A-Share Market Recap | 2026-09-07 Monday

Scope and methodology boundaries of this report (judge the strength of the conclusions accordingly) ① Quotes are on a 2026-09-07 close basis (Tencent qt.gtimg.cn, read at 20260907153045–20260907153138). Each stock is given all four prices — open / high / low / close — not just the close. ② Limit-up / limit-down / failed-limit / prior-day limit-up pools use the official East Money push2ex basis, with len(pool)==tc verified (limit-up 93==93, failed-limit 19==19, 9/4 prior-day limit-up 39==39). ③ Limit-up determination is tiered by board limit: SSE/SZSE main board ±10%, ChiNext/STAR Market ±20%, BSE ±30%. Every stock in this report is labelled with its board. ④ ⚠️ Sector moves and sector fund flows in this report are on a THS (Tonghuashun) basis (stock_board_industry_summary_ths, 90 industries). Its "net inflow" = inflow − outflow, which is not the same ruler as the East Money "main-force net inflow" used in the 9/7 pre-market report; the two reports cannot be subtracted against each other. Single-stock fund flows are likewise on a THS basis. ⑤ The Dragon-Tiger List (LHB) was not obtained for this report: the exchanges publish it after 18:00, and this report was generated at 15:30, so it does not yet exist at that timestamp. §5's seat analysis is therefore left blank, with no substitute data used. ⑥ No northbound-flow data (daily net buying has not been disclosed since 2024-08-19); this report gives no northbound figures. ⑦ Reconciliation uses two baselines: one column is versus the pre-market baseline price (the 9/4 close), i.e. the quote as of when the pre-market report was written; the other is versus today's open, i.e. the earliest executable price after reading the pre-market report. The two differ substantially — see §2.

⚠️ Data-retrieval and search log (this section is not sent to clients):

1. Channels that worked

• Indices/stocks via Tencent qt.gtimg.cn, which returns OHLC + turnover rate + turnover value in one call.
• The four pools — limit-up / limit-down / failed-limit / prior-day limit-up — via push2ex, normal; len(pool)==tc was verified one by one per memory [[limit-updown-must-use-official-pool]].
• Sector quotes / sector fund flows / single-stock fund flows all via THS (stock_board_industry_summary_ths, stock_fund_flow_individual); advance/decline counts via stock_market_activity_legu.
• In-window marketwide announcements via East Money np-anotice-stock; 600 items pulled and scanned one by one.

2. Failed channels / data-retrieval faults this run

• ⚠️ East Money push2 clist main cluster returned HTTPError and the push2delay delayed cluster returned TimeoutError — both down; this is the third consecutive trading day. ak.stock_board_industry_name_em accordingly returned RemoteDisconnected. Per memory [[ths-fallback-when-em-clist-down]] the whole line fell back to THS rather than degrading into "no sector-level data obtained"; the pitfall flagged in that memory was handled with zfill(6) (the THS code column is int64 and eats leading zeros), giving 24/24 matches with no silent MISSING.
• ⚠️ stock_lhb_detail_em('20260907') returned TypeError: NoneType — this is a scheduling problem, not an interface fault; running at 15:30 will always fail to get it, and memory [[lhb-publishes-after-1800]] has now closed this loop three times. This should no longer be investigated as a fault; the schedule should be changed: if the post-close recap slot is to include the LHB, it must move past 18:10, or a separate backfill slot must be split out.
• ⚠️ WebFetch on The Paper's "September 7 close" article returned 2026-04-13 content (the piece cites Shanghai Composite 3988.56 / combined turnover RMB 2,150.3 billion, which does not match today's 3932.70 / RMB 1,964.4 billion at all); the entire piece was discarded. Same shape as memories [[stale-source-silent-lag]] and [[recurring-column-headlines-hide-date]]. What exposed it was reverse-checking the index level, not looking at the headline date.

