A-Share Market Recap | 2026-09-15 Tuesday

Data read moment: 2026-09-15 15:30 (post-close settled values; Tencent quote field 30 timestamp verified name-by-name as 20260915153xxx). Index and turnover caliber: Shanghai + Shenzhen markets, at the close. Market-wide advance/decline counts are self-computed after paginating the full 5,915 names, item-by-item consistent with the Jiemian News closing wrap caliber. Limit-up / limit-down / failed-seal counts always use the official exchange-caliber pool (tc == len(pool) verified, no silent truncation). Price limits: SSE/SZSE main board ±10%｜ChiNext (30x)/STAR Market (68x) ±20%｜BSE ±30%. Every single stock in this report is tagged with its board.

---

0. One-Line Recap

1. Sentiment temperature: breadth is at freezing point, money is in single-point euphoria. Market-wide 1,120 up / 4,376 down (advancers only 20.2%), mean change −1.37%, median −1.50% — a textbook broad-decline day; yet the STAR 50 rose +1.55% against the tide, the only one of the four major indices to close green, and out of 90 industries only 3 recorded net capital inflow, of which the semiconductor line alone took RMB 4.461 billion (East Money main-force caliber), 9x the second-place name. This is not "the market fell," it is "all the money crowded into one line while the other 4,376 names bled out."

2. The single most central call in the pre-market list was falsified, and that has to be said first. The pre-market note wrote "the AI hardware chain carries the greatest absorption risk today" into §0.4 and ranked cybersecurity as branch No. 1. The actual outcome reversed on both counts:

   • AI hardware carried no absorption risk — the Philadelphia Semiconductor Index fell −5.86% overnight, but A-share semiconductors today were +1.34% (121 up / 64 down), main-force net inflow +RMB 4.461 billion (highest in the market); 中际旭创 (Zhongji Innolight, 300308, ChiNext ±20%) opened +0.04% and 新易盛 (Eoptolink, 300502, ChiNext ±20%) opened +0.02% — there was no gap-down at all, so the observation point set in pre-market §8, "can it recover after gapping down 3%~5%," never even had its premise satisfied.
   • Cybersecurity never became a leading theme — the cybersecurity concept was −0.91% (36 up / 96 down) for the full session, software development −1.25%, IT services −1.43%, computer equipment −1.37%; all three carrier industries fell and all three saw net capital outflow.

3. But one pre-market call was right, and it was the most counterintuitive one in that same piece — it must be given full credit. The pre-market note retracted its first-draft conclusion that "A-share optical modules will most likely see a second leg down today," on the grounds that the real cause of Corning's (GLW) −13.70% was its own $2 billion ATM equity offering (a dilution event, with Q3 guidance maintained), and that "a U.S. company's share issuance carries no demand-side read-through for Chinese optical module makers." Today 中际旭创 was −1.03% and 新易盛 −1.22%, flat opens and slight declines — the second leg down indeed did not happen. That retraction saved the conclusion of that entire section.

4. The honest read on hit rate: on the verification-point caliber, all 3 "watch closely" slots failed outright. 电科网安 (CETC Cyberspace Security, 002268), 三六零 (360 Security Technology, 601360) and 安恒信息 (DBAPPSecurity, 688023) all had "can it hold after gapping up" written as their verification point; today all three gapped up +1.78%+1.99% and all three closed below their open (open→close −2.62%−2.93%). The 4 executable names (excluding 中新赛克, which the pre-market note explicitly flagged as "not a buy candidate") would have lost an average of −2.21% if executed as listed at the open.

5. The most painful line: the Pass group beat the recommendation group, and systematically so. The watch-only group (10 names) averaged +2.37% with 9/10 closing green; the Pass group (15 names) averaged +2.62% with 8/15 green; while the 5-name recommendation list averaged +1.68% and all of that gain came from 中新赛克, the one name explicitly excluded from buying — the 4 executable names were −0.40% on a prior-close basis. All three groups beat the market-wide −1.37% — which says the pre-market note found the right "stock pool" but got the intra-pool ranking completely inverted.

6. The two genuine leading themes today, neither of which appeared anywhere in the pre-market note: wind power equipment +4.08% (26 up / 5 down, highest in the market, 4 limit-ups) and the CCL / copper foil / fiberglass price-hike chain (华正新材 limit-up, 宏和科技 +5.79%, 诺德股份 limit-up). For the latter the pre-market note actually derived the causality correctly (§7 explicitly wrote "upstream 生益's gross margin is +9.1pp over two years; downstream names are the cost absorbers"), yet landed the conclusion on "watch only," and never put a single genuine upstream name into any of its tables.

7. Tomorrow's baseline: follow the money, not the narrative. Semiconductors are the only direction with money of meaningful magnitude behind them and have now taken net inflow on consecutive days; wind power is strong on price, weak on money, unconfirmed on attribution — a one-day move; cybersecurity has entered a divergent state of "price down, money in"; agricultural planting and retail have entered genuine withdrawal (the two together account for 48% of the limit-down pool).

