A-Share Market Recap | 2026-09-08 Tuesday

Scope and caveat boundaries of this report (judge conclusion strength accordingly) ① Quotes are on a 2026-09-08 close basis (Tencent qt.gtimg.cn); for each stock we give prev close / open / high / low / close, plus a separate "open → close" column — the pre-market watchlist is published before the open, so the open is the executable price; reconciling only on the close-basis change would systematically overstate the hit rate. ② Limit-up / failed-limit / limit-down pools are East Money's official pools push2ex, with len(pool)==tc verified (73==73). ③ Limit-up determination is tiered by board limit: Shanghai/Shenzhen main board ±10%, ChiNext/STAR ±20%, BSE ±30%. Every stock in this report is tagged with its board. ④ Sector data is given with two different rulers, each labeled: Tonghuashun stock_board_industry_summary_ths (its "net inflow" = inflow − outflow) and East Money push2delay clist (its "main-force net inflow" is a large-order caliber). The two cannot be subtracted across calibers. ⑤ ⚠️ Today's Dragon-Tiger List (LHB) was not obtained: the LHB publishes after 18:00 and this report was generated at 15:34; stock_lhb_detail_em('20260908') returned empty. This report gives no LHB figures for today; the capital-flow conclusions in §5 rest solely on the main-force net inflow caliber. ⑥ No northbound flow data (daily net buy disclosure was discontinued as of 2024-08-19); this report gives no northbound figures. ⑦ ⚠️ Hong Kong readings are a 15:22 intraday snapshot, not closing prices — HK closes at 16:00 and had not closed when this report was generated. The A/H comparison in §2.4 is discounted accordingly. ⑧ Commodity futures are 15:00 domestic close + CNBC international quotes, with one dead contract removed (see internal note).

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

1. Channels that worked

• Quotes via Tencent qt.gtimg.cn; limit-up / failed-limit / limit-down pools via push2ex, with len(pool)==tc verified per memory [[limit-updown-must-use-official-pool]] (73==73), failed-limit pool 37==37, limit-down pool 0==0.
• ⚠️ The East Money clist main cluster is still 502 (fifth consecutive trading day), but the backup domain push2delay.eastmoney.com is fully usable — the previous report fell all the way back to Tonghuashun per memory [[ths-fallback-when-em-clist-down]]; in fact the backup domain was working the whole time, so that fallback was premature. This report therefore obtained main-force net inflow on the East Money caliber (Tonghuashun's "net inflow" is inflow − outflow, a different ruler), giving the first dual-caliber cross-check of sector and single-stock flows. This should be written back to memory: when clist returns 502, try push2delay first before switching sources.
• Advance/decline counts via stock_market_activity_legu (official caliber 3257/1859/91).
• Commodity futures via ak.futures_zh_realtime + the CNBC quote endpoint (the substitute channel designated by memory [[yfinance-rate-limit-root-cause]]).

2. Failed channels / data-fetch faults this run

• ⚠️ ak.stock_board_industry_name_em() returned RemoteDisconnected, the same cluster fault, bypassed via push2delay.
• ⚠️ ak.stock_lhb_detail_em('20260908') returned NoneType not subscriptable — this is not an endpoint outage, it is generated at 15:34 while the LHB only publishes at 18:00. Memory [[lhb-publishes-after-1800]] closes the loop for the fifth time. This is a structural defect of the recap slot: it can never obtain the same day's LHB, while the pre-market slot always can. This report leaves it honestly blank and does not pass off "main-force net inflow" as the LHB.
• ⚠️ Full-A pagination only retrieved 3,900/5,223 (the pagination ceiling in memory [[em-clist-pagination-one-sided-bias]]). But this time the sort key was f12 (stock code) rather than percent change, so the truncation happens in code order and is orthogonal to returns; it does not constitute the one-sided bias of the previous report. The median of +0.44% is computed on that sample; the authoritative breadth figures always use the legu official caliber, not this sample.

