A-Share Pre-Market Brief | 2026-09-17 Thursday

News window: 2026-09-16 15:00 → 2026-09-17 08:00 (Beijing time). Market-wide filings inside the window: 1,499 (East Money security/ann, publication moment set by eiTime; the latest one is 09-16 23:24:21). Price baseline: A-share price/volume figures are 2026-09-16 closing settlement values (Tencent qt.gtimg.cn; timestamps of all candidate names are 20260916). The market-wide snapshot comes from East Money push2delay, 5,915/5,915 with no pagination gap, valid sample 5,551 names. Price limits: Shanghai/Shenzhen main boards ±10% | ChiNext (30x)/STAR Market (68x) ±20% | Beijing Stock Exchange ±30%. Every name in this report is tagged with its board; limit-ups and limit breaks are judged against their own limit. Limit-up/limit-down/limit-break: official exchange-caliber pools, already validated with tc == len(pool) (ZT 89=89, DT 4=4, ZB 11=11). Dragon-Tiger List (LHB): this issue did obtain the 9/16 LHB (73 entries) and the seat detail for key names; seats are de-duplicated across the buy and sell lists and reconciled against net buy value. Fundamentals caliber: 11 companies independently verified by three sub-agents; cumulative and single-quarter figures are shown separately, and all PE figures are TTM (source: East Money income-statement/balance-sheet interfaces, 2026 interim report; gross margin back-computed from OPERATE_COST; valuation percentiles from stock_value_em). ⚠️ The "PE (dynamic)" shown on East Money quote pages is a half-year annualized figure, which produces a positive PE for companies that are loss-making on a TTM basis (赛意信息 (Ysstech Info-Tech) shows 98.60x, 康惠股份 (Kanghui Pharmaceutical) 235.86x, while both are in fact TTM loss-making) — this report does not use that caliber. ⚠️ Of the 22 names in this report, only 10 have been financially verified; unverified ones are marked ⬜ in §3 and their scores are capped. Unverified does not mean good quality. Northbound flows: since 2024-08-19 the exchanges have stopped disclosing daily net buy value, so this report gives no aggregate reading; we only use the "Shenzhen/Shanghai Stock Connect dedicated" seats at the single-stock level on the LHB. This issue has been independently QC-ed by risk-auditor; the items adopted and acted upon are listed in §10.

---

0. Today's Bottom Line

1. The chronology is real, but the conclusions drawn from it must be discounted. The broad A-share rally on 9/16 (75.1% of names up, semiconductors 184 up / 1 down) did all happen before the FOMC release — the A-share 9/16 15:00 close = 9/16 03:00 US Eastern, and the decision was released during the US session on 9/16. But the "hike" itself was not new information: before the meeting, CME FedWatch put the September hike probability above 90%. The only thing that genuinely arrived for the first time is the residual — the dot plot and the wording, not the hike.

2. That residual is indeed hawkish: a 12–0 unanimous 25bp hike to 3.75%–4.00% (the first since July 2023), the dot plot shows 16 of 18 officials expecting one more hike this year and 4 expecting two, and the 2% inflation target is not expected to be met until 2029.

3. ⚠️ But the overnight sector split runs exactly counter to the "tightening suppresses growth" intuition, and this is the one table to look at first in this report:

   Overnight decliners		Overnight flat/up
   Energy XLE	−2.88%	Semiconductors SMH	+0.64%
   Regional banks KRE	−1.77%	Tech XLK	+0.10%
   Financials XLF	−1.62%	Healthcare XLV	+0.07%
   Dow .DJI	−1.21%	Utilities XLU	UNCH
   Software IGV	−0.56%	A-share proxy ASHR	+0.12%
   S&P .SPX	−0.45%	Nasdaq .IXIC	−0.01%

   Table scrolls horizontally

   On a rate-hike day, banks fell, energy fell, while tech and semiconductors closed higher — this is not the textbook reaction to a hike. What the market read was not "rates are higher so banks earn more," but rather "tightening hurts credit and cyclicals, not growth." So the inference "hike → A-share growth names must catch down today" is not supported by the overnight tape.

4. The Treasury "5%" is background, not a same-day shock: 10Y 5.004%, 2Y 4.726%, 5Y 4.861%, 30Y 5.346%. ⚠️ Two readings this issue gave contradictory same-day bp changes (at 17:05 ET the 10Y read +0.32%, at 19:50 ET it read UNCH), a known rolling-previous-close problem with that data source, so this report only gives levels and does not claim a large same-day rate jump occurred. Curve shape: 10Y−2Y about +28bp.

5. The renminbi gives no stress signal: offshore USDCNH 6.7105 (−0.03%), onshore USD/CNY 6.706. This matters, because the core rationale for the top two names on this list is "foreign buying," and the exchange rate is the premise of that rationale.

6. Semiconductors held overnight, but with a discount: the Philadelphia Semiconductor Index +0.63%, SMH +0.64%; however the equal-weighted mean of the 25 SOX constituents is −0.03% (15 up / 10 down); and for "semiconductor equipment," the strongest A-share group on 9/16, the US peers were all down overnight: AMAT −1.37%, LRCX −0.61%, KLAC −0.39%.

7. ⚠️ Fundamental verification overturned my own initial ranking, and this has to come early: the two biggest gainers on 9/16 happen to be the two with the worst financial quality and the heaviest accounting cosmetics — 有研新材 (GRINM Advanced Materials) +9.99% limit-up, with an H1 gross margin of just 7.42%, TTM operating cash flow −RMB 804 million, PB at the 99th percentile of its full history, and the only one of the four verified names with no sell-side earnings coverage; 中微公司 (AMEC) +6.65%, whose +300% attributable net profit includes RMB 1.982 billion from the revaluation of outside equity investments (the company's own pre-announcement text), with contract liabilities −12.4% YoY and down sequentially for three consecutive quarters. 北方华创 (Naura Technology), the hardest name on financial quality, rose only 3.88% that day. Both have been removed from the recommendation slots.

8. The catalyst grade of the top name was also downgraded by verification: in the full text of 光迅科技 (Accelink)'s 2026 interim report, "CPO" appears 0 times, "silicon photonics" 0 times, "1.6T" 1 time and "800G" 2 times, all of them product-line positioning language with no revenue disclosure whatsoever; the H1 growth driver is the telecom-side "transmission" +50.5%, while datacom-related "data & access" is only +16.0% with a 23.6% gross margin. Under the "concept ≠ beneficiary" grading rule, its beneficiary grade toward AI datacom should be "industry-trend catalyst, not named." It still ranks 1st, but the reason is fund structure and seal quality, not fundamentals.

9. Driver type: policy (day 2 of pricing) + overseas industry, not new orders. The "15th Five-Year Plan for the Electronic Information Manufacturing Industry" was released on 9/15 (outside the window); among the 1,499 filings inside the window there were 0 earnings pre-announcements and 0 price-hike announcements.

