 13/13, with the last date asserted = 2026-09-15); quotes via qt.gtimg.cn.
2. push2delay.../api/qt/stock/get's f62/f184 return garbage values (f62 is always 0, f184 gives 873.64%), identified and discarded. New pitfall: fund-flow fields are not equivalent across different endpoints in the same cluster.
3. daykline returns only the last 1 day when lmt=0, so stock-level "net inflow on two consecutive days" could not be obtained, and §8's "3rd consecutive day" can only be executed at the industry level.
4. The Dragon-Tiger List was usable for the first time this session. Seat details were de-duplicated on the (branch, buy amount, sell amount) triple (the buy and sell lists contain completely identical rows), and 13/17 names closed with a 0.00 discrepancy. For the 5 names that failed to close (锡华 +0.15, 博汇 +0.41, 宏盛 +0.76, 气派 +0.49, 天融信 +0.03, in RMB 100 millions) their amount breakdowns were not used; for 天融信 only the "0 institutional seats" qualitative conclusion, which does not depend on amounts, was used.
5. ⚠️ The most important sub-agent error interception this session: the PCB-chain agent recorded 华正新材's code as 002843 (which is actually 泰嘉股份) and on that basis gave "net profit attributable to parent −30.7%." Three-step verification: (a) a Tencent reverse lookup confirms 002843 = 泰嘉股份 and 603186 = 华正新材; (b) turnover of RMB 2.48 billion ÷ turnover rate of 6.35% ≈ RMB 39 billion of free float, which is inconsistent in order of magnitude with "H1 revenue of RMB 836 million + PE 96.92"; (c) pulling 603186's report directly gives revenue RMB 2.960 billion (+41.3%), net profit attributable to parent +305.3%, ex-non-recurring +385.1%. The direction is completely opposite — had it been accepted directly, 华正 would have gone from a recommendation slot to a Pass.
6. ⚠️ risk-auditor was only launched this time after the first draft was complete, violating my own rule of "never skip QC before finalizing," and the cost was very concrete: it caught 7 "high"-severity red flags, of which at least 4 (the contradiction between the gross-margin ranking rule and its results, 长鑫's count of down days, the two volume-ratio calibers, and the comparative-statement factual errors) should have been blocked at writing time. Next time QC must be launched in parallel as soon as the §3 master ranking takes shape, not after the whole draft is written.
7. Substantive conclusions overturned and revised by QC (logged item by item):
   • (a) The gross-margin ranking was empirically refuted — locally recomputed Spearman(YoY gross margin, trailing 40 sessions) = −0.415, trailing 20 sessions −0.156, same day −0.042 (n=14, Tencent daily bars). Branch 1 was downgraded from A+ to A, the ranking basis was switched wholesale to fund flow + position, and gross margin was moved into the position-size column. This is the single largest revision of this session.
   • (b) Multi-window backtest of CCL vs the general-purpose board: 1 day +5.0% vs +1.1% (CCL wins), while 5/10/20/40 sessions were −2.6 / −22.3 / −22.2 / −46.7pp, all losses. Written into §0④ and §8① and set as the number-one judge.
   • (c) 华正新材's gross margin is actually +2.99pp (weakest of the four CCL names and below 沪电/深南/鹏鼎); the first draft's "sharp expansion" used the net-profit caliber to cover a missing gross-margin figure. Fixed; its recommendation slot was moved from 3rd to 4th and limited to a single day.
   • (d) 金安国纪's "biggest expectation gap" was overturned (trailing 5 sessions +22.4%, past year +596%), replaced with the verifiable statement "position on the low side," and downgraded from "priority deep-dive" to "watch closely"; its verification point was changed from a purely relative criterion to an absolute condition.
   • (e) Three comparative factual errors were corrected: 生益's cash content of 0.91x is not the lowest of the four (金安 0.469x, 南亚 −0.051x, negative); 生益's debt ratio of 50.33% is not the highest of the four (华正 65.09%, 南亚 58.70%). The first draft disclosed the same metric only for recommendation slot 1 and hid it for slots 2/4, which is one-sided disclosure. Now filled in.
   • (f) The "5 consecutive days of grinding down" claim for 长鑫科技 was wrong (9/10 was +0.58%, so it is really 3 down days); "volume ratio 0.81" was a homemade caliber, and the standard volume ratio of 0.96 = flat volume, not shrinking. The strength of the conclusion was downgraded from "a textbook distribution pattern" to "the positive news did not buy any incremental demand."
   • (g) The two counterexamples 胜宏科技 (−3.00pp) and 兴森科技 (+2.50pp) were skipped wholesale by the narrative in the first draft; they have now been written explicitly into §2②, §3 and §7; 鹏鼎控股 (Apple FPC, revenue +5.1%) as evidence for "high-end PCB has pricing power" is an attribution mismatch and has been removed from the evidence set.
   • (h) 拓荆科技's "breakout on expanding volume" does not hold (close 649.00 < 9/7 close 651.92 < 8/31 close 677.86); it has been changed to "a volume-expanding rebound that did not break the prior high," and the absolute threshold of 651.92 was added to the verification point; the double standard versus 臻宝科技 is explained explicitly in §9① and given the smallest position.
   • (i) The two volume-ratio calibers have been unified to Tencent's standard volume ratio f49 and stated in the caliber declaration; §8's threshold was accordingly moved from 1.2 to 1.3 (because 9/15 was already at 0.96).