---

1. Market Overview

• Two-market turnover was RMB 1,612.717 billion, down RMB 16.459 billion from the prior session (Jiemian News closing-wrap caliber; digit-for-digit consistent with the RMB 1,612.71 billion obtained by summing Tencent index turnover locally). The full-market caliber including the BSE is RMB 1,625.3 billion (the local pagination self-count and the Tonghuashun 90-industry sum are mutually consistent, at RMB 1,625.28 billion and RMB 1,625.26 billion respectively). Three independent routes close on each other.
• Advance/decline counts (self-computed; full market of 5,915 names fully paginated, len==total check passed; after removing 367 suspended names and names with no opening price, 5,548 valid): up 1,120 (20.2%) / down 4,376 (78.9%) / flat 52. Name-for-name consistent with the "1,120 advancers, 4,376 decliners" published in the Jiemian News closing wrap.

The caliber difference must be spelled out: the official limit-down pool has 25 names, the Legulegu snapshot gives "34 limit-down / 33 genuine limit-down (of which 6 are ST names)," and the Jiemian News closing wrap uses 33. The gap comes from differing treatment of ST price limits and the BSE. The body of this report uses the official pool figure of 25 throughout; where 33 is mentioned, the source is labeled.

Sentiment temperature verdict: breadth at freezing point, structure extremely polarized. Three things hold simultaneously: (a) advancer share 20.2%, four out of every five names down; (b) failed-seal rate up to 42.9%, close to "one seals, one breaks"; (c) yet the mean "open→close" of the top 50 names by turnover is +1.04% with 29/50 green — large-cap, high-liquidity names were being bought today; the bleeding was in the small- and mid-cap tail. This is not a release of systemic risk, it is a violent re-concentration of style.

---

2. Pre-Market List Reconciliation

① The 5 recommended names (pre-market §9①)

⚠️ 润泽科技 requires an ex-rights note, otherwise two change figures will contradict each other. The stock's ex-dividend/ex-rights date is today (2026-09-15) (record date 9/14, 2026 interim distribution, dividend of RMB 0.4288 per share, distribution implementation announcement eiTime 2026-09-08 18:31:47).

• The exchange-caliber prior close has been adjusted for the distribution to RMB 60.25 (check: 60.68 − 0.4288 = 60.2512, digit-for-digit consistent with the 60.25 given by the quote interface), which gives today's change as +1.10%, with the RMB 61.25 open being a gap-up of +1.66%.
• If you simply compare today's close of 60.91 against the Monday close of 60.68 cited in the pre-market list, you get +0.38% — that number is not wrong, but it omits the RMB 0.4288 cash dividend. A holder's actual return is (60.91 + 0.4288 − 60.68) / 60.68 = +1.10%, consistent with the exchange caliber.
• Incidentally, this is an omission in the pre-market note: pre-market §8 specifically laid out a "calendar for the next 5 trading days," yet did not include 润泽科技's ex-dividend/ex-rights date today, even though the distribution implementation announcement had already been published on 9/8. An ex-rights date changes both the prior close and the percentage-change baseline, which makes it an event that must go on the calendar. | 中新赛克 (Sinovatio, 002912) | SZSE main board ±10% | Watch only · sentiment gauge, explicitly not a buy candidate | +10.02% | +10.02% | 0.00% | Limit-up · 4 consecutive, seal broken 5 times intraday | "Can the one-word board hold" | Passed in form, changed in substance (see below) | Delivered in form |

中新赛克 has to be discussed separately — it is the single most informative name today. The pre-market note designated it "the thermometer for the whole line," and that designation was right, but the thermometer's reading is not "it held," it is "a different crowd is doing the sealing":

Conclusion: it did not "open and then fall quickly" (the failure condition set pre-market), so literally no bearish signal was triggered; but "one-word locked float" has turned into "high-turnover relay" — the float completed a full round of turnover today. Tomorrow it is a divergence point, not an acceleration point.

Hit rate — all four calculations laid out, with a statement of which one is actually binding:

The binding ones are the last two rows. "Beat the market 5/5" is a flattering number with no meaning — the market-wide mean today was −1.37%, and any high-attention, high-liquidity name would beat it (the top 50 by turnover also averaged +0.72%). The verification points, by contrast, were falsifiable conditions written down by the pre-market note itself, and they passed 0/3 — that is this list's real score.

The executable caliber of −2.21% is the single most important number in this reconciliation. All 4 names gapped up +1.66%~+1.99%, meaning the opening price was precisely one of their worse entry points of the day; looking only at "change on the day of −0.40%" systematically understates the real cost of executing as listed, the two differ by 1.81 percentage points.

② The design flaw in the verification points themselves (this matters more than the hit rate)

The three "watch closely" slots shared one and the same yardstick — "can it hold after gapping up." Measuring that yardstick's market-wide base rate today, in hindsight:

• Names that gapped up today (open > prior close): 1,707;
• Of those, the ones that "held after gapping up" (open→close > 0): only 297 = 17.4%;
• The gap-up group's mean "open→close": −1.72%.

In other words, this yardstick had a market-wide pass rate today of only 17.4%. That 电科网安, 三六零 and 安恒信息 failed it carries almost no stock-level information — what it measures is the day's market environment, not the merits of these three names. The pre-market note tied three recommendation slots to a single criterion, and put 4 of its 5 slots on a single branch, which means the list had only one independent variable.