10. Pre-market state call: a day of disagreement. Suppressors (hawkish dot plot, the absolute 5% level on Treasuries, 沐曦股份's RMB 6.670 billion lockup expiry today = 75.38% of its float) coexist with supports (ASHR +0.12%, flat renminbi, SMH +0.64%, semiconductor materials 31 up 0 down, real foreign buying at the single-stock level). Low position size; let the call auction and the "open→close" adjudicate.

---

1. News Overview

⚠️ Three verification notes: (a) Two search summaries were discarded because of date mismatch. A search for "US close on September 16" returned an article saying "the Dow fell 0.63% to 52093.10, the Nasdaq fell 0.78% to 25981.57" — those two numbers are exactly the "previous close" measured on CNBC (Dow 52,093.11 / Nasdaq 25,981.571). That article was about the night of 9/15. Another claimed "all three indices rose, Intel up over 6%," which contradicts the measured data (Dow −1.21%, INTC +4.03%) in both direction and magnitude. (b) The "311.11%" in the 赛意信息 filing corresponds to RMB 6.45 billion, not RMB 6.72 billion. Back-computation: 2025 audited revenue RMB 2.07323 billion, 6.45÷2.07323 = 311.11% (to two decimal places), while 6.72÷2.07323 = 324.13%. (c) The "78%" debt ratio at 康惠股份 back-solves consistently, but implies an unexplained gap. Currently 13.72/19.86 = 69.06% (matching the "69%" in the filing); if the entire RMB 1.141 billion were debt-funded it would be 80.35%; back-solving "78%" implies about RMB 807 million of new debt, i.e. the remaining roughly RMB 334 million must come from own funds — yet cash on the books is only RMB 61.16 million, and the source is not explained.

---

2. Strongest Catalyst Branches, Descending

Ranking method (made explicit this issue per QC): step one ranks by breadth using "share of decliners in the industry"; step two adjusts using cross-market/overnight evidence, and adjustments can only lower a branch, by no more than 2 notches. Both steps are shown so readers can re-check.

⚠️ One important deduction for branch 1 (QC adopted): the draft said "4 limit-ups concentrated in the same sub-industry," but it is in fact 3 — 晶升股份(688478) makes crystal-growth equipment and belongs to equipment, not materials. The limit-ups in semiconductor materials are 有研新材, 有研硅 and 立昂微 — 3 names. The branch's "logic hardness" is accordingly cut from "medium-high" to "medium." ⚠️ One under-discussed counter-logic for branch 2: Huawei's in-house 7.2T optical module is simultaneously a substitution risk for third-party module makers (including 光迅科技), not purely a clear positive. This report has seen no evidence that the Huawei product is supplied by A-share manufacturers.

Two branches excluded from the branch list:

---

3. Overall Catalyst-Strength Ranking (22 names, descending by stock)

⚠️ Two rulers are used for the seal metric this issue (the single most important item adopted from QC): the draft used only "seal order/turnover value," and that caliber ≈ (seal order/free-float market cap) ÷ turnover rate, which mechanically rewards low-turnover names — 中瓷电子's "seal ratio 0.311, highest on the main theme" and "turnover 1.4%, lowest in the market" are not two pieces of evidence, they are the same number stated twice. This table adds "seal order/free-float market cap" (seal/float ratio) alongside, and the Spearman correlation between the two rulers is only 0.718, with the disagreement landing exactly on the recommendation slots. ✅ = financials independently verified by a sub-agent; ⬜ = public price/volume and filings only, unverified, score capped at 9 points.

---

4. Single-Stock Scoring Model (100 points)

⚠️ The total score does not determine the conclusion. The conclusion is jointly determined by "evidence grade + position + liquidity," so a high score can coexist with a conservative conclusion (e.g. 永鼎股份 scores 62 yet is "watch only," because PE is 157.95 and it sits late in the chain). This issue states that explicitly per QC, so readers no longer have to infer conclusions backwards from scores.

---

5. Detailed Analysis of the Top 10

1光迅科技002281Shenzhen main board ±10% | Optical comms | Priority deep-dive (single day only)

• Related news: 09-16 Huawei showed the industry's first 7.2T optical module and launched the world's first 3D data-center deployment; CAICT said "NPO commercial deployment is imminent." On top of that, the 9/15 plan calls for "consolidating the foundation of the photonics industry."

• Catalyst grading (under the "concept ≠ beneficiary" rule): industry-trend catalyst, not named. Huawei named no A-share supplier; and Huawei's in-house 7.2T optical module is simultaneously a substitution risk for third-party module makers, with no evidence in this report that the product is supplied by A-share manufacturers.

• ⚠️ Four findings from verification (all re-checkable):

  1. Word frequency across the full 2026 interim report: "CPO" 0 times, "silicon photonics" 0 times, "1.6T" 1 time, "800G" 2 times, and the latter two both appear in product-line positioning language ("products cover 400G, 800G, 1.6T and other rates"), with no 800G/1.6T revenue amount, shipment volume or share disclosed anywhere in the text.
  2. The growth engine is telecom, not datacom: H1 "transmission" RMB 2.283 billion, +50.5% (gross margin 28.8%); "data & access" RMB 4.310 billion, only +16.0% (gross margin 23.6%, the lowest of all segments). Domestic revenue is 65.6%.
  3. TTM ROE is only 9.7% (Innolight 65.3%, T&S 42.5%), while PE(TTM) is 132.6, PE 10-year percentile 97.1%, PB 10-year percentile 96.9% — the combination of "most expensive + lowest return on equity."
  4. Inventory RMB 9.070 billion (+57.8% from end-2025), already exceeding all of H1 revenue (= 137%), turnover 270.9 days; RMB 193.55 million of inventory write-downs were booked in H1, and the company's own text says this "accounts for 33.25% of attributable net profit for the first half of 2026"; H1 operating cash flow −RMB 1.236 billion.

• The two positives are stated in full as well: H1 revenue +26.1%, ex-non-recurring +57.8%, Q2 alone +27.0%/+55.5%, with cumulative and single-quarter pointing the same way; contract liabilities RMB 1.146 billion, +674% YoY, the only strong advance-payment signal among the three. Omdia caliber: fifth globally in optical components (5.1%), third globally in access optical components (8.3%).

• ⚠️ One factual error already corrected: the draft described Accelink as part of the "CETC system," which is wrong. Its controlling shareholder is Fiberhome Technology Group and its actual controller is China Information and Communication Technology Group (CICT, the Wuhan Research Institute of Posts and Telecommunications system). The CETC-system name is 中瓷电子, also on this list (CETC 13th Institute); the two do not belong to the same group. The correction does not change the "national team" characterization, but it removes the draft's implied association of "same system as 中瓷."