   • (j) §8④4 originally read "can participate but limited to a single day, with a capped position and no overnight hold," which conflicted with four "avoid/Pass" characterizations elsewhere in the report and crossed the line into investment advice; it has been changed to a purely observational statement.
   • (k) "Upstream fund flow of +RMB 1.664 billion" was circular reasoning (it exactly equals the sum of the 4 names I recommended); that statement has been deleted and replaced with a name-by-name listing of net inflows.
   • (l) The scoring model's deduction cap of −15 conflicted with the worked example's −19; the cap has been changed to −20 (rather than forcing the score back into the range); items marked "not confirmable for now" are scored 0 and flagged next to the total as [incl. N unverified items] (the first draft had given the unverified 长鑫 specific scores).
   • (m) "10Y broke above 5%" lacked a basis (the change_pct field has been discarded, so there is no reading for the direction of change); it has been changed to "sits at 5.006%" with an explicit note that the direction was not obtained.
   • (n) Filing count 1,197 vs QC's 2,044 (a 41% difference), suspected to be incomplete local pagination. All statements depending on a complete count have been deleted (e.g. "the most numerous category is abnormal fluctuations at 21 filings") and replaced with bounded statements. The substantive judgment "there is no sector-level new catalyst today" holds after independent QC re-verification.
   • (o) 崇达技术's 26Q2 single-quarter gross margin of 20.39% has already recovered from 19.82% sequentially, so the characterization was changed to "still under pressure YoY, past the inflection point sequentially"; the Pass table was completed with the same-day moves for 满坤 +6.05%, 崇达 +2.56% and 景旺 −3.36% (the first draft gave only the decliner 中京 −1.01%, which is one-sided disclosure).
   • (p) 华正's verification point of "RMB 620 million in the call auction" equals about 1.59% of free-float market cap, a threshold that passes almost by construction, and it has been replaced with "does not break the opening price of 239.70."
8. Confirmed correct by QC, no changes needed: all arithmetic spot checks passed (limit-up pool 7/32=21.88%, limit-down pool 12/27=44.44%, seal-break rate 42.86%, first-boards 78.1%, gap-ups that held 17.4%, 华正's three prices 228.70×1.10=251.57, 意华 2.863 billion×28.72%=RMB 822 million, 双星 5.108 billion×0.3%=RMB 15.30 million, equipment gaps −18.6/−18.1pp); the primary financial data match the primary-source APIs 100%; 金安国纪's profit is clean (ex-non-recurring RMB 770 million > net profit attributable to parent RMB 766 million, with negative non-recurring items), so the reality of its earnings cannot be questioned on low-base grounds; §5⑥, the branch-level equipment warning, was rated the section in the whole report where evidence strength and conclusion strength match best, and is kept as is.
9. Not obtained / explicitly left blank: (a) an in-window primary source for CCL price increases; (b) a primary source for the wind-power catalyst (not obtained for a second consecutive session); (c) 7 names with unverified financials (拓荆, 中微, 长川, 宏和, 气派, 臻宝, 长鑫) — the body text records "not confirmable for now" throughout and scores them 0; (d) the high-speed connector share at 意华; (e) the source of 通鼎's 60.70% optical fiber and cable gross margin; (f) where about RMB 3 billion of the memory concept's +RMB 2.881 billion went (§5⑤ already flags that this gap would undermine that characterization); (g) 长鑫 has been listed for only 37 trading sessions, and its lockup-expiry schedule has not been checked; (h) attribution for the automation equipment and household goods limit-up industries; (i) stock codes and names for the lockup-expiry details (the API's return table contains no code/name columns, so §8⑤ gives amounts only and no stock names were fabricated).
10. Sub-agent usage: 2 × fundamentals-analyst (4 PCB-chain names + 5 semiconductor/connector names, actually extended to 14 peer comparisons) + 1 × risk-auditor (waited for synchronously, with the full feedback processed item by item as in 7). Scripts are left in work_ashare0916/.

---

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

Data sources: exchange-caliber limit-up/limit-down/seal-break pools (East Money push2ex, validated tc == len(pool)) | Dragon-Tiger List and seat details (East Money, buy and sell lists de-duplicated) | single-stock quotes and standard volume ratio (Tencent Finance qt.gtimg.cn, 9/15 closing settlement values) | daily bars and multi-window returns (Tencent web.ifzq.gtimg.cn, forward-adjusted) | stock-level fund flow (East Money push2delay) | financial data (akshare stock_financial_abstract / stock_zygc_em, reporting period 2026-06-30) | company filings and publication timestamps (East Money filings API eiTime field) | overnight US equities and Treasuries (CNBC, last_time freshness checked) | 15th Five-Year Plan for the Development of the Electronic Information Manufacturing Industry — China News Service¹ | August national economic performance — 21jingji³ | Apple accepts Samsung memory quotes — 21jingji² | YMTC IPO accepted — CLS⁵