A-Share Market Recap | 2026-09-02 Wednesday

Scope and caliber boundaries of this piece (please judge conclusion strength accordingly) ① Quotes are Tencent qt.gtimg.cn closing-settlement caliber, timestamps 20260902153031–20260902153055, later than the 15:00 close. The price triple has been reverse-computed and verified name by name (e.g. 苏垦农发 9.88×0.9089=8.98 ✓, 敦煌种业 9.12×0.9002=8.21 ✓). ② A "vs. pre-market price" column has been added to the reconciliation. The pre-market piece was published at 07:00, and the earliest price a reader could actually execute at that day is the opening price from the 09:25 call auction, not the previous close. Therefore every name in this piece carries two figures at once — "closing change (vs. previous close)" and "open→close (vs. today's opening price)" — because using only the former for reconciliation systematically overstates or understates the hit rate. ③ Limit-ups / consecutive limit-ups / sealing orders use the East Money official limit-up pool, 9/2 caliber, verified consistent at tc=52 / len(pool)=52, and stable across two consecutive pulls. ⚠️ There is a caliber discrepancy with the media close reviews' "54"; the cause has been identified and is explained in §1 — it is not a data failure. ④ Sector changes, sector net inflows and advancer/decliner counts use the THS 90-industry-sector closing snapshot caliber; single-stock net main-force flow uses the THS "real-time" caliber. The two differ from East Money's "net main-force inflow" caliber, and must not be subtracted across sources. ⑤ ⚠️ The LHB (Dragon-Tiger List) was not obtained for this piece. The exchanges only disclose it after 18:00, and this piece was generated at 15:30; both stock_lhb_detail_em and stock_lhb_jgmmtj_em returned empty. This is a scheduling problem, not an interface failure (yesterday's pre-market piece did obtain the 79 rows for 9/1 at 07:00 the next morning — confirmed). Therefore §5 of this piece contains no seat-level conclusions at all, and there is not a single piece of "institutional net buy / net sell" evidence of the kind in yesterday's pre-market piece. ⑥ No northbound flow data: since 2024-08-19, daily net buy figures for the Stock Connect have ceased to be disclosed; this piece gives no northbound figures and cites no interface value returning 0.0. ⑦ For the defense branch that led today's gains, this piece could not obtain a confirmable in-window catalyst. The primary announcement pages of 内蒙一机 and 长城军工 have been checked, and neither company has issued a new announcement since 8/28 and 8/21 respectively; the "Norinco Group asset injection" material retrieved was checked and is entirely old 2025 text, and is not used. This piece reports only the fact that "defense led today" plus its capital structure, and offers no attribution — see §3 note ②.

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

I. Interface failures that actually occurred this run

• East Money stock_zh_index_spot_em returned RemoteDisconnected; index quotes were switched to Tencent qt.gtimg.cn, which worked.
• East Money stock_zh_a_hist likewise returned RemoteDisconnected (all 4 symbols failed); single-stock daily bars were switched to Tencent web.ifzq.gtimg.cn/appstock/app/fqkline, and that interface is also unstable (2/5 returned abnormal structures, 1/5 timed out). In the end consecutive-limit-up height was read directly from the lbc field carried by the limit-up pool itself, without relying on recomputation from daily bars — which turned out to be more reliable.
• THS single-stock fund flow had two degenerate rows: 竞业达 (003005) and 国芳集团 (601086) returned inflow = 0.00 and net = −turnover; both are limit-up stocks sealed all day with extremely low turnover rates of only 3.27%/0.87%. These are degenerate rows, not genuine net outflows, and this piece does not cite fund-flow data for these two. This falls under the "freshness/validity must be checked down to symbol granularity" category.
• The third push2ex pull hit an SSL handshake timeout, but the first two runs agreed (tc=52/len=52, and 600540 was absent from the pool in both), which is enough to confirm stability.