3. Three methodological cross-checks actively performed that changed the conclusions

1. ⚠️⚠️ [The single most important one in this report] "Today's strongest theme is semiconductor equipment" was refuted by the industry breakdown. The draft was going to follow pre-market Branch 2 and write "the equipment line delivered." But spreading the 93 limit-up stocks across industries gives: components 14, communications equipment 8, 22 combined — while semiconductor equipment had not a single limit-up; sector gains likewise run components +6.83% > communications equipment +3.77% > semiconductors +2.54%. The real theme is CPO / optical modules / PCB, not front-end equipment. Without spreading out the industry breakdown, "rose by association" would have been written up as "the theme delivered." This is exactly the mandatory question required by memory [[ztpool-industry-mix-only-counts-target-branch]].
2. ⚠️ [The "equipment stocks >5%" criterion was recomputed name by name, and the conclusion is the opposite of "the sector is confirmed"] The pre-market criterion was "at least 3 equipment stocks up >5% simultaneously." Measured: of 8 equipment stocks only 芯源微 (Kingsemi, 688037) +6.94% cleared the bar, so the criterion was not met; yet over the same session the semiconductor sector's net inflow of RMB 23.978 billion was No. 1 marketwide and all 8 names saw net inflows. The price criterion says no; the fund-flow criterion says yes. This report does not fudge it — both are written down, along with the point that the criterion itself was mis-specified (see the §2 self-critique).
3. ⚠️ [A definitional gap in the limit-down count] The official limit-down pool holds only 2 names, both main board, but 摩尔线程 (Moore Threads, 688795) closed at −20.00%, exactly the STAR Market limit-down, and is not in the official pool (legu likewise reports "true limit-downs 0"). It has been confirmed that the limit-up pool itself does include STAR / ChiNext / BSE names (it contains +20.00% and +29.97% samples), so this is not a tiering issue but presumably a seal-order determination rule. This report handles it as "official pool of 2 + 摩尔线程 named separately," without changing the definition itself.

4. What this report did not obtain / did not do (listed honestly)

• The LHB (9/7) was not obtained, for the reason above — scheduling, not a fault.
• No exclusive attribution was obtained for the CPO / optical module / PCB theme. Exclusion work done: of the 600 in-window marketwide announcements, 0 came from optical module / PCB companies; three same-day financial pieces (Sina 09:51, 21 Finance 10:22, Jiemian) all describe the phenomenon without giving a cause. The two strongest candidates so far are ① the overnight MRVL +7.05% (a figure our own pre-market report recorded) and ② the CIOE optical expo opening 9/9, but neither is an exclusive attribution, so this report marks them "pending verification" rather than confirmed. Per memory [[no-news-found-is-not-no-news]], what is written here is "the announcements and three articles were checked and turned up nothing," not "there is no news."
• Same-day attribution for the 9/4 A-share semiconductor −2.82% / net outflow of RMB 16.038 billion is unobtained for the third report running and stays on the books.
• The primary source for "August breeding-sow inventory −2.94% MoM" (item 1 of the pre-market report's post-finalization to-do list) was not pursued in this report. Today's actual driver in agriculture has shifted to the planting/grain-price line, and the livestock leaders in fact fell, so the marginal significance of that figure for tomorrow has declined, though it remains unclosed.

---

0. Recap in One Sentence

Today was a clean case of "tech eating the heavyweights": ChiNext Index +3.41%, STAR 50 +2.42%, Shenzhen Component +1.91%, versus SSE 50 −0.41% and the Shanghai Composite up only +0.07% — a 3.8pp split between indices, exactly the reverse of last Friday (heavyweights holding, tech collapsing). Sentiment temperature: euphoric. Limit-ups 93 (39 last Friday), failed-limit rate 17.0% (55.2% last Friday), 3,039 advancers, and the 9/4 limit-up cohort today posted a mean of +4.89% with 13 names hitting limit-up again — the best day for follow-through quality in recent memory.

And today's strongest theme is the one that was not in the pre-market list at all. Spreading the 93 limit-up stocks across industries: components 14 and communications equipment 8 take the top two spots, with sector gains of components +6.83% (63 up, 0 down) and communications equipment +3.77%; the leaders are 中际旭创 (Zhongji Innolight, 300308) +10.38% on RMB 37.6 billion of turnover and 沪电股份 (WUS Printed Circuit, 002463) limit-up on RMB 9.6 billion. This is the CPO / optical module / PCB line, not the front-end semiconductor equipment line the pre-market report bet on — front-end equipment did rise (semiconductor sector +2.54%), but it ranked only third and had not a single limit-up.

Result of the pre-market list: of 5 recommendations, 2 rose and 2 beat the all-A median (+0.39%), with a 5-name mean of +0.66%. The 33 names judged Pass / watch only averaged +4.38%. In other words, this report got the direction right (tech) but the ordering backwards — what it recommended underperformed what it rejected by 3.7 percentage points. Branch 1 (the RMB 360 billion capital injection) turned out to be the two worst sectors of the day: insurance −2.45% (0 up, 5 down) and banks −1.45% (0 up, 42 down).

Tone for tomorrow: the theme is in its ignition phase, but has only one day of history. All 14 component limit-ups were first boards and all 8 communications-equipment limit-ups were too — this line ignited on day 1 today, it is not in day-2 fermentation, so tomorrow is its first test of divergence.