And the yardstick is also directionally misleading: grouping all 5,548 names by opening gap, "open→close" does indeed deteriorate monotonically as the gap-up widens (gap-down group −0.32% → gap-up >3% group −2.84%), but the absolute return on the day runs exactly the other way — the 51 names that gapped up >3% averaged +2.90% on the day with 64.7% green, the best-performing group in the entire market. Pearson correlations: corr(opening gap, open→close) = −0.157, corr(opening gap, change on the day) = +0.248 (n=5,548, opposite signs). "Gap up and fade" and "lose money on the day" are not the same thing, and the pre-market note treated them as the same thing.

③ Watch-only group / Pass group reconciliation (pre-market §3, §6)

Group statistics (equal-weighted, excluding 香山股份, which is a branch representative rather than a list holding):

Two counting-caliber disclosures, so the numbers are not made to look better than they are:

1. 中新赛克 (002912) belongs to both "the 5 recommended names" and "the watch-only group" (pre-market §3 gave it exactly the verdict "watch only," while also putting it on the §9① recommendation list as a sentiment gauge). Its +10.02% is the maximum value in both groups, so double counting inflates both at once. The table above therefore also gives the 9-name caliber excluding it, at +1.52% — that is the watch-only group's real score free of double counting, still above the executable recommendation group's −0.40%, so the direction of the conclusion is unchanged.
2. 任子行 (300311, ChiNext ±20%, −4.43%) is not counted in any group: pre-market it was only named as a "follower" in §7, and it was on neither the recommendation list nor the Pass list. Had it been counted in the Pass group, that group's mean would fall from +2.62% to +2.18%, and the direction of the conclusion would likewise be unchanged.

④ One-line self-critique: what was called right, what was called wrong, and how to change the next pre-market note

Three things called right (they cannot be erased just because the total score looks bad):

1. Retracting the "second leg down in optical modules" call was the most valuable judgment of this cycle. Decomposing Corning's decline into "a dilution event from its own share issuance" rather than "the AI narrative," and retracting the extrapolation on that basis — today 中际旭创 −1.03%, 新易盛 −1.22%, both opening essentially flat: the conclusion was completely correct.
2. The identification of "who is following" was precise. The pre-market note named 亚信安全 (688225), 绿盟科技 (300369) and 任子行 (300311) as follower names; today the three were −2.48% / −4.86% / −4.43%, all down with open→close of −6.16%~−7.08%, 3/3 correct.
3. Refusing to treat 凯盛新能 (Kaisheng New Energy, 600876, SSE main board ±10%) as "upstream in the CCL price-hike chain" was right. The QC pass had argued it corroborated the §7 reasoning; the pre-market note left it blank on the grounds that "it is photovoltaic glass, not electronic-grade fiberglass cloth." Today 凯盛新能 was −6.06% and broke its 3-consecutive streak, while the genuine electronic-grade fiberglass cloth name 宏和科技 (Honghe Technology, 603256, SSE main board ±10%) was +5.79%. Refusing that false corroboration avoided a piece of mistaken self-confirmation.

Three things called wrong, and specifically how to change them:

1. "Fundamental quality" was used as a short-term ranker. The pre-market note ranked the list from best to worst using ex-non-recurring profit, gross margin, goodwill and cash flow, and the ranking came out completely inverted: the names with the worst fundamentals, whose concepts their own companies had denied by announcement (闽东电力, 澳弘电子, 博汇科技), hit limit-up, while the relatively most solid ones (安恒, 电科网安) fell.

   • Fix: fundamentals should only be used to set the position-size cap and holding period, and must not participate in same-day direction ranking. Direction ranking must start from the money set (sector main-force net inflow + industry distribution of the limit-up pool), and only rank by fundamentals within that set. Had the money been looked at first today — semiconductors +RMB 4.461 billion, PCB +RMB 3.918 billion, storage +RMB 2.881 billion, cybersecurity only +RMB 1.410 billion — branch No. 1 would not have been cybersecurity.

2. "A company's own risk-warning announcement" was treated as the hardest reason to avoid, and it has close to zero predictive power for next-day direction. Pre-market §0.6 listed the risk warnings / concept denials from six companies on the evening of 9/14 as "a clear signal of regulatory cooling." Measured today (including 双星新材, which the pre-market note did not cover):

   Stock	Board (limit)	9/14 evening announcement (eiTime)	Result today
   闽东电力 (000993)	SZSE main board ±10%	19:45 abnormal fluctuation + 23:12 risk warning, denied compute-power/grid synergy	+10.02% limit-up · 5 consecutive
   澳弘电子 (605058)	SSE main board ±10%	17:52 unusual-movement announcement, "no hot-concept situation identified"	+9.99% limit-up · 3 consecutive
   博汇科技 (688004)	STAR Market ±20%	Evening risk warning (equity transfer agreement uncertain)	+19.98% limit-up · 2 consecutive
   双星新材 (Shuangxing New Material, 002585)	SZSE main board ±10%	17:45 abnormal fluctuation announcement	+10.04% limit-up · 3 consecutive
   众泰汽车 (000980)	SZSE main board ±10%	Evening warning on funding pressure	−3.00%
   风华高科 (000636)	SZSE main board ±10%	Evening warning on rising cost pressure	−2.36%

   Table scrolls horizontally

   4 of the 6 hit limit-up. This criterion is not useless, but what it constrains is risk exposure, not next-day direction.