II. Two search contaminations intercepted this run (one of which would have manufactured a nonexistent announcement)

1. ⚠️ The single most critical one: the East Money announcement interface tagged 阳光股份 (000608)'s notice_date as 2026-09-02, with the title "Announcement on the Signing of a Major Day-to-Day Operating Contract" — which looks like a second major contract newly added today. Pulling the art_code for comparison shows it is AN202609011828889050, the very same piece as the RMB 633 million server leasing contract cited in §5.10 of yesterday's pre-market piece (published 9/1 18:28). East Money's notice_date uses the "next trading day" rather than the publication day. Copying it verbatim would have conjured up "阳光股份 signed another major contract today" and attributed the +4.69% to a nonexistent piece of news. How to see through it: compare the timestamp inside the art_code, do not look at notice_date. The same trap also appeared on 万向德农 (a risk-warning announcement with notice_date 9/2 whose art_code is AN202609011828885603, actually published 9/1).
2. Searching "reasons for the September 2 defense rally" returned results that are entirely old text from 2025 and from March/April/July 2026 (analysis of 内蒙一机's April limit-up, long pieces on Norinco Group restructuring, etc.), not one of which falls inside this window. All have been discarded, and this piece gives no attribution for defense. ⚠️ But per the "no news found ≠ no news" lesson in memory, this piece's wording is "no confirmable catalyst obtained", not "there is no catalyst", and "re-check the defense catalyst" has been listed in §6 as the first to-do for tomorrow morning's pre-market.

III. Channels that worked

• Limit-up pool push2ex getTopicZTPool (dpt=wz.ztzt, sort=fbt:asc, date=20260902) tc=52/len=52, consistent across two pulls.
• Failed-seal stock_zt_pool_zbgc_em returned 15 rows and limit-down stock_zt_pool_dtgc_em returned 8 rows; both had complete column structures, not empty shells. ⚠️ Note the difference from yesterday: yesterday the limit-down pool returned 0 rows, which at the time could not be distinguished from "genuinely no limit-downs", so yesterday's pre-market piece gave no limit-down count; today it returned 8 rows of valid data, so today's is given. Same criterion, different conclusion, purely because the returned structure differed.
• Today's quotes were pulled name by name for the 9/1 limit-up pool (83 names) to compute follow-through statistics; all 83 matched successfully, 0 missing.
• Sectors used stock_board_industry_summary_ths() (90 rows) and single-stock fund flow used stock_fund_flow_individual(symbol='real-time') (5,208 rows); the East Money clist series continues to be bypassed via the established backup channel and was not attempted again.

IV. Three cross-checks actively performed on method that changed the conclusion

1. The 52 vs. media 54 limit-up-count discrepancy has a specific identified cause; it is not a data failure. 新赛股份 (600540) closed at 6.73 = the limit-up price, +9.97%, yet is not in East Money's official limit-up pool — because it is in stock_zt_pool_zbgc_em with 6 seal breaks, only resealing at the close. East Money's limit-up pool has exclusion logic for names that break their seal repeatedly during the day. This piece therefore gives both calibers in §1 and explains the source of the difference, rather than simply picking one number. ⚠️ This also directly affects the §2 reconciliation: 新赛股份 counts as a limit-up on the "did it limit up" measure, and as the worst possible form — 6 seal breaks all day — on the "seal quality" measure; both must be written together.
2. The follow-through statistics were changed from "ranked by industry mean" to "look at the full sample first". The draft was going to follow yesterday's pre-market approach of ranking industry means, but among today's 45 industries 31 have a sample size of n≤2, so the mean is entirely dominated by individual names (e.g. "ground weaponry +10.01%" is actually n=1). Per the "sample size must always be read alongside" lesson in memory, the full-sample fact (31 up, 52 down) has been moved to the front and industry means demoted to a footnote with n mandatorily labeled.
3. The recommendation-group mean was only computed on the closing caliber at first, −1.86%; after adding the open-execution caliber it is −1.47%, a difference of 0.39pp. The entire difference comes from 苏垦农发 (which kept falling after gapping down −1.82%). This piece gives both figures, and on that basis writes the §2 self-criticism — the three-tier contingency the pre-market §8.1 set for 苏垦 (gap up 0–3% / gap up >5%) had no tier covering "gap down", and gap down is exactly what happened.

V. What this piece did not do (listed honestly)

• The 9/2 LHB and institutional-seat data were not obtained (only disclosed after 18:00), so conclusions such as "institutions net sold RMB 127 million of 敦煌种业 yesterday" cannot be tracked into the next day. Recommend that tomorrow morning's pre-market piece re-pull the 9/2 LHB once; this is the biggest evidence gap in this piece.
• The in-window catalyst for the defense branch was not obtained (see above).
• No name-by-name fundamental check was done on today's 52 limit-up stocks; only ladder structure and seal quality were verified.