---

1. Market Overview

1.1 Indices

Turnover: Shanghai RMB 897.9 billion + Shenzhen RMB 1,048.1 billion + BSE 50 RMB 18.4 billion ≈ RMB 1,964.4 billion.

✅ Turnover has been cross-checked on two independent lines: the Tencent index sum gives RMB 1,964.44 billion versus the THS 90-industry sum of RMB 1,964.23 billion, a difference of RMB 210 million. Following the habit from memory [[ths-fallback-when-em-clist-down]], after switching data sources, find one quantity present on both sides and reconcile it before using anything else.

⚠️ The Shanghai Composite's +0.07% is a deceptive number. It is the net result of two offsetting forces — banks and insurers dragging −1.5% to −2.5%, tech pulling +2% to +7% — not "a narrow range-bound session." Reading the index says "nothing happened today"; reading the structure says "the theme changed today." Characterizing the day by the index would miss the most important information of the session.

At 3932.70 the Shanghai Composite is 1.71% below 4000 (the pre-market report put it at 1.78%). The pre-market report listed "an index breakout attempt driven by the heavyweights" as the thing most worth watching this week — today the heavyweights went down, the breakout did not happen, and the driver of index gains has already shifted to the small-cap tech end.

1.2 Sentiment Indicators

Sentiment temperature: euphoric (but only one day old). Three indicators improved simultaneously: the failed-limit rate was cut from 55.2% to 17.0%, the limit-up count multiplied 1.4x, and follow-through on the prior day's limit-up cohort flipped from negative to +4.89%. The pre-market report listed "the 9/4 limit-up cohort averaged −0.57% with only 34% advancing" as "the common background pressure on every theme on Monday"; that claim was fully falsified today — the wealth effect was not contracting, it expanded that very session.

Note on the limit-down basis: the official limit-down pool holds 2 names — 康欣新材 (Kangxin New Materials, 600076, SSE main board ±10%) −10.09% and 兴业科技 (Xingye Leather Technology, 002674, SZSE main board ±10%) −10.02%. ⚠️ One more must be named separately: 摩尔线程 (Moore Threads, 688795, STAR Market ±20%) closed at −20.00%, exactly the limit-down magnitude, yet is not in the official limit-down pool (legu likewise reports "true limit-downs 0"). The cause was the same-day unlocking of 25.7745 million IPO and public-offering off-market placement shares (21 Finance, 9/7 10:22); the stock made a new post-listing low and its market cap fell below RMB 200 billion. This report follows the official pool and reports 2 names without redefining anything, but lists this one separately because it is the most direct sample of today's risk in high-priced names.

---

2. Reconciliation of the Pre-Market List

2.1 The 5 Recommendations (pre-market §9① "the 5 stocks most worth watching today")

Note on the two baselines: "vs pre-market price" = versus the 9/4 close (the quote as of when the pre-market report was written); "vs today's open" = versus today's opening price, i.e. the earliest executable price after a reader has read the pre-market report. The A-share pre-market report is written before the open with no auction data, so today's open is its executable price.

Hit rate and benchmarks:

⚠️⚠️ The line that has to come first: for both names that delivered, most of the gain was handed over in the opening jump. 兆易创新 was +3.65% versus the pre-market price but only +0.69% after the open; 北方华创 was +3.52%, only +1.29% after the open. Measured against today's open — the only executable price — the 5-name mean is −0.34%, not +0.66%. The pre-market report's own §9 wrote that "the catalyst on a short-duration branch may be fully priced in the opening jump" — that call happened exactly as predicted, including on the two names it got right.

2.2 Reconciling the Pass / Watch-Only List — the Ugliest Table in This Report

Pre-market §6 and §9③ together judged 33 names as Pass or "watch only." Statistics are broken out by the type of Pass reason given pre-market (not by ex-post moves, to avoid hindsight):

The worst misses, name by name (all judged Pass or watch only pre-market and limit-up today):

The correct calls (a minority, but very concentrated by type):

2.3 Verdict on Each Verification Point Set Pre-Market

Pre-market §8 and §9 together gave 6 falsifiable verification points. Judge the verification points first, then talk about conclusions — per memory [[verification-point-beats-headline-conclusion]], this section carries more weight than the §2.1 hit rate.