3. Four cross-checks actively performed on method that changed the conclusion

1. ⚠️⚠️ [Most important item in this report] "Thermal coal +5.87%" is a dead contract. The first draft had written it into the evidence chain for a broad commodity rally. Checking revealed that ak.futures_zh_realtime('动力煤') returns ZC2212, tradedate=2022-11-28, position=0, volume=2 — a contract that stopped trading back in November 2022, with price fields filled in so they look like normal values. This is exactly isomorphic to memory [[cboe-stale-symbol-silent-trap]]: freshness must be pushed down to contract granularity; you cannot trust the product name. The whole line was deleted and replaced with East Money's "thermal coal" industry sector +2.01% and coke J2701 +1.81% (position 67,240, a live contract).
2. ⚠️⚠️ [The search engine stitched in a six-month-old story] Searching "phosphorus chemicals / glyphosate / limit-up / catalyst" returned as the top result "the US designates phosphorus and glyphosate as national security priorities", which looks like a perfect explanation for today's phosphate fertilizer & phosphorus chemicals +7.01%. But that item is the executive order Trump signed on 2026-02-18 invoking the Defense Production Act, published six months ago (the Pengpai/Sina originals are dated 2026-02-24/25). Had it been copied, the attribution for today's strongest branch would be entirely fabricated. Memories [[ai-attribution-stitches-old-news]] and [[recurring-column-headlines-hide-date]] recurred simultaneously. This report instead honestly labels it "attribution not closed", see §3.
3. ⚠️⚠️ [Testing the agriculture branch against futures — the conclusion contradicts the share prices] Today agrochemicals +4.62%, sugar-substitute concept +5.50%, phosphate fertilizer & phosphorus chemicals +7.01%; the narrative is "agricultural upcycle". But pulling up the corresponding commodities: urea −0.22%, rapeseed meal 0.00%, cotton +0.18%, soybean meal +0.32%, corn +0.44%, white sugar +0.95% (SR2701); on CNBC international, sugar UNCH, corn +0.56%, soybeans +0.15%. Only wheat +3.17% actually rose. That is: fertilizer stocks surged while urea futures fell, sugar-substitute stocks surged while international sugar was flat. This is the core evidence behind this report's judgment that "today's agriculture move is capital rotation, not fundamentals-driven"; the method follows memory [[etf-proxy-detects-stale-spot-quotes]] (using a tradable proxy to falsify a textual narrative).
4. ⚠️ [The seal-quality criterion from the pre-market piece: the full sample contradicts the very sample it named] Pre-market §8 specifically named "the four highest seal-to-turnover ratios vs the two lowest" as a test of the criterion. Measured: the high-quality group +3.21%, the low-quality group +4.25% — the named sample gives the reverse result. But pulling all 93 names, corr(seal ratio, today's change) = +0.221, weakly positive; the criterion was not falsified. Both conclusions must be stated: the criterion is weakly valid, but the 6-name test it designed for itself is too small to support any conclusion. Meanwhile corr(failed-limit count, today's change) = −0.097 ≈ 0, while inside the PCB sub-sample failed-limit count looks perfectly monotonic (Fastprint 0 breaks +9.17% > CIG 1 break +7.33% > Shennan 3 breaks +4.80% > Kinwong 6 breaks +0.94%) — that is a sub-sample coincidence and must not be extrapolated.

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

• Today's LHB was not obtained (publishes at 18:00), so the hardest pre-market evidence — "are institutional seats still buying PCB" — cannot be verified today; it must wait for tomorrow morning's pre-market piece.
• The first-hand catalyst for today's surge in agriculture / phosphorus chemicals was not identified. Ruled out: ① the US phosphorus/glyphosate executive order (2026-02 old news); ② commodity-price drive (futures barely moved, see above). What remains is capital rotation, expectations for the 9/11 USDA WASDE, or some policy item I did not scan; this report treats it as "attribution unknown" and invents no reason. The "which industry does the highest consecutive-limit-up stock belong to" question required by memory [[ztpool-industry-mix-only-counts-target-branch]] has been answered as a mandatory item (§4).
• The catalyst for general retail remains unidentified for a second straight day (Baida Group is already at 4 consecutive limit-ups). Carried over from the pre-market piece; still not closed here.
• The driver behind crude +4.66% (SC2610) / WTI +3.00% / Brent +1.76% is unconfirmed. The OPEC+ JMMC was scheduled to meet 9/6, but this report did not obtain the original meeting outcome, so "OPEC+" is not written as the attribution.

---

0. One-Line Recap

Sentiment temperature: warming, on the strong side, but the leadership changed hands. Today was neither a broad decline nor a broad rally, it was a high-to-low rotation: at the index level the ChiNext Index −1.15% and STAR 50 −1.52% led the losses, while market-wide 3,257 advancers / 1,859 decliners (official caliber), breadth even better than yesterday (3,039/1,987). What fell was heavyweight tech; what rose was agriculture, chemicals, resources, real estate, and pharma.

The strongest leadership is the agriculture + chemicals complex: phosphate fertilizer & phosphorus chemicals +7.01% (9 up, 0 down), agrochemicals +4.62% (61 up, 2 down), sugar-substitute concept +5.50%, glyphosate +5.28%, crop farming +3.72%; of today's 73 limit-ups, agrochemicals 8 + agricultural product processing 4 + crop farming 3 = 15, i.e. 20.5% of all limit-ups, the only direction that formed a cluster.