---

0. One-Sentence Recap

Sentiment: turning from euphoria to ebb, but this is not a crash — it is a rotation of the leading theme. The Shanghai Composite −0.97%, ChiNext Index −2.39%, combined turnover of RMB 1,791.179 billion, RMB 242.2 billion lighter than yesterday; limit-ups fell from yesterday's 83 to 52, limit-downs 8, failed seals 15, the seal-break rate rose from yesterday's 6.7% to 22.4%, and the highest consecutive limit-up count fell from 7 to 4.

The strongest theme was none of the ones judged pre-market — it was defense. Among 90 industry sectors, defense equipment +2.06% ranked 1st and defense electronics +0.59% ranked 8th, while defense electronics had the largest net inflow of the whole market (+RMB 190 million) — on a day when the 90 sectors combined saw net outflows of RMB 95.964 billion and only 6 sectors had positive net inflow, that position says more than the price change. None of the pre-market piece's five branches was defense; this is the single biggest miss in this piece.

Pre-market list hit summary: 21/30 (70%) overall, but the structure is "big financials right, agriculture badly wrong, Pass calls most right of all." Of the 5 recommendation slots, 3 beat the Shanghai Composite; the fatal one is the No. 1 slot, 苏垦农发 (601952) −9.11%, which single-handedly dragged the recommendation-group mean from −0.04% down to −1.86%. Passing on the entire seed sub-branch pre-market was correct (that group was broadly −5% to −11% today), but at the same time it staked the one remaining agriculture recommendation slot on the same receding theme — the direction was half right, and the position was placed on the wrong half.

Tomorrow's tone: defensive, and the theme must be rotated. The agricultural ebb is confirmed (crop farming & forestry −4.41%, dead last of the 90 sectors, with 0 crop-farming names in the limit-up pool and 2 in the limit-down pool); continuing to avoid semiconductors is correct (−1.34%, net outflow RMB 4.895 billion, 27 up 158 down); the only thing worth prioritizing tomorrow morning is defense — and defense is precisely the thread with the thinnest evidence in this piece: the best price action, the most genuine money flow, and no catalyst to be found.

---

1. Market Overview

1.1 Indices and Volume

Combined turnover of the two markets was RMB 1,791.179 billion, roughly RMB 242.2 billion lighter than yesterday.

Cross-caliber validation: this piece sums Tencent's per-index turnover to get 835.368 + 955.811 = RMB 1,791.179 billion; the Jiemian close review¹ independently gives RMB 1,791.178 billion, RMB 242.224 billion lighter. The two independent channels differ by RMB 1 million (rounding), which is credible.

⚠️ One divergence that should not be averaged away: the main boards fell across the board, yet the BSE 50 rose 2.50%. Among today's 52 limit-ups, 3 were BSE 30cm limit-ups — 智新电子 (Zhixin Electronics, 920212, +29.95%, consumer electronics), 维琪科技 (Weiqi Technology, 920176, +29.99%, cosmetics), 晟楠科技 (Shengnan Technology, 920006, +29.96%, aviation equipment). The pre-market piece did not cover the Beijing Stock Exchange; this is the second structural blind spot (the first is defense, see §3).

1.2 Sentiment Temperature

⚠️ Caliber discrepancy in the limit-up count (cause identified — please do not treat it as conflicting data): East Money's official limit-up pool gives 52 (tc=len(pool)=52, consistent across two consecutive pulls), while the Jiemian close review¹ gives 54. One confirmed source of the difference is 新赛股份 (Xinsai Co., 600540): closing price RMB 6.73 = the limit-up price, +9.97%, but it broke its seal 6 times during the day and only resealed at the close, so it is not in East Money's official limit-up pool. The body of this piece uses the official pool's 52 throughout, and in the §2 reconciliation 新赛股份 is tagged with both facts — "sealed at the close" and "broke seal 6 times all day".

Sentiment judgment: the first ebb day after a peak, but not a freezing point.

• Supporting "ebb": limit-ups halved, the seal-break rate up 3.3×, the highest consecutive count cut from 7 to 4, turnover RMB 242.2 billion lighter, 3,901 names down.
• Supporting "not a freezing point": still 52 limit-ups, an intact 4-board ladder (2 names at 4 boards / 4 at 3 boards / 7 at 2 boards), the BSE 50 up 2.50% against the tide, and defense forming a new money-making effect.
• The most accurate description is "theme rotation", not "sentiment collapse": the largest cluster in yesterday's limit-up pool was crop farming with 8 names, and today crop farming has 0; today's limit-up pool has auto parts 5, power grid equipment 4, and ground weaponry + aviation equipment 4 combined. The money did not leave, it just moved.