   • Fix: risk-warning announcements go in the "position size and holding period" column without exception, and must not go in the "direction" column; the wording changes from "avoid" to "may participate, but single-day only, size-capped, no overnight holding."

3. Neither of today's two genuine leading themes made it into a pre-market table. See §3. The CCL price-hike chain is especially regrettable — pre-market §7 derived the causal chain entirely correctly ("upstream 生益's gross margin is +9.1pp over two years, ex-non-recurring +111%; downstream 中京/奥士康/满坤 are the cost absorbers"), yet (a) landed the conclusion on "watch only," and (b) not a single genuine upstream name appears in any of the four tables — 华正新材 (603186) and 宏和科技 (603256) do not appear anywhere in the text.

   • Fix: whenever it has been written that "a given link is a beneficiary," at least one tradeable name from that link must be given in the list, otherwise that causal derivation is commentary, not a list. And it must be checked in reverse: "which names in this link did I fail to list."

---

3. Verification of Today's Leading Themes

① Did the pre-market note call the strongest theme right? — No, and the direction was inverted

The pre-market note's branch No. 1 (cybersecurity) was −0.91% today and branch No. 2 (innovative drugs / CRO) was around −1.5% with across-the-board net outflow; while the AI hardware chain, which the pre-market note explicitly listed as carrying "the greatest absorption risk today," took the only net capital inflow of any magnitude in the market (semiconductors +RMB 4.461 billion) and pushed the STAR 50 into being the only index to close green.

This is not "one theme was missed," it is "the strongest theme was written up as the one most to be avoided." The root cause can be pinpointed exactly: the pre-market note mapped from the Philadelphia Semiconductor Index's overnight −5.86%, but it had itself written in §0.2 that "NVIDIA was only −3.36%, while optical modules were −9%−13% and semiconductor equipment −6%−8%; compute chips ≠ optical modules ≠ PCB, so the A-share mapping must be done layer by layer" — the layering principle was written down correctly, but it was not then used to ask "so did the overnight counterparts of the A-share storage/equipment/materials layer actually fall?" The principle was written; it was not applied to its own conclusion.

② Why wind power gets tracked and not chased — attribution could not be confirmed, and I am writing out the falsification process

Wind power is the strongest branch on price today, but I could not find a catalyst; all four searches and cross-checks came back empty or pointing the other way:

1. It was not driven by overnight/pre-market news, it was driven intraday. All wind power names opened with no premium: 新强联 opened +0.21%, 泰胜风能 −0.41%, 日月股份 −0.11%, 明阳智能 −0.10%; the first seal times of the four limit-up names cluster between 10:12 and 11:04 (吉鑫科技 10:12:04, 大金重工 10:17:18, 天顺风能 10:46:39, 锡华科技 11:04:17). A pre-market list is structurally incapable of catching this kind of move in advance — so this one does not count as a pre-market miss.
2. The only company event in the window that can be checked points the other way. Checking East Money announcement eiTime name by name: 大金重工, 吉鑫科技 and 锡华科技 have had no announcements since 9/12; 天顺风能's only in-window announcement is 9/14 19:58 "2025 Private Placement of A Shares — Prospectus (registration draft)" — that is a dilution event, directionally unsupportive of a limit-up, and cannot be used as a catalyst.
3. The money calibers contradict each other and do not support "concentrated large-money accumulation." Tonghuashun "wind power equipment" net inflow is −RMB 77 million (net outflow), while East Money "wind power equipment" main-force net inflow is +RMB 1.157 billion (net inflow) — the two calibers have opposite signs. (Their constituent sets and their definitions of "main force" both differ, so they cannot be subtracted and compared; they are placed side by side here only to show that no piece of money evidence here is robust.)
4. Public searching produced no primary source on any policy / tender / order dated 9/14–9/15. The wind power policy material retrieved is all from February 2026 or earlier, or framework language about the 15th Five-Year Plan, none of it inside this cycle's window.

So this report's conclusion is "attribution unclear," rather than manufacturing a reason. It is handled as "capital rotation, not fundamentally driven": strong price + no catalyst + contradictory money calibers + an intraday pulse — those four together are the standard signature of a one- to two-day move. Tomorrow it gets "track," not "chase."

⚠️ At the same time one reminder unfavorable to this report has to be written down: the pre-market note had itself disclosed, in the industry distribution of the limit-up pool, that "3 power grid equipment names (including 通达股份 (Tongda Cable) at 2 consecutive) — their driver could not be verified, left blank as such." Two days after being left blank, the adjacent wind power sector became the strongest thing in the market today. "Driver could not be verified" does not equal "no need to track" — tomorrow morning's pre-market note must break out "industries in yesterday's limit-up pool whose driver I could not attribute" into a separate table, instead of leaving them blank and dropping them.