4 right, 2 wrong. But the two wrong ones are wrong in completely different ways and deserve separate treatment:

• ④ is a mis-specified criterion, not a market failure to deliver. The "3 equipment stocks >5%" threshold puts a ±10% SZSE main-board stock (北方华创) and ±20% STAR Market stocks (拓荆/盛美/中微) on the same ruler; for a main-board name 5% is half a limit-up, for a STAR name only a quarter. More importantly, it uses price to verify something that is about fund flows. On the same day the fund-flow criteria (③⑤) both passed, the semiconductor sector's net inflow of RMB 23.978 billion was No. 1 marketwide, and not one of the 8 equipment stocks saw a net outflow. Price said no, flows said yes, and flows were right.
• ⑥ applied the criterion to the wrong object. 牧原/温氏 are the leaders of the livestock (hog price) line, whereas today's agricultural strength was entirely in the planting / grain-price line (planting & forestry +4.33% vs livestock farming +2.21%). The pre-market report used sub-branch A's leaders to verify the persistence of sub-branch B.

2.4 One-Sentence Self-Critique: What Was Called Right, What Was Called Wrong, and How to Change the Next Pre-Market Report

What was called right is "what to avoid": precious metals (−2.77%, weakest marketwide), the three digital-RMB names with 27%–38% turnover rates (−6.69% / −3.70% / −0.24%), and the pseudo-beneficiary designation of 中国人寿 (−1.84%). What these three share is that their criteria are "this thing is physically impossible" or "this position is already overextended" — independent of where the money went that day.

What was called wrong is "what to recommend" — and it was systematically wrong in the same place:

1. ⚠️⚠️ The heaviest one: the overnight SOX was read as "equipment + memory," missing the leg that was actually rising. In the table in pre-market §1.2, the second-largest gainer was MRVL +7.05%, and it was filed under "other"; the pre-market report then used the fact that "NVDA only +0.84%, AVGO only +0.21%" to explicitly rule out the "AI compute demand" channel, asserting that "what was priced that night was the supply-side chain of capacity expansion → equipment orders." That exclusion was wrong: MRVL is neither NVDA-style compute demand nor KLAC-style front-end equipment; it is AI interconnect / optical DSP / custom ASIC — precisely the line the A-share market bought today. The data was sitting in our own table at the time; the classification buried it. This is the same shape as memory [[narrative-first-screening-blind-spot]]: the "supply side" narrative came first, names were stuffed into it, and whatever didn't fit the narrative was silently dropped.
2. Within that same line, fundamental quality was once again used as a short-term ranker. Today's moves for the 8 semiconductor equipment names: 芯源微 +6.94% (Pass) > 精测电子 +4.98% (Pass) > 拓荆 +4.82% (Pass) > 中科飞测 +4.79% (Pass) > 盛美 +4.15% (watch only) > 北方华创 +3.52% (recommendation No. 5) > 中微 +2.56% (watch only) > 华海清科 +1.11% (watch only). The pre-market report used "ex-non-recurring share 99.1%, OCF conversion 1.12" to rank 北方华创 first among equipment names, and on the day it finished 6th of 8. Memory [[financial-quality-is-not-a-short-term-ranker]] already recorded this once (盛美上海 with ex-non-recurring −15.31% was Passed and finished +12.16%, No. 1 marketwide); this is the second occurrence of the same error, and pre-market §7.2 itself ended with the reminder "financial quality is not a short-term ranker" — and then ranked by it anyway.
3. Branch 1's mechanism analysis was right and its directional call was wrong, and both were written in the same paragraph. Pre-market §2.1 spent three thousand characters arguing that "issuance at market price offers no premium protection, dividend-per-share dilution of −3.34%/−6.16%/−4.07%, and the 2025 natural experiment shows dilution passing 100% through to dividends" — and the market delivered that analysis point by point today (banks 0 up 42 down, insurers 0 up 5 down, the non-recapitalized 邮储/交行 outperforming). Yet the conclusion column of that same section read "Direction: still probably up," on the grounds of "an extremely strong signal from the Ministry of Finance's 5-year lock-up + the broad rally in state-owned banks the day after the 2025 announcement." The result: its own computations said down, its own cited precedent said up, and it went with the precedent — staking 2 of the 5 recommendation slots on it.

Three executable changes for the next pre-market report:

• Change 1: overnight overseas gainer lists must be re-bucketed by "industry-chain stage," and after bucketing there must be an "unclassified" column processed name by name — nothing may be swept into "other." Today's lesson is that the name that was missed (MRVL) was not unseen; it was seen and put in the wrong bucket. Concretely: equipment / memory / compute / interconnect and optics / foundry — five buckets, and any empty bucket must be written up explicitly as "no notable move in this stage during this window."
• Change 2: when a branch's "own computation" fights its "historical precedent," the two must be split into two lines with separate conclusions, and it must be stated explicitly which one drives slot allocation. The correct write-up today would have been: "mechanically bearish (own computation), sentiment-wise bullish (precedent), and the precedent is already explained away by the premium-issuance conditions of 2025, which differ from this round — therefore no recommendation slot."
• Change 3: verification points must give both a "price criterion" and a "fund-flow criterion," and must state in advance which one wins when they conflict. Today ④ conflicted with ③⑤, and the pre-market report had no rule for conflicts, leaving "is the sector confirmed or not" to be adjudicated on the spot after the close. Suggested default: on ignition day listen to flows, on fermentation day listen to price.