• Technical sentiment: on 9/16, limit-up +10.00%, turnover RMB 7.405 billion, turnover rate only 5.24%, volume ratio 1.45, seal order RMB 340 million, 0 limit breaks, first seal 10:35. ⚠️ But on the seal/float ratio it is only 0.233%, the lowest among the main-theme limit-up names — the absolute seal size is not small, it is just thin relative to its RMB 145.581 billion free float.

• LHB (the hardest evidence in this report, but the denominator must be stated clearly): net buy +RMB 1.015 billion, first in the market. After seat de-duplication (Shenzhen Stock Connect and the institutional-dedicated seats appear on both the buy and sell lists, already netted and de-duplicated, with the total of RMB 1.0150 billion reconciling to the LHB net buy):

  Seat	Net amount	Nature
  Shenzhen Stock Connect dedicated	+RMB 461.4 million	Northbound
  Institutional dedicated A / B / C	+RMB 125.8 / 92.0 / 35.2 million	Institutional, total +RMB 253.0 million
  Guotai Haitong, Wuhan Ziyang East Rd	+RMB 225.7 million	Hot money
  Lianchu Securities, Hangzhou Huancheng North Rd	+RMB 183.4 million	Hot money
  China Merchants, Shenzhen Shennan Ave Chegongmiao	−RMB 64.4 million	Hot money selling
  Guosen, Zhejiang Internet Branch	−RMB 44.1 million	Hot money selling

  Table scrolls horizontally

  Northbound plus institutional totals +RMB 714.4 million. ⚠️ The denominator of that "70.4%" is the net amount of the LHB's top-5 seats (RMB 1.015 billion), not turnover value — RMB 714.4 million is only 9.6% of the day's RMB 7.405 billion turnover. The correct statement is therefore "institutions and northbound dominate within the top-5 seats," not "the whole day's money was institution-led."

• Near-term calendar (eiTime checked): 9/21 (Monday) 14:30 extraordinary shareholders' meeting, to review the H-share issue and listing proposal; 12/16 the 6-month lockup on the other participants in the private placement expires (issue price RMB 167.55, a 10.5% paper gain versus the current 185.21). There was no new company filing at all on 9/16, the day of the limit-up.

• Final call: priority deep-dive, single day only, limited size. What supports it is seat structure and the cleanliness of the seal, not fundamentals — the fundamental verification skews negative. That RMB 461 million of northbound buying happened before the hike was announced and cannot be treated as evidence of foreign attitude today.

2北方华创002371Shenzhen main board ±10% | Semiconductor equipment | Watch closely ——【Upgraded this issue】

• The upgrade rests on three things holding at once, not just "good quality": the hardest verified quality + the smallest gain on 9/16 (+3.88%, against 有研新材 +9.99% and 中微 +6.65%) + the only positive order proxy variable.
• Verification highlights: ex-non-recurring/attributable = 99.13% (almost no non-recurring items); TTM operating cash flow RMB 9.098 billion = 1.60x TTM attributable net profit of RMB 5.684 billion, the only one of the four above 1, with H1 operating cash flow at +RMB 3.774 billion versus −RMB 3.191 billion in 2025H1, a YoY swing of RMB 6.965 billion; contract liabilities RMB 5.320 billion, +6.3% YoY, the only positive of the four; EV/ex-non-recurring TTM 86.6x, the lowest of the four; net cash RMB 5.215 billion; no equity-financing dilution.
• Where the positive expectation gap comes from (an accounting base effect, not an operating improvement): consensus 2026E attributable net profit is RMB 7.35 billion, which requires H2 to deliver RMB 3.98 billion, +72% against 2025H2's RMB 2.314 billion — aggressive on the surface. But 2025Q4 attributable net profit alone was only RMB 392 million (net margin 3.56%), and of the RMB 3.433 billion of 2025 full-year admin expenses, about RMB 2.376 billion fell in H2. With expenses normalized, H2 of RMB 3.98 billion ≈ about RMB 1.95 billion in Q3 (2025Q3 actual was RMB 1.922 billion, so only +1.5% is needed) + about RMB 2.03 billion in Q4.
• ⚠️ The counter-evidence is equally hard: operating leverage has already reversed (revenue +24.9% converted into only +5.0% ex-non-recurring, H1 gross margin 40.07%, −2.10pp YoY); all four valuation calibers sit at the 91st percentile of the past 3 years; overseas equipment peers were all down overnight.
• ⚠️ A methodological risk that must be flagged to readers: "good financial quality" has repeatedly been shown not to be a short-term ranking variable — in the 9/16 recap, 北方华创 was rated the weakest on fundamentals of the 5 recommendation slots, yet was the second-best performer. Upgrading it here is not "quality means it will rise," it is that it plays the low-volatility ballast role among the recommendation slots.

3有研硅688432STAR Market ±20% | Semiconductor materials | Watch closely (limit size)

• Branch stage: day 2 of the launch phase. Semiconductor materials on 9/16 were +6.14%, 31 up 0 down, flows +RMB 3.936 billion; limit-ups within the branch number 3 (有研新材, 有研硅, 立昂微).
• Technical sentiment: +20.00% limit-up, turnover RMB 3.914 billion, turnover rate 6.15%, 1 limit break, first seal 10:09; opened −1.11% lower (open 44.55 against the previous close of 45.05) and then pushed to limit-up — the entire gain was earned intraday, not handed over by an overnight gap.
• LHB (wording corrected this issue per QC): net buy +RMB 706 million. Listed buying seats: Shanghai Stock Connect net +RMB 461.7 million, Guotai Haitong Wuhan Ziyang East Rd +RMB 130.7 million (hot money), Guotai Haitong HQ net +RMB 43.3 million, J.P. Morgan Shanghai Yincheng Middle Rd +RMB 102.5 million, Goldman Sachs Shanghai Pudong Century Ave net +RMB 34.3 million.
  • ⚠️ The draft's "foreign-led, not hot money" fails on two counts: ① the second-largest buyer is a hot-money seat, so "not hot money" is a universal negation that cannot stand; ② the branches of J.P. Morgan and Goldman Sachs are brokerage seats of foreign securities firms, and the clients behind a seat are not necessarily foreign — domestic private funds and quants have long used such seats. Correct statement: among the top-5 buyers, Shanghai Stock Connect plus foreign-broker seats total about RMB 599 million, roughly 77% of the listed seats, but "the nationality of the money" cannot be inferred from seat ownership, so that characterization is marked pending verification.
  • ⚠️ The reconciliation check failed: the listed seats sum to RMB 772.5 million ≠ the net buy of RMB 706 million, a gap of RMB 66.5 million from unlisted selling seats. Accelink was reconciled and shown; this name was not — the same ruler should be applied to both, so the gap is disclosed here as is.
• ⚠️ The most important reservation: this name's financials are entirely unverified. What the sub-agents covered was 有研新材(600206), part of the same GRINM system, and that verification came out among the worst in the market. Same system does not extrapolate. Also: its PE(TTM) is 285.87.
• Final call: watch closely (limit size). The basis is foreign-seat buying and branch breadth; not one item comes from fundamentals.