③ One self-correction that has to be stated: the 深信服 downgrade — the reasoning was wrong but the conclusion survives

The pre-market note downgraded 深信服 (300454, ChiNext ±20%) from a recommendation slot to "watch only," on the grounds that "it is in substance an AI compute name, and that is precisely the line that was slaughtered overnight — buying it as a defensive asset gets the direction exactly backwards."

Today proves the second half of that sentence wrong: the AI compute line was not only not slaughtered today, it was the strongest money direction in the market. 深信服 was +0.54% with an intraday high of +5.97%. If the (correct) fact that "it is an AI compute name" had been used to follow today's money direction, it should have been upgraded rather than downgraded.

• The factual part was right (82.6% of revenue growth comes from cloud computing and AI infrastructure — hard data from East Money's revenue breakdown);
• The inferential part was wrong ("therefore it is a headwind asset today").
• Lesson: after reclassifying a stock, its direction within the new category must be re-assessed, rather than carrying over the direction conclusion from the old category. The pre-market note completed the reclassification but then simply carried the old bearish conclusion across.

---

4. Limit-Up Tier / High-Rung Boards

Tier structure: extremely flat, first boards are 78%

Yesterday's 55 limit-ups → only 7 limit-up again today, a promotion rate of 13.0%. In two days the tier has collapsed from "a 55-name base + a 4-rung top" to "a 32-name base + a 5-rung top" — the base shrank 42% while the top gained only 1 rung — which is textbook theme-relay breakdown, not a breakout in height.

Seal quality of the high-rung boards: the higher, the worse

The two high-rung boards (闽东电力 at 5 consecutive, 澳弘电子 at 3) broke their seals 66 times combined, with seal order / turnover ratios of 0.006 and 0.008 — the two worst boards in the pool. Their limit-ups today were not the result of money locking positions, they were the result of repeated two-way fighting followed by a barely-achieved close on the board. The opening premium on this kind of board is extremely unreliable tomorrow.

The heavyweights: recognizability sits in semiconductors and the price-hike chain, not in cybersecurity

Names with large turnover that rose (turnover > RMB 4 billion and closed green):

The cybersecurity branch has not one heavyweight. The three "heavyweight candidates" named pre-market — 电科网安 (RMB 510 million), 三六零 (RMB 1.645 billion), 安恒信息 (RMB 368 million) — all fell today, and not one of them made the top ranks of market-wide turnover. Pre-market §8 wrote, among the "conditions for the sector signal to hold," that "at least one of 电科网安 / 三六零 / 深信服 rises on expanding volume (heavyweights following)" — that condition was not met today.

Broken boards: gapping up and fading was today's most expensive pattern

Yesterday's limit-up names as a whole nevertheless outperformed today, and that must be stated truthfully; it does not contradict the "tier breakdown." Self-computed (taking yesterday's official limit-up pool of 55 names, joining each to today's quotes, leaving 54 valid after removing the suspended 华之杰 (Huazhijie)): 27 up (50.0%) / 27 down (50.0%), mean +0.68%, median +0.39%, against a market-wide advancer share of only 20.2% and a mean of −1.37%. Yesterday's limit-up names beat the market on breadth by 29.8 percentage points. The real structure is: yesterday's limit-up names carry excess return in aggregate, but are extremely polarized internally — best +19.98% (博汇科技), worst −10.04% (桂林旅游) — and the cost of betting on the wrong pattern is enormous (gapping up +6%~+10% and then hitting limit-down, single-day "open→close" of −15% to −17.5%). So yesterday's limit-up names can be traded tomorrow, but the "buy when the board opens" approach must be abandoned in favor of watching the opening premium.

---

5. Capital Flows

① Northbound flows: not provided this cycle, and this is not a data-retrieval failure

Since 2024-08-19, Stock Connect has ceased disclosing daily northbound net buying, retaining only quarterly / post-close calibers such as quotas and holdings. The relevant interfaces silently return 0.0 after the disclosure stopped, and copying that through would produce a pseudo-conclusion like "northbound net inflow of zero, neutral stance." This report does not provide that indicator.

② Dragon-Tiger List: no data this cycle (a scheduling reason, not an interface failure)

The Dragon-Tiger List is published daily by the exchanges after 18:00, while this piece is generated at 15:30, so structurally it cannot be retrieved. Today's seat preferences and the moves of institutions and hot money must wait for tomorrow's pre-market note. This report does not speculate about seat composition in any way.

③ Sector money: only 3 of 90 industries had net inflow

Tonghuashun caliber (net inflow = inflow − outflow, RMB bn):

East Money main-force net inflow caliber (industry level; a different caliber from the table above, not subtractable or comparable):

East Money main-force net inflow (concept level), top ranks:

⚠️ One caliber conflict must be self-disclosed; I cannot pick whichever one favors me. The Tonghuashun "components" industry is −RMB 3.319 billion (net outflow), while the East Money "PCB" concept is +RMB 3.918 billion (net inflow) — opposite signs. Their constituent sets are entirely different (the PCB concept has 199 names, the components industry has 63) and their definitions of "main force" also differ. I cannot determine which is more accurate. Therefore the §3 conclusion that "the CCL chain continues tomorrow" is supported by price and limit-ups (华正新材 and 诺德股份 at limit-up, the Jiemian wrap listing copper foil / CCL at the top of the gainers), not by money — the money evidence on this line is contradictory.