---

3. Verification of Today's Themes

3.1 Master Branch Table

3.2 Did the Pre-Market Report Identify the Strongest Theme? — No, and What It Missed Was No. 1

Today's sector rankings for the three branches given pre-market:

⚠️ The pre-market report assigned its highest S rating to the two sectors that finished second-to-last and fifth-from-last marketwide, while the No. 1 sector was not on the list at all. This is the single thing most worth taking away from this report.

Was this missed theme something the pre-market report "could not possibly have known"? — No; two leads were already in hand at the time:

1. The overnight MRVL +7.05%, written in the pre-market report's own §1.2 table, was the second-largest gainer of that session (behind only KLAC +7.32%), yet it was filed under "other"; and when the pre-market report used flat NVDA/AVGO to rule out the "AI demand" channel, it never pulled MRVL out to ask "then which channel does it belong to?"
2. CIOE, the China International Optoelectronic Exposition, opens September 9–11 at the Shenzhen World Exhibition & Convention Center (the 27th edition, held alongside IICIE, the integrated circuit innovation expo, and elexcon, the Shenzhen international electronics show). This is a calendar event, not something inside the "weekend news window," so the pre-market methodology of "scanning all marketwide announcements from 9/4 15:00 → 9/7 07:00" was structurally incapable of catching it. This is the same class of failure as memory [[scan-window-misses-multiday-themes]]: a news window scans "what has just happened" and cannot see "what is about to happen."

⚠️ Attribution honesty statement: the two items above are candidate explanations, not confirmed attribution. Exclusion work done in this report: ① of the 600 in-window marketwide announcements, 0 came from optical module / PCB companies; ② three same-day financial pieces (Sina 09:51, 21 Finance 10:22, Jiemian newsflash) all describe the phenomenon without giving a cause. This report therefore offers no exclusive attribution for this theme, only the evidence and the candidates.

3.3 Agriculture: The Theme Did Not Ebb, It Switched Legs

The pre-market criterion was "the leaders (牧原/温氏) not following = a sign of an ebbing theme." The leaders indeed did not follow today (−0.14% / −0.52%), yet agriculture was the 3rd strongest sector marketwide. The criterion failed because it was watching the wrong sub-branch:

Two judgment errors that must be acknowledged:

1. 亚盛集团 (600108) was Passed and is 3 boards one-line today. The pre-market Pass reason was "the company's own 8/14 announcement says drought will adversely affect 2026 results, so the market's reason for buying it runs opposite to what the company disclosed." That fact was right and is still right today — but it is not a short-term reason to sell. On the contrary, drought is itself part of the rising-grain-price narrative: the market is not buying 亚盛's current-period earnings, it is buying the macro chain of "reduced harvest → grain prices rise," and 亚盛 is the purest and most recognizable name on that chain. The pre-market report used a company-level earnings impact to negate an industry-level price narrative; the two are not on the same level.
2. 敦煌种业 (600354) was judged "watch only" on the grounds of seal orders/turnover of 0.008, the worst marketwide, plus a 27.93% turnover rate — and it limit-upped today. ⚠️ But the pre-market report deserves a fair word here: its seal-order quality today was still very poor — seal orders RMB 92 million / turnover RMB 569 million = 0.161, turnover rate 11.51%, third from last among the 13 consecutive-board names; and money was a net outflow of RMB 109 million. The observation "the seal orders are bad" was not wrong; what was wrong was treating "bad seal orders" as "it won't rise," when all it can actually tell you is "even if it rises you can't hold it." This matches memory [[position-metrics-size-not-direction]]: positioning metrics size a position, they do not set direction.

By contrast, the two best seal-order qualities today belong precisely to two Passed names: 亚盛集团 at seal orders/turnover 4.473 (No. 1 marketwide) and 播恩集团 at 3.716 (second). On 9/4 the pre-market report recorded 亚盛 at 0.167 and 播恩 at 0.341 — seal-order quality changed by an order of magnitude within two trading days, and the pre-market report was using 9/4 seal-order data to judge 9/7.

3.4 The Recapitalization Branch: Analysis Entirely Right, Direction Entirely Wrong

Measured today:

Sector level: banks −1.45%, 42 constituents 0 up 42 down; insurers −2.45%, 0 up 5 down. Flows: banks −RMB 3.245 billion, insurers −RMB 2.010 billion.