4中际旭创300308ChiNext ±20% | Optical comms | Watch closely (ballast position) ——【Upgraded this issue】

• Why it was added: the original 4th recommendation slot was 中瓷电子, whose selection rationale, "seal ratio 0.311, highest on the main theme," does not hold under the other ruler (see §5⑦). This issue substitutes 中际旭创, on the grounds that its operating data is the hardest among the three verified optical-comms companies.
• Verification highlights: single-quarter revenue up sequentially for 8 straight quarters (RMB 4.843→…→19.496→22.281 billion), single-quarter gross margin up for 7 straight quarters (32.76%→…→46.42%); 26Q2 revenue RMB 22.281 billion, +174.6% YoY, ex-non-recurring RMB 7.374 billion, +206.4%; physical expansion with construction in progress at RMB 1.422→3.707 billion (+161%); a buyback of no less than RMB 4 billion and no more than RMB 8 billion, with a price cap of RMB 1,200/share, was disclosed on 2026-09-01 and executed for the first time; in an 8/23 investor-relations record the company said "major customers have already given 2027 guidance and orders."
• ⚠️ Three pieces of counter-evidence:
  1. PE 10-year percentile 50.1% vs PB 10-year percentile 93.2%, a 43.1pp divergence. The mechanism is that TTM ROE has reached 65.3% — the PE is low not because the price is cheap, but because the denominator E is at a historical high. If ROE reverts from 65% to 30%, PE doubles at an unchanged PB. But what this flags is asymmetric odds, not "the top is already in": the company says it already has 2027 order guidance, and the expansion is being paid for in cash.
  2. A cash-to-profit ratio of only 0.13 (H1 operating cash flow +RMB 1.800 billion vs attributable net profit RMB 13.651 billion), with Q2 single-quarter operating cash flow at −RMB 1.568 billion. The company attributes this to early stocking. Cross-check: H1 revenue +182.5% while the receivables balance rose +136.5% YoY, so receivables grew more slowly than revenue, which does not support "buying revenue by loosening credit terms."
  3. Turnover of RMB 21.471 billion on 9/16 was first in the market, but the volume ratio was only 1.04 — no incremental money, it is a game among existing holders.
• Final call: watch closely (ballast position). It is the only name on this list that is "verified strong on fundamentals and has a buyback price floor," but it is also the least likely to deliver short-term elasticity. Its role on the list is the same as 北方华创's: reducing overall volatility, not chasing gains.

5澜起科技688008STAR Market ±20% | Digital chips | Watch closely

• The positives from verification: core interconnect chips at RMB 3.111 billion in H1 (93.28% of the total, gross margin 69.29%), Q2 alone RMB 1.694 billion, +28.2% YoY / +19.5% QoQ; ex-non-recurring PE at the 62nd percentile of the past 3 years, the lowest of the four verified; debt-to-asset ratio 7.51%, net cash about RMB 17.9 billion; on an ex-non-recurring basis OCF/ex-non-recurring = 1.00x.
• One important clarification: the Q2 blended gross margin fell −7.96pp QoQ (69.79%→61.83%), which looks like deterioration, but back-computing by segment — the low-margin Jintide® line (gross margin 9.68%) saw volume roughly quadruple QoQ in Q2, and recomputing on the Q2 mix implies 63.5%, only 1.7pp away from the actual 61.83%. The vast majority of the gross-margin decline is explained by product mix; volume and pricing on the core lines have not deteriorated.
• The negatives from verification: of the +72.33% attributable growth, RMB 675 million comes from fair-value changes (RMB 430 million) plus investment income (RMB 252 million), while ex-non-recurring is only +21.17%; inventory +148%, turnover days 65→147; the 76th-percentile PB is depressed by roughly RMB 17.9 billion of net cash from the H-share IPO, and stripping that out PB is about 62x.
• Final call: watch closely. "A low percentile ≠ cheap" — its EV/ex-non-recurring of 100.3x is still above 北方华创's 86.6x.

6有研新材600206Shanghai main board ±10% | Semiconductor materials | Watch only ——【Downgraded from a recommendation slot】

• I changed my conclusion during the writing of this report, entirely on the basis of re-checkable financial data:
  1. H1 gross margin of only 7.42% — in financial profile this level is closer to metal processing/trading than to semiconductor materials.
  2. Revenue +50.35% but ex-non-recurring only +16.98% — operating leverage is negative, and the growth very likely comes from passing through metal prices rather than from volume/mix improvement.
  3. H1 operating cash flow −RMB 950 million (2025H1 was −RMB 225 million, so it deteriorated further), TTM operating cash flow −RMB 804 million, negative or near zero in all of the last 5 periods.
  4. Notes and accounts receivable +57.5% YoY (RMB 1.016→1.600 billion) — the revenue growth is being tied up in working capital.
  5. Valuation is at historical extremes on every dimension: PB at the 98th percentile of the past 5 years and the 99th of its full history; PS at the 99th of its full history; ex-non-recurring PE at the 95th of the past 3 years; EV/ex-non-recurring TTM 185.1x, the highest of the four.
  6. The only one of the four verified names with no sell-side earnings coverage — an RMB 43.7 billion market cap priced with no earnings model as an anchor.
• The price and volume are still good, no denying it: limit-up +9.99%, seal/float ratio 0.883%, second on the main theme, 0 limit breaks, first seal 10:07:38, turnover RMB 3.759 billion.
• But two things must be kept apart: "it sealed well yesterday" and "it deserves a recommendation slot today" are two different questions. When a limit-up happens at the 95th percentile of ex-non-recurring PE, the 99th percentile of full-history PB, and −RMB 800 million of TTM operating cash flow, putting it in a recommendation slot is betting on the best pattern with the worst margin of safety.
• One more note: its LHB seat detail was not obtained (the interface returned NoneType), so whether the buyers were institutions or hot money cannot be judged.

7中瓷电子003031Shenzhen main board ±10% | Optical comms | Watch only ——【Downgraded by QC this issue】

• This is a recommendation slot directly overturned by QC this issue, and the process is worth writing down: the draft listed it 4th, with the core rationale "seal ratio 0.311, the highest within the main theme."

• The problem QC pointed out: seal order/turnover value ≈ (seal order/free-float market cap) ÷ turnover rate, a caliber that mechanically rewards low-turnover names. 中瓷's "seal ratio 0.311, the highest" and "turnover 1.4%, the lowest in the market" are not two independent pieces of evidence, they are the same number stated twice.