Pre-market watchlist scorecard: stock picking won, track selection lost — and it lost on the exact defect the list itself had already spelled out. The pre-market Top5 averaged +1.21% on a close basis, beating the full-A median of +0.44%; but on the executable "open → close" caliber it was −0.22%, with 3 of the 5 closing below their open. More seriously: the Pass group averaged +1.06%, essentially level with the recommended group's +1.21% — this list failed to separate strong from weak. Pre-market §9① itself wrote "this list has only one variable; five names are not five independent judgments"; today was precisely the worst day for that variable.

⚠️ But the pre-market verification-point framework judged correctly: §8 stated "a gap-up of <1% or a gap-down = none of the triple positives were accepted by the market, the strongest bearish signal" — Zhongji Innolight opened +0.08% and YOFC opened +0.09%; that signal triggered by 09:25, after which the optical module line spiked and faded all day. The conclusion was wrong, and the gallows built for that conclusion judged right (details in §2.3).

Tomorrow's tone: defensive counterattack. Yesterday's limit-up stocks averaged +2.89% today (60 up, 33 down, 19 limiting up again), showing sentiment has not broken, the money just moved; but the failed-limit rate rose from 17.0% to 33.6% and the top consecutive-limit count fell from 6 to 4, so continuation is clearly harder. Tomorrow's central tension: the agriculture-chemicals line has price but no fundamental confirmation (futures did not move), while the compute line has fallen to the "downgrade" level set by the pre-market criteria yet the money is still there (Innolight main-force net inflow +1.349 billion, second largest in the market).

---

1. Market Overview

1.1 Indices and Volume

1.2 Breadth and Sentiment Indicators

⚠️ The one thing this table most needs to be read for: the indices fell while the market rose. The ChiNext Index −1.15% and STAR 50 −1.52%, yet 3,257 stocks advanced, 73 limit-ups, 0 limit-downs, median +0.44%. This is the classic "heavyweights killed, single stocks up" high-to-low rotation, not a risk event. Labeling today a "correction day" would miss the structure; labeling it a "broad rally" is equally wrong — yesterday's strongest tech growth names were the weakest bucket in the market today.

⚠️ Second thing: the failed-limit rate doubling (17.0% → 33.6%) and the consecutive-board ladder thickening (2nd boards up from 9 to 12) happened at the same time. These two signals point in opposite directions: the former says absorption is deteriorating, the latter says the wealth effect is continuing. The reasonable reading is that money is withdrawing from the old leadership and rebuilding a ladder inside the new one, a process that necessarily produces many "old-line failed limits". The grouped data supports this reading: yesterday's 93 limit-up stocks averaged +2.89% today, with the agriculture complex (crop farming +5.62%, general retail +4.49%, publishing +9.68%) far ahead of the tech complex (components +2.74%, communication equipment +2.10%, auto parts −0.91%).

---

2. Pre-Market Watchlist Reconciliation

2.1 Top5 Recommendations (core reconciliation)

⚠️ This table must be read on two calibers. The pre-market list is published before the open, so the price a reader could actually get is today's open, not yesterday's close. Looking only at "close change %" would record "gapped up then bled all day" as a hit.

Hit rate: 2/5 = 40% (counting "open → close positive"); on a close-change basis it is 4/5 closing green, but 3 of those closed below their own open, so executing the list as written loses money.

2.2 Pass List Reconciliation — the most damning table in this report

⚠️ Three slaps that must be acknowledged:

1. 亚盛集团 (Gansu Yasheng, 600108, SH main board) rose +10.00% today for a 4th consecutive board, one of the highest board counts in the market. The pre-market Pass rationale was "no first-hand catalyst inside the window + turnover of only 102 million, extremely poor liquidity". Turnover was indeed still only 89 million (the liquidity judgment was right), but the inference "no catalyst therefore it will not rise" was wrong. Memory [[no-news-found-is-not-no-news]] recurred once again.
2. 科森科技 (Kesen Technology, 603626, SH main board) hit the limit today at +9.99%. The pre-market piece hammered the Pass with four pieces of evidence (the company itself admitted the share is tiny, ex-non-recurring −168 million, gross margin 3.15%, the same script drew a regulatory inquiry in 2024). All four pieces of evidence are true, and the stock limited up. This is exactly what memory [[financial-quality-is-not-a-short-term-ranker]] says: fundamental quality is a position-size manager, not a short-term ranker.
3. 崇达技术 (Suntak Technology, 002815, SZ main board) +6.68%, and on expanding volume. The pre-market piece explicitly wrote "⚠️ its PB of 2.22 is the lowest in the PCB chain; the Pass rationale is trading structure, not valuation — if it seals the limit again on renewed volume, it should be re-evaluated". Today turnover expanded 157% with a 25.55% turnover rate, triggering that re-evaluation condition, and the list contained no mechanism to execute it.