Three conclusions that can all hold at once but point in different directions:

① The pre-market dilution calculation was delivered by the market point by point. The decline ordering of the three recapitalized names (人保 −1.75% < 工行 −1.97% < 农行 −2.16%) matches the pre-market self-computed net-asset-per-share dilution ordering (+0.159% > −0.897% > −0.913%) exactly; and the two non-recapitalized names (邮储 −0.55%, 交行 −1.35%) outperformed all of the recapitalized ones. The market priced on "the more you are diluted, the more you fall," which is exactly what those three thousand characters in pre-market §2.1 argued.

② But the pre-market directional conclusion ("still probably up") was wrong, and it was the most wrong call marketwide. Banks were 0 up 42 down, the only one of the 90 industries with zero advancers (insurers were 0 up 5 down but have only 5 constituents).

③ 建设银行 −2.46% fell the most, and this one name does not fit the "dilution pricing" model — it was not recapitalized, its CET1 of 14.24% is the highest of the six, and pre-market judged it "watch only." This report did not obtain an attribution for why it alone fell the most, and will not invent one. One possible but unconfirmed direction is its 9/7 turnover rate of 1.42%, the highest of the six (ICBC was only 0.13%), implying stock-level fund behaviour — but that is not enough to serve as a conclusion.

④ There is one more layer in the timing: today was the first trading day after the recapitalization plan was announced, and not one of the four approval procedures (shareholders' meeting + NFRA + SSE + CSRC) has been completed. In other words, this is not "a decline after good news was delivered," it is "good news traded as bad news on the very day it was announced." The reason the 2025 precedent cited pre-market (state-owned banks broadly rising the day after the announcement) does not apply was already spelled out by the pre-market report itself: that round was a fixed-price issuance at a 6%–19% premium, this round is at market price with no premium subsidy. The difference in conditions between the precedent and this round was argued in full in the pre-market report, and yet was never used to overturn the precedent itself.

---

4. Limit-Up Ladder / Ladder Height

4.1 Consecutive-Board Structure

⚠️⚠️ The most important row in this table is the last one: 80 of today's 93 limit-ups (86%) are first boards, and all 22 of the leading component and communications-equipment names are, without exception, first boards. This theme's ladder height is 1, not 2 and not 3. "The sector that rose the hardest" and "the most fragile ladder" are the same thing today — with no 2-board or 3-board names carrying the baton, all 22 face their "first consecutive board" test simultaneously tomorrow, with no intermediate layer as a buffer.

The answer to another mandatory question: the highest consecutive-board name marketwide (6-board 龙版传媒) belongs to publishing, and does not belong to any of today's strongest themes. Per memory [[ztpool-industry-mix-only-counts-target-branch]], this item must be answered separately — you cannot merely count names within the target branch. What it means is: the height of sentiment sits on an old theme that pre-market judged to be "an ebbing signal," while the main theme's height is zero.

4.2 Leader Identification (large turnover + limit-up/large gain + recognizability)

⚠️ The quality stratification among the leaders is obvious: 中际旭创, 沪电股份 and 光迅科技 show the healthy pattern of "volume expansion + large gain + net inflow"; whereas 剑桥科技 (603083) limit-upped on a net outflow of RMB 1.703 billion, 景旺电子 (603228) limit-upped on a net outflow of RMB 169 million, and 敦煌种业 (600354) limit-upped on a net outflow of RMB 109 million — these three are "sealed but the money is leaving," the weakest for follow-through tomorrow.

4.3 Failed Limits and Broken Ladders

19 failed limits, a failed-limit rate of 17.0%. Most frequent failures: 旷达科技 (Kuangda Technology, 002516, SZSE main board) broke 7 times and closed +8.69%, 东易日盛 (Dongyi Risheng, 002713, SZSE main board) broke 5 times and closed +7.08%, 德展健康 (Dezhan Health, 000813, SZSE main board) broke 4 times and closed +9.33%, 恒盛能源 (Hengsheng Energy, 605580, SSE main board) broke 4 times and closed +8.63%.

Results for the "signal light" stocks named pre-market:

• 龙版传媒 (605577): no failed limit; on the contrary it sealed a 6th board. The pre-market criterion was "opening the board = media ebb confirmed, sealing = extreme sentiment." The result is the latter: extreme sentiment. Same-sector 中国出版 (601949) at 2 boards and 读者传媒 (Reader Media, 603999) at a first board followed, giving 3 publishing limit-ups. Media has not ebbed, but it is now at the high point of sentiment.
• 楚天龙 (Chutian Dragon, 003040): no limit-up (closed +8.66%, high of 23.18, limit-up price about 23.23), and the RMB 543 million of seal orders seen pre-market did not reappear. The pre-market criterion was "the seal orders disappearing means it is over"; strictly speaking the seal orders were indeed gone, but the stock ran from a −0.52% lower open all the way to +8.66% (+9.23% versus today's open) with a 25.01% turnover rate. The criterion triggered, but the conclusion pointed the wrong way. Internal divergence within the digital-RMB sector was extreme: 楚天龙 +8.66% while 四方精创 −6.69% and 恒宝股份 −3.70%.