• Measured results under the other ruler (seal order/free-float market cap):

  Stock	Seal order (RMB bn)	Free-float mkt cap (RMB bn)	Seal/float %	Original seal ratio
  永鼎股份(600105)	0.642	66.111	0.971 (1st on the main theme)	0.097 (near the back)
  有研新材(600206)	0.386	43.708	0.883	0.103
  东材科技(601208)	0.297	54.863	0.541	0.067
  星网锐捷(002396)	0.127	29.410	0.430	0.024 (judged "extremely weak" → Pass)
  中瓷电子(003031)	0.256	62.119	0.413	0.311 (originally judged "highest on the main theme" → recommendation slot 4)
  光迅科技(002281)	0.340	145.581	0.233	0.046

  Table scrolls horizontally

  The Spearman correlation between the two rulers is only 0.718, and the disagreement lands exactly on my recommendation slots: 星网锐捷 (0.430), which I passed on, and 中瓷电子 (0.413), which I recommended, are in the same tier under the new ruler.

• Two other misstatements in the draft: ① the draft called it "small in size," but its free-float market cap is in fact RMB 62.119 billion, larger than 有研新材's RMB 43.7 billion — the correct phrasing is "small turnover value" (RMB 825 million), i.e. extremely low turnover in a large cap, not a small cap; ② the volume ratio is only 0.57, meaning the day's turnover was below its recent average, and "holders unwilling to sell" can equally be read as "nobody trading," while the draft offered only the first reading.

• Final call: watch only. The 09:35 early seal and 0 limit breaks remain facts, but the number that put it into a recommendation slot does not survive a change of ruler.

8中微公司688012STAR Market ±20% | Semiconductor equipment | Watch only ——【Downgraded from a recommendation slot】

• At the core this is a caliber problem: H1 attributable net profit was RMB 2.825 billion, +300.22% YoY, but about RMB 1.982 billion of that comes from fair-value changes and investment income on outside equity investments — the company itself wrote this in its 8/4 pre-announcement ("an increase of about RMB 1.815 billion over the RMB 168 million in the first half of 2025"). Income-statement cross-check: fair-value changes RMB 992 million + investment income RMB 1.086 billion = RMB 2.078 billion. Stripping that out, core operating profit of the equipment business is about RMB 794 million.
• So the "low PE" is a trap: attributable PE(TTM) 79.8x, 60th percentile of the past 3 years; ex-non-recurring PE(TTM) 158.3x, 90th percentile of the past 3 years. That 30pp percentile gap is 100% manufactured by unsustainable investment income.
• Conflicting directions: contract liabilities RMB 2.772 billion, −12.4% YoY, down sequentially for three straight quarters (2025Q3 peak 4.389 → 3.044 → 2.853 → 2.772 billion) — the only one of the four verified names whose order-proxy variable is negative YoY. The valuation percentile is high at 90% while the forward indicator is falling.
• Two accounting-quality items: the capitalized R&D share went 25.2%→35.8%, with development-expenditure balance doubling in a year from RMB 1.004 to 2.106 billion; two placements in June lifted capital reserve by about RMB 4.19 billion net, so the current PB of 11.24x is built on freshly enlarged net assets, and restating to end-2025 net assets it is about 14.9x.
• ⚠️ An uncertainty that must be flagged: contract liabilities are only a proxy for advance payments and are affected by payment terms. Whether AMEC's interim report directly discloses an "order backlog" figure was not obtained this issue — if the backlog it disclosed is in fact growing, this judgment must be overturned. This is the single most falsifiable item in the report.

9港迪技术301633ChiNext ±20% | Oil & gas equipment automation | Watch only ——【Downgraded from a recommendation slot】

• Related news (09-16 18:14): several new sales contracts supplying energy-storage workover rigs and drilling-rig automation control systems, about RMB 100 million (incl. tax) / about RMB 88.5 million (excl. tax); customer names exempted from disclosure. That equals 18.73% (incl. tax) / 16.57% (excl. tax) of 2025 revenue of RMB 534.02 million.
• ⚠️ Three items that force a downgrade after verification:
  1. 2026H1 has already swung from profit to loss: revenue RMB 164.14 million, −21.87%; attributable net profit −RMB 24.18 million, ex-non-recurring −RMB 25.30 million (2025H1 was +RMB 18.42 / +15.00 million).
  2. Gross margin collapse: 43.96% (2025FY) → 28.14% (−15.0pp); within that, the intelligent control system line fell from 37.67% to 14.44%. Under different gross-margin assumptions, the gross profit on this RMB 100 million order ranges between RMB 12.8 million and 38.9 million, a threefold spread, and the product mix of the contract is not disclosed.
  3. The story has already been played once and blew up: after peaking at 66.80 on 9/7 it fell five sessions straight to 51.20 (−23.4%), with turnover value shrinking from RMB 209 million to RMB 37 million. This order announcement came after the slump.
• The oil & gas share cannot be confirmed: the periodic reports have only one line item by industry, "industrial automation" (99.4%). The company profile lists "petroleum" as one of its application industries, but the financial reports have no caliber for oil & gas drilling equipment revenue share, so it cannot be confirmed for now.
• Two positives: the cleanest balance sheet of the four (net cash +RMB 184 million, debt ratio 22.30%, interest-bearing debt ≈ 0); and the news genuinely is not priced in (volume ratio 0.61).
• Risks: receivables/TTM revenue of 78%, the highest of the four; free-float market cap RMB 1.346 billion and turnover of RMB 46.73 million, liquidity so poor that even a correct direction would be hard to execute at the expected price.

10力源科技688565STAR Market ±20% | Thermal-power environmental equipment | Watch only

• Related news (09-16 18:40): signed a chemical water-treatment equipment contract for the Weihai thermal-power 2×660 MW project with Northeast Electric Power Design Institute Co., Ltd. of China Power Engineering Consulting Group (named; a central-SOE design institute with registered capital of RMB 1 billion), worth RMB 68.156 million (incl. tax) / about RMB 60.31 million (excl. tax), equal to 23.96% / 21.20% of 2025 revenue of RMB 284.51 million.
• The "most solid order" of the four order-driven names: 100% of the order falls within its core line items (by industry, thermal power 84.1%; by product, condensate polishing 56.6% and demineralized-water treatment 40.0%); the buyer is named and checkable; contract liabilities RMB 263 million = 103% of TTM revenue; net cash +RMB 77 million.
• But "most solid order" ≠ "best fundamentals": ex-non-recurring attributable profit has been negative for four straight years (2022 −RMB 42.2 million / 2023 −88.5 / 2024 −27.6 / 2025 −12.0 million), turning positive for the first time only in 2026H1 at +RMB 1.33 million; revenue −20.88%; PE(TTM) 158x; ⚠️ the 2025 gross margin of the demineralized-water treatment business was −5.26% (a loss), so if this contract is mainly that category it may not be profitable; turnover of only RMB 22.28 million.
• ST risk has been checked: the current stock short name carries no ST/*ST prefix; the STAR Market financial delisting condition is "net profit (the lower of pre- and post-non-recurring) negative in the most recent year and revenue below RMB 100 million," and while the 2025 lower-of figure of −RMB 12.0 million is negative, revenue of RMB 284.51 million > the RMB 100 million threshold, so it is not triggered. The only path not yet closed off is the type of audit opinion for fiscal 2025, which was not obtained this issue and cannot be confirmed for now.
• Final call: watch only. Recorded for long-term testing of whether "evidence hardness" and "same-day gains" are correlated — the two are frequently uncorrelated.