---

2. Pre-Market List Reconciliation

Reconciliation caliber: "vs. pre-market price" = the change from today's opening price to the closing price, i.e. the real result for a reader who acted on the 07:00 pre-market conclusion and bought in the 09:25 call auction. "Delivered" is judged by whether it beat the Shanghai Composite (−0.97%), with absolute return noted separately.

2.1 The 5 Recommendation Slots (pre-market §9.1)

Recommendation-group mean: −1.86% on the closing caliber, −1.47% on the vs.-pre-market-price caliber (Shanghai Composite −0.97%). The recommendation group underperformed the market by 0.89pp overall. ⚠️ But excluding 苏垦农发, the 4 big financials average: −0.04% at the close, +0.02% vs. pre-market price, beating the Shanghai Composite by 0.93pp.

This line must come first: today was not "the big-financials call was wrong", it was "one agricultural stock ate all the excess return the four big financials earned." Pre-market gave 4 of the 5 slots to big financials and 1 to agriculture, and that 1 lost more than 8× what the 4 earned.

2.2 The 7 "Watch Only" Names (pre-market §6.2–6.4)

Decision rule: underperforming the Shanghai Composite = the "do not buy" call was correct.

"Watch only" group: 5/7 correct (71.4%), group mean −5.44% (−7.52% vs. pre-market price), a heavy underperformance of the Shanghai Composite — for a set labelled "not recommended for purchase", that is correct.

2.3 The 18 Pass / Avoid Names

Pass group: 13/18 correct (72.2%), group mean −5.31% (Shanghai Composite −0.97%), a heavy overall underperformance of the market — the direction of the Pass calls was right.

2.4 Hit Rate and Group-Level Self-Criticism

Overall hit rate: 21/30 = 70.0% (recommendations 3/5 + watch only 5/7 + Pass 13/18).

⚠️ But breaking the Pass group down by pre-market's own categories reproduces the same pattern for the third time today:

The "position too high" group averaged exactly 0.00% today and beat the Shanghai Composite — the third time this pattern has held. The pre-market piece's own §6.2 wrote "do not read Pass as will-fall", citing the historical +6.18% / −0.21% contrast from reconciliation. Today it held again: evidence-based Passes (institutions are selling) averaged −10.76%, position-based Passes averaged 0.00%, a gap of 10.76pp. ⚠️ But the sample size must be stated alongside: these five groups have only 3–4 names each, drawn from a single trading day; they do not constitute a statistical prior and can only serve as directional corroboration. Holding three times does not mean the pattern is solid, only that "it has not yet been falsified."

Four self-criticisms, ranked by importance:

① The biggest one: the branch call was half right, yet the position was staked on the wrong half. Pre-market used two pieces of primary evidence (corn futures 9/1 −0.26%, and the abnormal-movement announcements from 万向德农 and 登海) to Pass the entire seed sub-branch — and that judgment was fully validated today: 登海 −8.50%, 隆平 −5.82%, 敦煌 −9.98% (limit-down), 农发种业 −9.75%, 神农 −11.39%. But it simultaneously kept one agriculture recommendation slot for the "grain-selling" 苏垦农发, and today's hardest-hit name was precisely that one (−9.11%). Where the problem lies: pre-market treated "selling grain vs. selling seed" as the main axis of risk, but today's actual risk axis was "agriculture vs. non-agriculture". Crop farming & forestry −4.41% ranked 90th and dead last, agrochemicals −2.55% ranked 88th, animal husbandry −2.02% ranked 83rd, agricultural product processing −1.36% ranked 70th — all four agricultural sub-sectors sat in the bottom 25%; a branch-level ebb does not care whether you sell grain or seed. How to fix this next pre-market: when a theme has already been judged by oneself to be at "week 3 of pricing + net main-force outflow of RMB 1.131 billion + institutions net selling on the LHB + already having limit-downed collectively once", the correct action is to zero out the entire branch, not to pick a "cleaner" name inside it. When a branch ebbs, the weight of single-stock quality in the ranking drops to very little.