By the same token, the money evidence for wind power equipment is also contradictory (Tonghuashun −RMB 77 million vs East Money +RMB 1.157 billion), which is one of the reasons §3② gives it "track only, do not chase."

Only on the semiconductor line do the two calibers point the same way with the same magnitude (+RMB 6.122 billion / +RMB 4.461 billion, and the 121 up / 64~66 down breadth is identical under both calibers). This is the only direction today whose money evidence is robust.

④ One counterintuitive but important reading: cybersecurity shows "price down, money in"

The cybersecurity concept was −0.91%, 36 up / 96 down, which looks like withdrawal; but it took +RMB 1.410 billion of main-force net inflow, 8th on the concept board, while over the same period the "data security" concept was −RMB 44 million (net outflow).

• Possible explanation one: money is absorbing supply as the sector pulls back; the theme is not dead, it has merely shifted from "sentiment-driven broad rally" to "stock picking."
• Possible explanation two: the net inflow is concentrated in the seal orders of a handful of limit-up names (中新赛克, 启明信息, 博汇科技) and does not represent sector-level accumulation.
• I lean toward explanation two, because there is a back-computable piece of side evidence: the three limit-up names traded a combined RMB 1.566 billion, while the sector's 96 declining names saw outflows all day. +RMB 1.410 billion landing on 3 limit-up names is more likely than it landing on 132 constituents.
• The reading that can falsify this tomorrow: if cybersecurity puts up 2 or more limit-ups while the sector index still falls, explanation two holds (money is only doing high-level relay); if the sector index turns positive and the limit-up names are new faces, explanation one holds (genuine absorption).

---

6. Tomorrow's Outlook (2026-09-16 Wednesday)

① Judgments on theme continuation

② Tomorrow's focus list (with board tags and falsifiable verification points)

Verification-point design principles (a direct product of this reconciliation): ① set structural criteria first (limit-up counts, breadth, tier promotion, money direction), and only then price criteria — a price criterion like today's "can it hold after gapping up" had a market-wide pass rate of only 17.4%, so it measures the index, not the stock; ② no more than one criterion from the same source may appear in a single list; ③ every verification point must target the variable that will actually lose me money.

③ Avoid tomorrow

1. The two extremely weak high-rung boards — 闽东电力 (000993, SZSE main board ±10%) and 澳弘电子 (605058, SSE main board ±10%). The basis is not "the company denied the concept" (today proved that criterion has no predictive power for direction), it is seal quality: the two broke their seals 66 times combined, with seal order / turnover ratios of 0.006 and 0.008, the two worst boards in the pool. The opening premium on a weak board is unreliable — this is a position-size criterion, not a direction criterion.
2. Agricultural planting, retail, and tourism and hotels. Tourism and hotels 0 up / 35 down, planting 0 up / 30 down, retail 2 up / 69 down — "zero advancers" is harder evidence of withdrawal than the size of the decline. 敦煌种业 (600354, SSE main board) and 新农开发 (600359, SSE main board) are already at 2 consecutive limit-downs.
3. Yesterday's limit-up names that gapped up and faded today. 桂林旅游 (000978, SZSE main board) gapped up +6.45% → limit-down (open→close −15.48%); 欧亚集团 (600697, SSE main board) gapped up +10.03% → −9.21% (open→close −17.49%). This is today's most expensive pattern, with a single-day execution cost of −15% to −17.5%.
4. 香山股份 (002870, SZSE main board ±10%) — included pre-market as a "representative good-news name" in branch 3, hit limit-down today. Its 9/14 19:02 announcement shows the transaction is an issuance of shares plus cash payment to acquire assets (a major asset restructuring, including "the top ten shareholders as of the trading day before suspension" and "an explanation of share price movements before the suspension") — it is a resumption-of-trading event, not simply "an RMB 800 million acquisition as good news." A restructuring name resuming trading cannot be classified simply as an "acquisition positive."
5. Position size in the innovative drugs / CRO sector. 万邦医药 (301520, ChiNext ±20%) turnover rate 50.53%, 近岸蛋白 (688137, STAR Market ±20%) open→close −4.35%, and all three of the sector's carrier industries saw net outflow.
6. Lone survivors and the small-cap tail. First boards are 78.1% of the limit-up pool, the promotion rate is 13.0%, and the failed-seal rate is 42.9% — relay money is no longer sufficient to support the tier.