---

6. Pass List (split into two buckets)

Conclusion from the 9/16 recap: "falsified Pass" produced same-day excess of −0.98pp (continuing to pass was correct); "position Pass" produced same-day excess of +1.65pp with 7/7 outperforming ("too high a position" does not mean "it will not rise"). This is the second consecutive time the same conclusion has appeared, so the list is split into two buckets this issue.

⑥-A Falsified Pass (primary evidence negates the catalyst itself — excluded)

⑥-B Position Pass (the catalyst may hold, but position/structure makes it unchaseable — cut size, keep in the watch pool, do not exclude)

⑥-C Not Participating (structural)

• Precious metals across the board: on 9/16, 13 up / 0 down, +3.83%. ⚠️ The exclusion reason has been softened this issue — overnight GLD was only −0.61% and SLV −0.83%, and on the day of the first hike plus a 10Y above 5%, gold fell only 0.6%, which is actually on the strong side, so it does not constitute a strong falsification. The real reason is position: 13 up 0 down consistency is already an extreme.
• Rubber sector: on 9/16, 23 up / 0 down, +3.08%, but the natural-rubber futures front contract RU2701 traded at RMB 18,845/tonne in the night session, essentially flat against the previous close of 18,875 (−0.16%) (the tradedate/position/volume fields were checked and the contract is active). Stocks all up while the commodity does not move = rotation of money, not fundamentals.
• Oil chain: see §2.
• Banks: on 9/16 A-share banks were 4 up / 37 down (88.1% decliners), and overnight US XLF was −1.62% and KRE −1.77% — bank stocks fell on a hike day; both markets point the same way, with no evidence of a reversal.

---

7. Intra-Branch Ranking

Branch 1 · Semiconductor materials/silicon wafers (31 up / 0 down, +6.14%)

The single most important line for this branch: the two names with the highest trading identifiability are one with the worst fundamentals and one with unknown fundamentals. The breadth (31 up 0 down) is real, but not one name's fundamentals can support it.

Branch 2 · Optical comms/optical modules (comms equipment 81 up / 8 down, CPO flows +RMB 15.720 billion)

The single most important line for this branch: the strongest fundamentals (中际旭创) come with the lowest trading identifiability, and the highest trading identifiability (光迅) comes with negative-leaning fundamentals. This is the classic structure of "the back end of a branch" — money is broadening out rather than adding to positions. The approach this issue is to take one name from each end: 光迅 for elasticity, limited to a single day, and 旭创 as ballast.

Branch 3 · Semiconductor equipment (25 up / 0 down, +5.03%)

The shared risk across the whole branch: overnight, US equipment peers AMAT −1.37%, LRCX −0.61% and KLAC −0.39% were all down. This branch gets no "priority deep-dive."

---

8. Today's Opening Verification Signals

Hard rule: every criterion must first come with a base rate. Measured "day after a limit-up" base rates this issue (official limit-up pool → next day):

Limit-up day → next day	Valid sample	Next-day green-close rate	Next-day mean gain	Mean open→close	Share with positive open→close
9/08 → 9/09	48 (pool 73, 25 missing)	52.1%	+1.30%	−0.58%	43.8%
9/11 → 9/14	40	72.5%	+3.21%	+0.94%	57.5%
9/14 → 9/15	54	50.0%	+0.68%	−2.06%	27.8%
9/15 → 9/16	32	71.9%	+3.65%	+0.83%	50.0%
Equal-weighted mean		61.6%	+2.21%	−0.22%	44.8%
Pool-weighted (n=174)		59.8%			43.1%

Table scrolls horizontally

⚠️ A methodological error was corrected this issue per QC. The draft used the equal-weighted mean of 61.6% > 60% to scrap the "next-day green close" criterion outright. But the pool-weighted figure is 59.8%, on the other side of the 60% threshold; and at n=174 the 95% confidence interval around 61.6% is [54.4%, 68.8%], which straddles the threshold. Throwing a point estimate at a hard threshold is not statistically valid. The corrected rule: a criterion is discarded only when the point estimate and its entire 95% confidence interval sit above 60%. On that basis, "next-day green close" is not discarded, but is demoted to a secondary criterion; the primary criterion remains "positive open→close" (equal-weighted 44.8% / weighted 43.1%), because its range is far wider (27.8% to 57.5%) and therefore more discriminating.

① Call-auction signals (09:15–09:25)

② Sector signals

③ Single-stock signals

④ Risk signals

⚠️ The trigger rule was restructured this issue per QC. The draft's "any two of six" had a problem: items 1/2/4/5 all fire together on a poor-sentiment day, which is equivalent to "any one." It is now changed to: at least one item from the 【 sentiment dimension 】 and at least one from the 【 structural dimension 】 must hold simultaneously.

【 Sentiment dimension 】

1. The market-wide share with a positive open→close is < 40% (base 44.8%; on 9/14→9/15 it was as low as 27.8%)
2. The limit-up count falls from 89 to < 40, with the largest thematic cluster below 4 names
3. The promotion rate of the 89 names that hit limit-up on 9/16 is < 20% today (on 9/16 it was 37.5%, on 9/15 13.0%). ⚠️ This threshold rests on a sample of only n=2, with the two observations straddling the threshold, so its reliability is low

【 Structural dimension 】 4. The core names do not follow: two of the three 光迅科技(002281), 北方华创(002371) and 有研硅(688432) turn red (i.e. decline) 5. FX turn: USDCNH breaks above 6.75 (9/16 was 6.7105) — this would directly overturn the "foreign buying" premise for two names on this list 6. 沐曦股份(688802) falls on expanding volume, dragging the domestic GPU/AI chip direction lower

---

9. Final Recommendation

① The 5 names most worth watching today

✅ One positive commonality (a direct response to the top criticism in the 9/16 recap) The heaviest conclusion of the 9/16 recap was that "almost all of the recommendation group's excess came from the overnight gap; executed at that day's open, the return relative to the index was 0." Measured gaps for this issue's 5 names:

Stock	Previous close	Today's open	Gap	Close	Open→close
光迅科技(002281)	168.37	169.00	+0.37%	185.21	+9.59%
北方华创(002371)	638.16	638.20	+0.006%	662.89	+3.87%
有研硅(688432)	45.05	44.55	−1.11% (gap down)	54.06	+21.35%
中际旭创(300308)	864.01	869.02	+0.58%	907.80	+4.46%
澜起科技(688008)	187.83	188.30	+0.25%	199.89	+6.16%

Table scrolls horizontally

All 5 gaps are ≤ +0.58%, and 有研硅 even gapped down −1.11% before running to +20% — the gains were almost entirely earned intraday. That is the opposite profile from the 9/16 list. But this only proves "yesterday's gains were executable"; it does not imply "the same will be true today" — today's gap depends on the pricing of the hike, and that happened only after yesterday's close.