---

5. Fund Flows

⚠️ Basis: sector and single-stock fund flows in this section are all on a THS basis (net = inflow − outflow), because East Money clist was unavailable across the board. This is not the same ruler as the East Money "main-force net inflow" cited in the 9/7 pre-market report; figures from the two reports cannot be subtracted against each other.

5.1 Northbound Flows

No data. Since 2024-08-19 Stock Connect no longer discloses daily net buying; this report gives no northbound figures and does not pass off any interface-returned 0.0 or quota-ceiling figure as one.

5.2 Dragon-Tiger List

Not obtained for this report. The exchanges publish the LHB after 18:00 on each trading day, and this report was generated at 15:30, so the data does not yet exist at that timestamp. This section is left blank; main-force fund flow or large-order data is not used as a substitute for seat analysis.

5.3 Sector Flows (THS basis, unit: RMB billion)

Top net inflows — heavily concentrated in the electronics complex:

Top net outflows — heavily concentrated in financials and the old themes:

Today's fund structure is extremely clean: 7 of the top 8 net-inflow sectors are electronics/equipment, and 3 of the top 8 net outflows are financials (brokers/banks/insurers, −RMB 10.09 billion combined). This is not a broad rally, it is a directional migration between sectors.

⚠️ There is also a corroborating piece of evidence independent of the sector basis — 7 of the 8 largest turnovers marketwide belong to this theme: 中际旭创 (300308) RMB 37.635 billion, 新易盛 (300502) RMB 20.918 billion, 长鑫科技 (688825) RMB 20.493 billion, 东山精密 (Suzhou Dongshan Precision, 002384) RMB 13.689 billion +6.65%, 天孚通信 (300394) RMB 13.607 billion, 兆易创新 (603986) RMB 13.412 billion, 生益科技 (Shengyi Technology, 600183) RMB 12.169 billion +7.75%, 工业富联 (Foxconn Industrial Internet, 601138) RMB 11.506 billion +4.52%. This is a list ranked by "where the money went," dependent on no sector classification at all, and it agrees with the conclusion the sector basis gives.

⚠️ Components 63 up 0 down and semiconductors 155 up 30 down — this combination of "an overwhelming advantage in advancers + large net inflow" is strong evidence that a theme is real (breadth, not a single ETF or a single leader). Per the lesson of memory [[volume-threshold-must-be-notional-not-shares]], sector conclusions use stock-by-stock breadth, not a single representative name.

5.4 Single-Stock Flows (THS basis, RMB billion)

⚠️ All 8 semiconductor equipment stocks saw positive net fund flows today (+RMB 105 million to +RMB 417 million). The pre-market second-layer criterion was "if net inflows remain concentrated in 北方华创 alone, this is leader-siphoning rather than sector repair" — today it was inflows across the board, which is sector repair, not siphoning. This one was called correctly.

---

6. Outlook for Tomorrow

6.1 Judgment on Theme Continuity

6.2 Focus for Tomorrow (branch + stocks + verification points)

Position discipline (carrying over the pre-market report's self-constraint, tightened by today's lessons): this theme has only 1 day of history and zero ladder height, making it a textbook short-duration branch. The cap on watch slots in this section is set at 3, and each one's falsification condition is written out explicitly. This section gives no position sizing or entry/exit timing advice.

⚠️ Agriculture and banks are deliberately left out of this section. Agriculture (the planting line) faces a divergence point tomorrow after a 3-board one-line move, and having already missed this line once pre-market does not mean it is now right to chase in the opposite direction — 亚盛集团 (600108) combines a 3-board one-line move with drought-driven earnings risk, i.e. "the height is already there + a hard fundamental flaw," which makes it something to observe, not something to focus on. Banks are an explicit avoid.

⚠️ Are the three verification points independent of each other? — Partly not, and that is stated here. ① and ② are two legs of the same variable ("AI hardware" — optics vs PCB); ③ is a different variable (memory demand). So this list actually has 2 independent variables, not 3. Per memory [[one-variable-watchlist-fakes-independence]] this must be written out, otherwise if the AI hardware line ebbs as a whole tomorrow, 2/3 of the list fails at once rather than 1/3.