④ Input notes for tomorrow morning's pre-market list (six items, all direct products of this reconciliation)

1. Look at the money set first, then rank by fundamentals — the order cannot be inverted. Had "semiconductors +RMB 4.461 billion / PCB +RMB 3.918 billion / storage +RMB 2.881 billion vs cybersecurity +RMB 1.410 billion" been looked at first today, branch No. 1 would not have been picked wrong. Fundamentals determine position size and holding period; money determines direction.
2. Move "a company's self-initiated risk warning / concept denial announcement" from the "direction" column to the "position size" column. 4 of 6 hit limit-up today. The wording becomes "may participate, but single-day only, size-capped, no overnight holding."
3. Verification points must be structural first, price second, and must not share a source. Today "can it hold after gapping up" had a market-wide pass rate of only 17.4%, so tying 3 recommendation slots to it left only one independent variable. Also: "open→close" and "change on the day" are two rankers that point in opposite directions (correlations of −0.157 and +0.248 respectively, n=5,548), and which one is meant must be stated whenever either is cited.
4. Whenever it has been written that "a given link is a beneficiary," at least one tradeable name from that link must be given in the list. The CCL upstream hit limit-up today, while 华正新材 and 宏和科技 never appear once in the four pre-market tables. And it must be checked in reverse: "which other names in this link did I fail to list."
5. New standing section: break out "industries in yesterday's limit-up pool whose driver I could not attribute" into a separate table. After the pre-market note left power grid equipment and glass-and-fiberglass blank for two days, the adjacent wind power sector became the strongest thing in the market today. After leaving something blank, it must still be tracked; blank does not mean dropped.
6. After reclassifying a stock, its direction within the new category must be re-assessed. The classification of 深信服 (an AI compute name) was right, but the conclusion carried over the bearish direction of the old category, which led to a downgrade on the strongest line of the day.
7. Another new standing action: tomorrow morning the "zero advancer" industries of today must be re-examined. Tourism and hotels and planting and forestry had 0 advancers today — such extremes can either keep collapsing the next day or bounce after being oversold; it is tomorrow's single largest source of uncertainty and must be actively characterized rather than passed over in silence.
8. The "calendar for the next 5 trading days" must include ex-dividend/ex-rights dates and distribution implementation dates. 润泽科技 went ex-rights today (dividend of RMB 0.4288 per share), and the pre-market calendar table missed it even though the distribution implementation announcement had been published on 9/8. An ex-rights date changes both the prior close and the percentage-change baseline, and missing it produces two change figures in the next day's reconciliation that both "look right" (+1.10% and +0.38%). A general rule alongside it: before citing any stock's change percentage, back-compute it once with "(close − prior close) / prior close"; when the prior-close field disagrees with the daily K-line close, first check whether there was an ex-dividend/ex-rights event that day.

---

⚠️ Data-retrieval failures and channel switches this cycle (not sent to clients)