⚠️ Three self-imposed constraints: (a) Concentration: 2 of the 5 were limit-up names on 9/16 (Accelink +10.00%, GRINM Semiconductor Materials +20.00%), and the base rate for "positive open→close" the day after a limit-up is only 44.8%, so those two must be limited to a single day; the other three (北方华创 +3.88%, 中际旭创 +5.07%, 澜起 +6.42%) did not hit limit-up and serve as ballast. This issue has cut the number of slots bet on the same pattern from 3 to 2 per QC. (b) Uneven evidence grades: 1 of the 5 (有研硅) is entirely financially unverified, and should be sized lower accordingly. (c) Independence of verification points: the 5 criteria respectively test — chip-loosening (Accelink), cross-market divergence (Naura, Montage), whether the evidence itself is overturned (GRINM Semiconductor Materials), and incremental money (Innolight). This issue has fixed the "three criteria testing the same variable" problem, but readers should still weight the "cross-market divergence" and "evidence overturned" types most heavily.

② The 3 strongest branches today

③ Directions not to chase today

1. 沐曦股份(688802, STAR Market) and the domestic GPU direction — today's lockup expiry of RMB 6.670 billion = 75.38% of its existing float, 89% of the market-wide unlock for the day, while its free-float market cap is only RMB 8.849 billion. This is the only purely computable negative this issue that depends on no judgment at all.
2. The two cross-industry compute names — 康惠股份(603139, Shanghai main board): RMB 61.16 million of cash to fund an RMB 1.141 billion purchase (18.7x), and it was net repaying debt rather than stocking up in 2026H1; 赛意信息(300687, ChiNext): the increment is only RMB 270 million, annualizing to 2.60% of revenue, while the market already rejected the larger RMB 6.45 billion contract on heavy volume on its 8/20 disclosure day (closing down 0.46%, −7.1% over three days).
3. The substrate theme at 崇达技术(002815, Shenzhen main board) — in substance it is an RMB 21.32 million land payment, the project was announced back in January, and the share price has already front-run it (9/16 +7.94%).
4. 闽东电力(000993, Shenzhen main board) — 6 consecutive limit-ups, the highest count in the market, but 20 limit breaks, seal ratio 0.024, PE 494.81, institutional seats net selling, and heavy selling from Huatai Raffles City, with no second-highest-count name in the branch to carry the baton.
5. The oil chain and rubber — oil prices gave back overnight (BNO −2.91%, XLE −2.88%), A-share oil-industry flows have been in continuous net outflow; the rubber sector was 23 up 0 down while natural-rubber futures barely moved (−0.16%).
6. Precious metals — downgraded to "not participating" rather than "bearish." Overnight GLD was only −0.61%, and on the day of the first hike plus a 10Y above 5%, gold fell only 0.6%, which is actually on the strong side. The reason to avoid it is the A-share position of 13 up 0 down, not overseas falsification.

④ The one-line final call

The hike itself was more than ninety percent priced before the meeting, and the only genuinely new arrival was the hawkish dot plot; meanwhile the overnight tape delivered a split that runs counter to the "tightening suppresses growth" intuition — banks fell, energy fell, while semiconductors and tech closed higher, the A-share proxy ASHR even closed slightly green and the offshore renminbi barely moved, so what most needs defending today is not an index-level catch-down but evidence collapse at the single-stock level: fundamental verification this issue has already pulled the two biggest 9/16 gainers (有研新材 +9.99%, 中微 +6.65%) out of the recommendation slots, because they happen to be the two with the worst financial quality and the heaviest accounting cosmetics, and QC pulled a third (中瓷电子) as well, because its supporting "highest seal ratio on the main theme" turns into mid-pack the moment the ruler is changed; of the remaining five slots only two are day-after-limit-up names, and the historical win rate for a positive open→close the day after a limit-up is only 44.8%, while one more (有研硅) is a completely unverified financial black box — so the correct posture today is low position size, letting the call auction and the "open→close" rather than "a green close" adjudicate, and putting the highest weight on the two genuinely independent criteria: whether the foreign seats in 有研硅 turn around, and whether 北方华创 can beat the semiconductor-equipment industry mean.

---

10. QC Items Adopted This Issue

Before finalization this report was independently QC-ed by risk-auditor, which raised 28 red flags. The following 14 were adopted and the main text was revised accordingly (those not adopted are mostly items already flagged as blanks in the text, or items that require intraday data to handle). They are listed because these revisions changed the recommendation conclusions themselves, and readers are entitled to know which parts were rewritten after being overturned.

Main items not adopted but flagged in the text: ① §8②'s constituent caliber is not fixed (flagged as "the close re-check must follow the names listed in §7"); ② the FOMC primary source was not obtained (marked "pending verification"); ③ ex-tax amounts and revenue bases have been added for the "earnings elasticity" of order-driven names; ④ "the methodology flipped within two days on n=1" — this criticism is valid but was not acted on this issue, because acting on it requires ≥5 trading days of parallel samples; it has been logged as a to-do.