② The packaged "big financials" branch was pulled apart today, and once pulled apart the three pieces moved in completely different directions. Banks −0.22% (rank 24), insurance −1.11% (rank 57), brokers −2.13% (rank 84, net outflow RMB 5.425 billion, the largest of the whole market). Pre-market wrote all three as a single branch and gave 2 bank and 2 insurance names across the 4 recommendation slots. The result: both banks (邮储 +1.46%, 中行 +0.15%) delivered, while of the two insurers (太保 −0.73%, 平安 −1.05%) one weakly delivered and one did not. How to fix this next pre-market: "big financials" is not one branch, it is three. Yesterday's fund-flow data had already written the difference out — banks +RMB 8.381 billion vs. insurance +RMB 2.716 billion is a 3× gap, yet pre-market allocated slots at the same intensity.

③ None of the three contingency tiers pre-market set for 苏垦农发 covered what actually happened. Pre-market §8.1 wrote: "with a gap up of 0%–3% the premise still holds; with a gap up >5% the odds turn poor and it is downgraded to watch"; §8.4 wrote "gap-up dive: 苏垦农发/深粮控股 turning red-to-green within half an hour after gapping up → the agriculture branch is void for the day." What actually happened was a gap down of −1.82%. All three tiers presupposed a gap up, and the gap-down tier was empty, so the very downgrade signal that should have triggered that day had no corresponding clause. How to fix this next pre-market: verification points must cover the "worse than expected" side. A contingency plan that only tiers the upside is no plan at all. (深粮控股, for its part, followed the standard gap-up dive: open +2.02% → close −9.34%, so §8.4 item 1 holds for it.)

④ Once again, the industry distribution of the limit-up pool was only used to count the target branch. Pre-market obtained the 83-name 9/1 limit-up pool and ran industry statistics, but the body only used it to count "14 agriculture-linked names". In the same data set, general retail had 6 (of which 国芳集团 was already at 3 boards) and auto parts had 4, and those two clusters today produced respectively the highest consecutive limit-up count of the whole market (国芳集团 at 4 boards, with a seal-to-turnover ratio of 4.752, the highest of the market) and the largest cluster in today's limit-up pool (auto parts, 5 names). How to fix this next pre-market: make "which industry does the market's highest consecutive limit-up stock belong to" and "which are the 2nd- and 3rd-largest industry clusters in the limit-up pool" mandatory questions, rather than only tallying the branch one is already writing about.

---

25 Special Focus: 阳光股份000608the Pass call was wrong, but what was wrong was not the nine pieces of evidence

Pre-market used nearly two thousand characters and nine pieces of primary evidence to downgrade it from a recommendation slot to Pass, and explicitly wrote one forecast: "It will very likely gap up or even limit up today. 'Pass' is not a prediction that it will fall."

What actually happened today: open 7.46 (+2.90%), close 7.59 (+4.69%), high 7.66 (+5.66%), net main-force flow +RMB 58 million, turnover rate 7.85%.

⚠️ First, a fact this piece nearly got wrong. The East Money announcement interface today listed under this stock an "Announcement on the Signing of a Major Day-to-Day Operating Contract" with notice_date = 2026-09-02, which looks like another new contract signed today. Pulling the art_code for comparison confirms it is AN202609011828889050, the very same RMB 633 million contract published 9/1 at 18:28 that the pre-market piece cited — East Money's notice_date uses the next trading day rather than the publication day. There was no new announcement today; the +4.69% corresponds to no new news.

Now the verification point pre-market set. §8.1 wrote: "if the gap-up magnitude prices in the full RMB 633 million (gap up >7% or a locked limit-up), the market is reading the headline; if it gaps up <3% and then weakens, the market has read the absence of a minimum-guarantee clause". What actually happened was a gap up of +2.90% (<3%), but it did not weaken — it rose all day to close +4.69%. Neither preset branch was hit. The correct reading is a third case: the market neither priced the RMB 633 million headline (or it would have locked limit-up) nor rejected it (or it would have drifted lower), but gave it a "small-position trial" pricing. Pre-market had not preset this state, yet it is in fact the most common one.