6.3 What to Avoid Tomorrow

1. Banks / insurers / brokers (the entire financial complex) — 0 up 42 down / 0 up 5 down / 1 up 48 down today, with RMB 10.09 billion of net outflows combined. Not one of the four approval procedures for the RMB 360 billion recapitalization has been completed, and there is no new near-term timed catalyst.
2. 龙版传媒 (605577, SSE main board) and the publishing coattail names — 6 boards, the highest consecutive-board count marketwide; the SSE regulatory warning has not been lifted, AI video revenue is <0.01% of the total, and H1 net profit was −34.46%. Position risk now far exceeds theme risk.
3. The three "limit-up with net outflow" names — 剑桥科技 (603083, SSE main board, net outflow of RMB 1.703 billion), 景旺电子 (603228, SZSE main board, −RMB 169 million), 敦煌种业 (600354, SSE main board, −RMB 109 million). Sealed but the money is leaving; the tier most likely to gap down tomorrow.
4. The livestock-line leaders — 牧原股份 (002714) and 温氏股份 (300498) closed lower today while the agriculture theme surged; the theme has switched legs, don't wait on the old one.
5. Precious metals — −2.77% today, weakest marketwide, 0 up 14 down; the FOMC is September 15–16, so this week is still a rate-hike-expectation pricing week.
6. High-priced stocks with share unlocks — 摩尔线程 (688795, STAR Market ±20%) fell limit-down −20.00% today on the unlocking of 25.7745 million shares, making a new post-listing low. Tomorrow one should proactively sweep the unlock calendar for high-priced tech names; this was the only large-cap stock at a limit-down magnitude today.
7. The remnants of the digital-RMB names with extreme turnover — 楚天龙 (003040) had a 25.01% turnover rate today and failed to seal, 四方精创 (300468) −6.69%, 恒宝股份 (002104) −3.70%. The pre-market Pass on this line was correct; do not go back on it just because 楚天龙 rose 8.66%.

6.4 Input Prompts for Tomorrow Morning's Pre-Market List (mandatory items)

1. ⚠️⚠️ Top priority: the overnight US gainer list must be re-classified into five buckets — equipment / memory / compute / interconnect and optics / foundry — and any empty bucket must be written up explicitly as "no notable move in this stage," with nothing merged into "other." Today's entire lesson condenses into one sentence: MRVL +7.05% was written in our own table at the time, and it was buried by the "other" column.
2. ⚠️ A "calendar" must be scanned in addition to the news window. The root cause of missing CIOE (9/9–11) this time is structural: pre-market scanned "announcements newly issued between 9/4 15:00 and 9/7 07:00," while industry expos / industry conferences / unlock dates / index rebalances are not newly occurring events, they are events fixed long in advance. The recommendation is to add a permanent section to the pre-market process, "calendar events over the next 5 trading days"; this single item guards against today's kind of miss better than any single news item.
3. Tomorrow's pre-market report must answer three questions before writing any branch:
   • How many of the 14 component first boards advanced to a 2nd board? (the only hard indicator of this line moving from ignition into fermentation)
   • Which industry does the highest consecutive-board stock marketwide belong to? (today's answer is publishing at 6 boards, belonging to no theme at all — that signal is itself important)
   • What are the mean and median moves today of yesterday's (9/7) 93 limit-up stocks? (today those figures were +4.89%/+5.13%, the single most sensitive indicator for judging sentiment)
4. Exclusive attribution for CPO / optical modules / PCB is still unobtained and must be closed out tomorrow. Already excluded: 0 in-window company announcements, and three same-day financial pieces all giving no cause. Candidates: ① the overnight MRVL +7.05%; ② the CIOE opening on 9/9. Tomorrow one should check the 9/7 overnight performance of the US optical communications names (COHR/LITE/AAOI) — if they keep surging, attribution leans toward ①; if they are flat while A-shares keep rising, attribution leans toward ② (a domestic calendar event). This is a clean criterion capable of separating the two candidates.
5. Attribution for 建行 (601939) −2.46%, the worst of the six, was not obtained — on the books.
6. The LHB (9/7) needs to be backfilled after 18:00, together with the seat structure of today's three "limit-up with net outflow" names — that is the most direct evidence for judging follow-through tomorrow. It is also recommended that the post-close recap slot be rescheduled to 18:10, otherwise the LHB will never be obtainable (this is now the third time).
7. The primary source for "August breeding-sow inventory −2.94% MoM" is still unclosed (an item carried over from the pre-market report). Today's agricultural driver has shifted to the planting / grain-price line, so the marginal importance of that figure has declined, but it should still be settled.

---

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