1. East Money's push2.eastmoney.com/api/qt/clist was again down market-wide with 502s (stock_board_concept_name_em threw HTTPError 502 directly). Per existing memory the delay cluster push2delay.eastmoney.com was tried first → HTTP 200, usable, so sector/concept money flows were still retrieved on the East Money caliber and did not degrade into "no sector-level money flow obtained." Market-wide pagination also went through the delay cluster, with total=5915 / 5915 retrieved / len==total verification passing, avoiding the "clist pagination one-sided bias" pitfall.
2. The limit-up / limit-down / failed-seal pools went through push2ex (a different cluster from clist, unaffected), with all six calls verifying tc == len(pool): 9/15 ZT 32=32, DT 25=25, ZB 24=24; 9/14 ZT 55=55, DT 16=16, ZB 30=30. sort was always passed unencoded as fbt:asc/fund:asc, and the silent failure of "tc correct but pool empty" did not occur.
3. stock_lhb_detail_em failed (TypeError: 'NoneType' object is not subscriptable) — but this is not an interface failure, it is a scheduling issue: the Dragon-Tiger List is only published after 18:00 and this piece was generated at 15:30. The body states this truthfully as "a scheduling reason," does not write it up as a retrieval failure, and makes no speculation about seat composition. This is exactly the memory item "the Dragon-Tiger List is only published after 18:00."
4. stock_zt_pool_previous_em was not used at all. Per the existing rule (that endpoint returns the same-day pool, not the next-day performance of yesterday's limit-ups), it was constructed by hand instead: take the 9/14 official pool of 55 names → join each to today's Tencent quotes → compute locally. Result: 27 up / 27 down, mean +0.68%, 7 promoted. Misusing that endpoint would have produced "all 55 names rose."
5. Tencent qt.gtimg.cn succeeded in every call, with field 30 timestamps verified line by line as 20260915153xxx, confirming post-close settled values. Three codes guessed wrong while investigating causes (920099 is actually 瑞华技术 (Ruihua Technology), 688511 is actually 天微电子 (Tianwei Electronics), 301311 is actually 昆船智能 (Kunchuan Intelligent)) — these were identified and not written into the body; 688797 臻宝科技 and 688401 路维光电 (Newway Photomask) returned empty on the Tencent route, and their data was supplied by the official failed-seal pool instead.
6. Turnover closes on itself across three independent routes: Tencent index sum RMB 1,612.71 billion｜Jiemian News closing wrap RMB 1,612.717 billion｜clist pagination self-count RMB 1,625.28 billion and the Tonghuashun 90-industry sum RMB 1,625.26 billion (the latter two include the BSE). The self-computed advance/decline counts of 1,120/4,376 are name-for-name consistent with the values published in the Jiemian wrap, which is the hardest cross-check of this cycle.
7. All four lines of wind power attribution came up empty, and this was written into the body as "attribution unclear" rather than manufacturing a reason: two WebSearch runs returned only descriptions of the phenomenon ("the sector strengthened") and policy material from February 2026 or earlier; the East Money announcement API was checked name by name on eiTime, with 大金重工 / 吉鑫科技 / 锡华科技 having no announcements since 9/12 and 天顺风能's only in-window announcement being the 9/14 19:58 private placement registration draft (dilutive, opposite direction). Following the "use commodity futures to falsify a theme" approach, the intention here was to cross-check with an upstream commodity, but wind power has no corresponding active futures contract, so it could not be executed.
8. One caliber conflict has been self-disclosed in the body, with no cherry-picking: Tonghuashun "components" −RMB 3.319 billion vs East Money "PCB concept" +RMB 3.918 billion (opposite signs); Tonghuashun "wind power equipment" −RMB 77 million vs East Money "wind power equipment" +RMB 1.157 billion (opposite signs). Constituent sets and definitions of "main force" both differ, so they cannot be subtracted and compared. The body therefore explicitly limits the support for the CCL chain to "price and limit-ups," and downgrades wind power to "track only." Only semiconductors have both calibers pointing the same way with the same magnitude, and the body lists it on that basis as the only direction with robust money evidence.
9. One search result actively refused. NBD's "Nearly 4,000 stocks fell! Wind power, PCB and cybersecurity concepts strengthened…" looks like a same-day closing wrap by its headline, but is actually from its "A-shares morning session" column; another China News Service piece was a "midday wrap" (its "the Shenzhen index closed up" contradicts the close of −0.72%). The body uses locally self-computed closing data throughout, treating these media pieces only as attribution leads. This is the memory pitfall "searching for 'today's closing wrap' returns the opening broadcast," and it was not stepped into this time.
10. No sub-agents were used this cycle (per session instruction). Fundamental data follows the conclusions already verified in the pre-market list, with sources labeled, and no unverified fundamental figures were added. The two weak-evidence points left over from pre-market (the segment revenue and AI security share of 电科网安 002268 and 三六零 601360) remain uncovered this cycle and remain unconfirmable; they have not been filled in or papered over in any way in the body.
11. One silent trap caught by "back-computing the price triple," worth archiving on its own. While writing, it emerged that 润泽科技 (300442)'s real-time "prior close" field of 60.25 disagreed with the 9/14 daily K-line close of 60.68, and that 60.68 is exactly the value cited in the pre-market list. Both numbers "look normal," and neither would expose anything on its own.
    • Back-computation locates it: 60.68 − 60.25 = 0.43, which has the shape of a cash-dividend ex-rights adjustment.
    • Primary source confirms: East Money RPT_SHAREBONUS_DET returns ex-dividend/ex-rights date = 2026-09-15 (today), record date 2026-09-14, dividend (pre-tax) RMB 4.288 per 10 shares = RMB 0.4288 per share; checking, 60.68 − 0.4288 = 60.2512, digit-for-digit consistent with the quote interface's 60.25. The real-time field is right (ex-rights caliber), the daily K-line is the unadjusted close; neither is wrong — what is wrong is subtracting across calibers.
    • What would have happened without back-computation: either writing +0.38% based on 60.68 (omitting the dividend, understating the return), or writing +1.10% that does not reconcile with the Monday close cited pre-market — both would create an unexplainable numeric conflict within the same piece. The body now carries an ex-rights note and gives both figures along with the conversion between them.
    • data.eastmoney.com/yjfp/detail/300442.html is JS-loaded and WebFetch retrieved only an empty table-header skeleton; the datacenter-web JSON API had to be used to get the dividend detail.
    • Only the key names cited in the body were back-computed this time; the 60+ names on the full list were not each checked for ex-rights events, so if tomorrow's reconciliation again shows "the prior-close field disagrees with the daily K-line," ex-rights should be suspected before an interface failure.
12. Data that could not be obtained (left blank or explained in the body, not fabricated): Dragon-Tiger List seats (published after 18:00), daily northbound net buying (disclosure ceased from 2024-08-19), a primary source for the wind power branch's catalyst, and the specific triggering events for the withdrawal in agriculture and retail (searching produced only generalized language about "excessive prior gains and profit-taking," which was not adopted as a conclusion).

---

⚠️ Risk disclosure: this recap is post-close information review and observation only, and does not constitute investment advice. Data may differ in timeliness or caliber; please defer to exchange/company disclosures. It must not be used directly as a basis for trading.

Data sources: exchange-caliber limit-up/limit-down/failed-seal pools (East Money push2ex)｜market-wide quotes and sector money flows (East Money push2delay, Tonghuashun)｜individual stock quotes (Tencent Finance qt.gtimg.cn, post-close settled values)｜advance/decline cross-check (Legulegu, Jiemian News closing wrap¹)｜company announcement publication times (East Money announcement API eiTime field)｜National Business Daily²｜China News Service³