On how this wrong call should be booked, one point must be made clearly:

• Not one of those nine pieces of evidence was falsified by today's price action. No minimum unit count and no guarantee clause, a signing entity with negative net assets bought for RMB 100,000, an anonymous unguaranteed counterparty, annualized gross profit attributable to parent capped at RMB 15.05 million, and the company's own statement that it "does not constitute a material impact" — these are all judgments about the value of the contract, and one day of share price can neither verify nor falsify them.
• ⚠️ But pre-market landed that judgment on the trading label "Pass", and in this list's system Pass carries the meaning "do not participate". Today's +4.69% shows that pre-market implicitly assumed a nonexistent equals sign between "how much is this contract worth" and "will this stock rise tomorrow". This is the same structural problem as "position-too-high Passes outperform" in the §2.4 table, in two different guises — the fundamental conclusion and the short-term direction are two orthogonal dimensions.
• One observation that still holds: as of this piece's data pull (15:30), the SZSE had not issued a letter of concern, so the thing pre-market §5.10 flagged for monitoring did not happen today. ⚠️ But letters of concern are usually issued after the close, and this piece's data-pull time is earlier than their common issuance window; "not seen" does not equal "will not come", and it must be rechecked tomorrow morning.

---

3. Verification of Today's Leading Themes

Note ① Did pre-market call the strongest theme? No. Pre-market's branch No. 1 was big financials (today banks rank 24, insurance 57, brokers 84), and branch No. 2 was agriculture (today rank 90). Today's genuinely strongest theme was defense, and none of pre-market's five branches was it. ⚠️ But one point in fairness to pre-market: it picked the right axis of judgment — "the direction money is going". Pre-market wrote that "the capital caliber and the price caliber must be read separately today", and on that basis demoted the price-strongest agriculture to No. 2 and ranked the capital-strongest direction first. That methodology continued to hold today — it is just that yesterday it was applied to banks, while today the answer from fund flows became defense electronics (+RMB 190 million, the market's only significant positive net inflow among industry sectors). The method was right, the target changed.

Note ② ⚠️ On defense leading the market, this piece gives facts only and no attribution, and the reason must be spelled out. Verified facts: ① the defense equipment sector was +2.06%, ranked 1st of 90 sectors, 56 up 24 down; ② defense electronics' +RMB 190 million net inflow was the largest of the market, while the 90 sectors market-wide had combined net outflows of RMB 95.964 billion, with only 6 sectors positive; ③ the limit-up pool held 2 ground-weaponry and 2 aviation-equipment names; ④ 内蒙一机 (600967) is on its 2nd consecutive limit-up day (first board 9/1, second board 9/2), 长城军工 (601606) turned over RMB 2.140 billion, and 博云新材 (002297) turned over RMB 1.316 billion with net main-force inflow of +RMB 69 million — all genuine limit-ups with volume, not thin-volume markups. Not obtained: this piece checked the primary announcement pages of 内蒙一机 and 长城军工, and their most recent announcements stop at 8/28 and 8/21 respectively, with no new disclosure in the window; the "Norinco Group asset injection" material retrieved was checked and was all published in 2025, outside this window, and has been discarded. This piece therefore offers no catalyst attribution. ⚠️ One calendar fact, stated without any causal inference: September 3 is the Victory Day of the Chinese People's War of Resistance Against Japanese Aggression (established in law in 2014). This piece does not claim any causal relationship between that and today's market — and this is precisely the first thing tomorrow morning's pre-market must go back and do (see §6④). Per "no news found ≠ no news", this piece's wording is "no confirmable catalyst obtained", not "there is no catalyst".

---

4. Limit-Up Ladder and High-Board Names

4.1 Ladder Structure

⚠️ Which industry does the market's highest consecutive limit-up stock belong to: general retail (国芳集团) and IT services (竞业达); neither belongs to today's leading defense theme, nor to yesterday's leading agriculture theme. This is exactly the same structure as yesterday — yesterday's highest, the 7-board 海鸥住工, belongs to household products, likewise unrelated to the leading theme. Two consecutive days of "the highest consecutive limit-up stock does not overlap with the strongest theme" indicate that high-board names are driven by single-stock events and do not play the role of the theme's core force.

国芳集团 (601086) had the best seal quality today: sealing order RMB 306 million / turnover RMB 64 million, a seal-to-turnover ratio of 4.752, the highest of the market, with a turnover rate of only 0.87%. It has now limit-upped 4 sessions in a row (8/28 +10.03%, 8/31 +9.95%, 9/1 +10.03%, 9/2 +10.01%), and issued an abnormal share price movement announcement on 9/1 and a further share trading risk warning announcement on 9/2 — exactly the same form as 万向德农 yesterday: the company issues risk warnings one after another, and the share price keeps sealing.

4.2 Broken Boards and Failed Promotions (the most important table today)

Of yesterday's 5 names at 4 boards or above, not one advanced today. The highest consecutive limit-up count fell from 7 to 4, and the two names at 4 boards (国芳集团, 竞业达) are yesterday's 3-board names advancing in place, not a new height.