---

⚠️ Data-collection failures and open items this issue (not for clients)

1. push2his.eastmoney.com was rate-limited during the base-rate computation: after 290 consecutive kline requests it turned to RemoteDisconnected. Bisecting the parameters confirmed this is not a parameter problem: push2delay with the same parameters returns HTTP 200 but klines is always 0 entries (confirming that the kline interface has the same host difference as daykline), and push2.eastmoney.com returns non-JSON. The base rate was instead assembled from three channels: 9/15→9/16 used the already-saved market-wide snapshot (zero requests), 9/11→9/14 and 9/14→9/15 used Tencent web.ifzq.gtimg.cn, and 9/08→9/09 used the batch collected before the rate limiting (only 48/73, gap not filled, flagged in the main text).
2. Some LHB seat details were not obtained: stock_lhb_stock_detail_em returned TypeError: NoneType for 有研新材(600206) and 永鼎股份(600105), while working normally for 002281/688432/000993. The main text states this explicitly and weights those items down.
3. Concept clist pagination gap: m:90 t:3 has total=504 but only 100 entries were retrieved. Because it is sorted descending by fid=f62, the TOP20 net-inflow list is complete; the "net outflow TOP10" was in fact the tail of the first 100, so that table has been deleted and not used.
4. Industry secid collision: in a9.py, "optical & optoelectronics" and "comms equipment" mistakenly used the same 90.BK0448. All industry breadth figures in the main text (including comms equipment 81 up/8 down) come from a5.py — grouped and counted from the market-wide snapshot's SW industry field f100, independent of BK secid and unaffected by that bug (this clarifies the concern in QC item 22). That script also had three failed secid requests, and its multi-day flow data was not used.
5. Unit trap in the lockup interface: LIFT_MARKET_CAP is in units of RMB 10,000 and CURRENT_FREE_SHARES is in units of 10,000 shares. Validated by back-computation: "13.966 million shares × RMB 477.56 ≈ RMB 6.67 billion."
6. ⚠️ Four self-interceptions / interceptions this issue:
   • (a) Nearly wrote "semiconductors were confirmed overnight" as a pillar — after pulling the 25 SOX constituents, the equal-weighted mean was −0.03% (15 up / 10 down), and AMAT/LRCX/KLAC, the peers of the strongest A-share equipment sub-industry, were all down.
   • (b) Nearly wrote "the Saudi supply cut pushed oil prices up" as a branch — four ETFs (USO −3.52%, BNO −2.91%, XLE −2.88%, OIH −2.83%) consistently show about 3% of give-back on 9/16, corroborated by @LCO.1 at 105.83 against 108.75 on 9/15 (−2.68%). The −0.48% on @CL.1 uses a rolling intraday previous close.
   • (c) The draft's recommendation slots were overturned wholesale by fundamental verification — 有研新材, 中微公司 and 港迪技术 were all downgraded out of recommendation slots, with 北方华创 and 澜起科技 added. Without running those three fundamentals-analyst passes, the list would have put the two worst names in the top five, and they happen to be yesterday's biggest gainers. On price and volume they genuinely are the best; the flaw shows only in the financial statements.
   • (d) ⚠️ The most serious one was caught by QC, not found by me: the precious-metals line "overnight −1.88%, a direct reverse falsification" used a futures reading based on a rolling previous close; on the ETF caliber it was actually −0.61%. In the same report I ran the rolling-previous-close check on @CL.1 and wrote it into the internal record, but I did not run the same check on gold and silver — and the gold/silver line was my sole basis for excluding an entire branch. "The same ruler must be applied to every object being compared" — I applied it to oil and not to gold.
7. One factual error corrected by a sub-agent: the draft attached 赛意信息's "311.11%" to RMB 6.72 billion. Back-computation: 2025 revenue RMB 2.07323 billion, 6.45/2.07323 = 311.11% (to two decimal places), so RMB 6.72 billion should be 324.13%.
8. One entity-attribution error corrected by QC: 光迅科技 is not part of the CETC system; its controlling shareholder is Fiberhome Technology Group and its actual controller is China Information and Communication Technology Group. This kind of "status fact" is self-consistent throughout a text and will never expose itself, so the registry must be checked.
9. Items the sub-agents marked "pending verification" are all left blank in the main text as such: the true issue price and total proceeds of AMEC's June placement (two akshare records conflict, determining whether PB is 11.24x or 14.9x); whether the interim reports of AMEC and Naura directly disclose an "order backlog" figure (if AMEC disclosed growth, the §5⑧ judgment must be overturned, as stated explicitly in the main text); Montage's actual H-share price and the A/H premium-discount; the type of audit opinion for Liyuan's fiscal 2025 (the only ST path not yet closed off); and 有研硅, 中瓷电子, 永鼎股份 and 12 names in total are entirely financially unverified.
10. Not obtained / explicitly left blank: (a) LHB seats for 有研新材 and 永鼎股份; (b) daily northbound totals (disclosure discontinued); (c) among the 1,499 filings in the window there were 0 earnings pre-announcements and 0 price-hike announcements (9 groups of keyword regexes were run over the full set, so this is a fact, not a miss); (d) the counterparty names for 港迪 and 康惠 (actively exempted from disclosure); (e) 港迪's oil & gas drilling revenue share (no such caliber in the financial reports); (f) the full text of the FOMC statement; (g) about 60 of the 89 limit-up names on 9/16 were not attributed.
11. QC conclusion: 28 red flags, 14 adopted and revised (see §10 in the main text). Of the 12 that QC judged would change conclusions, 11 have been handled; the 1 outstanding item is "the methodology flipped within two days on n=1" (the basis for branch ranking was changed twice between 9/16 and 9/17). That criticism is valid, settling it requires ≥5 trading days of parallel samples, it has been logged as a to-do, and it is not resolved this issue.

---

⚠️ Risk warning: this list is pre-market information gathering 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 industry-chain mapping errors; it must not be used directly as a basis for trading.

Data sources: filings within the window (East Money np-anotice-stock, publication moment by eiTime, 1,499 entries) | filing originals (East Money/CNINFO PDFs read directly, with links attached item by item in the text) | market-wide snapshot and industry/concept fund flows (East Money push2delay, 5,915/5,915 with no pagination gap; industry breadth grouped independently by the SW industry field) | exchange-caliber limit-up/limit-down/limit-break pools plus seal orders and free-float market cap (East Money push2ex, validated with tc == len(pool): ZT 89, DT 4, ZB 11) | LHB and seat detail (akshare stock_lhb_detail_em / stock_lhb_stock_detail_em, 2026-09-16, seats de-duplicated across the buy and sell lists and reconciled to net buy value) | single-stock price/volume and volume ratio (Tencent Finance qt.gtimg.cn, 2026-09-16 closing settlement values, timestamps checked name by name) | financials and valuation (East Money stock_profit_sheet_by_report_em/stock_balance_sheet_by_report_em, 2026 interim-report caliber, single quarters derived by subtracting cumulative figures and cross-checked against the segment tables; gross margin back-computed from OPERATE_COST; valuation percentiles from stock_value_em and Baidu Stock 10-year series; consensus from stock_profit_forecast_ths; all PE figures are TTM) | lockup-expiry calendar (East Money RPT_LIFT_STAGE) | overnight US equities/ETFs/commodities/Treasuries/FX (CNBC quote-html-webservice, last_time trade timestamps checked symbol by symbol) | commodity-futures liveness (akshare futures_zh_realtime, tradedate/position/volume checked) | previous-session reconciliation baseline: reports/a-share/2026-09-16-recap.md in this repository