4.3 Follow-Through Win Rate (full sample)

Yesterday's 83 limit-up stocks today: 31 up, 52 down, mean −0.76%, median −2.12%, open→close mean −1.74%. Follow-through win rate 37.3%.

The downgrade line set in pre-market §8.4 item 4 was "deteriorating below 30%"; the actual figure is 37.3%, not triggered but clearly deteriorated. ⚠️ The median of −2.12% is lower than the mean of −0.76%, meaning a handful of big gainers (e.g. 新赛 +9.97%, 恒宝股份 +10.02%) pulled the mean up, and the typical stock's real experience was worse than the mean. And the open→close mean of −1.74% means: yesterday's limit-up stocks generally gapped up and drifted lower today, and those chasing them lost more than the closing change suggests.

4.4 Core-Leader Identification

Screening on "large turnover + rising + recognizable", today's core-leader candidates cluster in defense and power grid equipment:

Core-leader judgment: 博云新材 (002297) is the highest-quality name today — limit-up + RMB 1.3 billion turnover + net main-force inflow of RMB 69 million + a seal-to-turnover ratio of 0.212, above the market median of 0.156. ⚠️ But it was still −1.89% yesterday and only took its first board today, making it a "just started" rather than an "established" core leader; its recognizability awaits confirmation tomorrow.

---

5. Fund Flows

5.1 Sector Flows (THS 90-industry-sector caliber)

⚠️ Market-wide, the 90 sectors' net inflow totalled −RMB 95.964 billion, with only 6 sectors positive. This is today's single most important aggregate fact — it is not one theme bleeding, it is the whole market bleeding.

Two readings: ① Defense electronics' +RMB 190 million is small in absolute terms, but it is the market's only industry direction where "sector up + positive net inflow + a limit-up ladder" all hold at once. Against an aggregate backdrop of −RMB 95.964 billion, the relative meaning of +RMB 190 million far exceeds its absolute size. ② Brokers' −RMB 5.425 billion was the market's largest outflow, and the pre-market piece did not cover brokers. Brokers are a coincident indicator of market sentiment, and 1 up 49 down + the market's largest outflow is a signal that should have entered the pre-market risk list.

⚠️ The direction of fund flows reversed completely between yesterday and today, and this must be stated plainly: yesterday banks ranked first market-wide with +RMB 8.381 billion of net inflow, and today banks show a net outflow of −RMB 221 million (though the sector fell only −0.22% and ranked 24, still resilient). Pre-market treated "banks +RMB 8.381 billion" as the core basis for the No. 1 recommendation slot, and that basis vanished within a single day. This points to the same thing as self-criticism ② in §2.4 of this piece: single-day fund-flow extremes have no cross-day persistence, and pre-market §2.1 used one as the main axis for branch ranking, treating a high-frequency noise as a low-frequency signal.

5.2 Single-Stock Net Main-Force Flow (THS "real-time" caliber)

Largest net outflows (within the list):

Largest net inflows (within the list):

⚠️ Data quality note: 竞业达 (003005) and 国芳集团 (601086) returned inflow = 0.00 and net = −turnover from this interface; both are limit-up stocks sealed all day with turnover rates of only 3.27% / 0.87%, so these are degenerate return values rather than genuine net outflows, and this piece does not cite fund-flow data for these two.

5.3 LHB and Northbound Flows

• ⚠️ LHB: not obtained for this piece. The exchanges disclose 9/2 data only after 18:00, and this piece was generated at 15:30, so the interface returned empty. This is not an interface failure, it is a scheduling-timing problem (yesterday's pre-market piece successfully obtained the 79 rows for 9/1 at 07:00 the next morning — confirmed). This creates a piece-wide evidence gap: seat-level conclusions from yesterday's pre-market piece such as "institutions net sold RMB 126.57 million of 敦煌种业" and "institutions had 0 buys and 1 sell in 芒果超媒" cannot be tracked or verified at all today. 敦煌种业 limit-downed today and 芒果超媒 gapped up and drifted lower, directionally consistent with yesterday's institutional-distribution judgment, but that is only price corroboration, not confirmation from seat data. Recommend tomorrow morning's pre-market piece re-pull the 9/2 LHB.
• Northbound flows: since 2024-08-19, daily net buy figures for the Stock Connect have ceased to be disclosed; this piece gives no figures and cites no interface value returning 0.0.

---

6. Tomorrow's Outlook (2026-09-03 Thursday)

① Theme Continuity Assessment

② Tomorrow's Watch List (with sectors and verification points)

⚠️ Advance statement: the evidence strength for every name in this section is weaker than in yesterday's pre-market piece, because the LHB was not obtained and the defense catalyst was not obtained. Please read at a lower confidence level.

③ Tomorrow's Avoid List

1. The entire agriculture complex (regardless of grain-selling or seed-selling) — crop farming & forestry (90), agrochemicals (88), animal husbandry (83), agricultural product processing (70), all four sub-sectors in the bottom 25%. ⚠️ Special note: today's limit-down 敦煌种业 (600354) was a locked limit-down (opening price = high = the delivery of yesterday's pre-market judgment), and 国投丰乐 (000713) had its seal broken 23 times during the day; forms like these still carry inertia tomorrow.
2. 万向德农 (600371) (SH main board) — +3.30% today, closing green against the tide, but after the close it announced "completion of the industrial and commercial registration for the company name change" (published 9/2, art_code AN202609021828919713). ⚠️ This piece did not obtain the new post-change name and makes no thematic inference whatsoever; it only flags this: a seed company that has broken a 6-board run, that describes its own share price as "severely deviating from fundamentals and a reasonable valuation range", and that has issued 3 abnormal-movement announcements plus 2 risk warnings, closing green against the tide on an ebb day while completing a name-change registration — that combination has more uncertainty than analyzability.
3. Brokers (1 up 49 down, net outflow RMB 5.425 billion) and semiconductors (27 up 158 down, net outflow RMB 4.895 billion).
4. Indiscriminate follow-through on yesterday's limit-up stocks — win rate 37.3%, open→close mean −1.74%.
5. New listings on day two — 马矿股份 (601123) −23.00% today, 贝特利 (301697) −10.64%; the pre-market Pass calls were delivered, and the day-two drawdown on new-listing speculation has kept widening this week.

④ Input Notes for Tomorrow Morning's Pre-Market List (by priority)

1. ⚠️ First priority: go back and find defense's in-window catalyst. Today it was the market's No. 1 gainer and the only industry direction with significant net inflow, and this piece could not find a catalyst. What to check: recent announcements and state-asset developments at Norinco Group / the AVIC complex, policies or events related to September 3, and why 内蒙一机's net main-force flow was −RMB 74 million on a limit-up day. Until a catalyst is confirmed, defense should not be given a recommendation slot — "money without a narrative" is the most common precursor form of a one-day wonder.
2. ⚠️ Second priority: re-pull the 9/2 LHB and institutional-seat data (available after 18:00). This is the biggest evidence gap in this piece. Focus on three things: ① whether institutions kept net selling 敦煌种业 on its limit-down day; ② whether the buy-side seats in the defense limit-ups (长城军工, 博云新材, 内蒙一机) were institutions or hot money; ③ who was sealing 新赛股份, which broke its seal 6 times and still sealed at the close.
3. Split "big financials" into three independent branches: banks / insurance / brokers. Today the three ranked 24 / 57 / 84 respectively, moving in completely different directions; continuing to package them will repeat today's positioning error.
4. Give 苏垦农发 (601952) a "gap down" contingency tier. Today it gapped down −1.82%, while all three pre-market tiers presupposed a gap up. If it gaps down again tomorrow, that is a continuation of the branch's ebb, not an oversold opportunity (crop farming saw RMB 2.192 billion of net outflow overall, with no sign of the bleeding stopping).
5. Make "which industry does the market's highest consecutive limit-up stock belong to" and "the 2nd- and 3rd-largest industry clusters in the limit-up pool" mandatory pre-market questions. For two days running the highest consecutive limit-up has been unrelated to the leading theme (海鸥住工 — household products, 国芳集团 — general retail), while today's largest cluster in the limit-up pool is auto parts with 5 names, followed by power grid equipment with 4, neither of which was covered pre-market.
6. Cover the Beijing Stock Exchange. The BSE 50 rose 2.50% against the tide today, and 3 of the 52 limit-ups were BSE 30cm limit-ups, yet none of the pre-market piece's branches touched it.
7. Recheck whether 阳光股份 (000608) received an SZSE letter of concern. It had not been seen at this piece's 15:30 data pull, but letters of concern are usually issued after the close, and "not seen" does not equal "will not come".

---

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