✅ There were also elegant calls: 天孚通信 (Suzhou TFC Optical, 300394) −3.18% was the only name downgraded from "watch only" to Pass in the pre-market piece, on the rationale "the most expensive valuation + the worst volume + the weakest capacity evidence all stacked on one name" — on a day when most of its sector closed green it fell 3.18%, so that downgrade was effective. 中巨芯 (688549) was also directionally right, but the magnitude was badly overestimated: the pre-market piece predicted "limit-down at the open" on the Moore Threads script, whereas it actually opened +0.22%, bottomed at −2.20%, and closed −1.71%. The 150.7% lockup-expiry ratio did not translate into price, showing that "lockup share" alone is not enough of a variable to predict the drawdown.

2.3 ⚠️ The Verification-Point Framework: the conclusions were wrong, the gallows judged right

Pre-market §8 set out a batch of falsifiable verification points. Judged one by one, 5 of 7 were right, 1 was reversed and 1 could not be judged — the verification points are far more accurate than the recommendation conclusions themselves. This is isomorphic to memory [[verification-point-beats-headline-conclusion]].

⚠️ On point 7 the statement must be completed, because it negates two things at once:

• The 6 names specifically called out pre-market give a reverse result (high quality +3.21% < low quality +4.25%). Using 4 names against 2 to "test" a criterion is an invalid design in itself.
• Across the full sample of 93 names, seal ratio vs today's change gives corr = +0.221 (the ≥10% group +3.70%, the 2–5% group +0.26%), so the criterion is weakly valid and was not falsified, though it is not monotonic (the <2% group at +1.63% is actually above the 2–5% group).
• Whereas the "failed-limit count" criterion gives corr = −0.097 on the full sample, essentially zero. What is dangerous is that inside the PCB sub-sample it looks perfectly monotonic — 兴森科技 (Fastprint, 002436) 0 failed limits +9.17% > 剑桥科技 (CIG, 603083) 1 time +7.33% > 深南电路 (Shennan Circuits, 002916) 3 times +4.80% > 景旺电子 (Kinwong, 603228) 6 times +0.94%. Looking only at this sub-sample one would conclude "failed-limit count is an effective ranker", while the full sample says it is noise.

2.4 One-Line Self-Criticism

What was right was the valuation/position half of "what not to buy" (−2.28%); what was wrong was the entire track selection. Three specific, actionable lessons:

1. ⚠️ The list has only one variable, and the pre-market piece spelled this out itself yet did not change its actions accordingly. Pre-market §9①, verbatim: "this list has only one variable... forcing in a stock with no catalyst to fake diversification is more dangerous than admitting concentration." That judgment was right, but what it should have implied is "cut total exposure to this line or add a genuinely independent leg", not "hand out 5 recommendation slots as usual". Memories [[one-variable-watchlist-fakes-independence]] and [[short-duration-branch-must-cap-slots]] say the same thing: identifying a risk ≠ constraining it.
2. ⚠️ Agriculture was downgraded wholesale as "already-fermented old news", when it had already been the market's leadership for 6 straight trading days. The pre-market piece listed agriculture as branch 8 (B+) on the rationale "the catalyst is out of window (late August) and already fermented", and put the 4-board 亚盛集团 into Pass. But "already fermented" is not "already over" — today the agriculture complex delivered 15 of 73 limit-ups and 1 of the market's 2 four-board names. This is a joint recurrence of memories [[withdrawal-confirmed-is-not-permanent]] and [[scan-window-misses-multiday-themes]]: the scan window is only 24 hours and cannot see a line that has already been running a week; what is needed is "pricing progress" as a second dimension, not the single dimension of "is the catalyst inside the window".
3. ⚠️ The limit-up pool industry-distribution table was obtained pre-market and fully laid out, and still did not save the call. Pre-market "Appendix 1" already listed crop farming 4 names and general retail 6 names, and honestly flagged "general retail catalyst unidentified". Today general retail produced the 4-board 百大集团 (Baida Group, 600865) and crop farming produced the 4-board 亚盛集团. The action memory [[ztpool-industry-mix-only-counts-target-branch]] requires is "look at the distribution before setting branches"; the pre-market piece managed the "look" but not the "set branches accordingly" — both of the highest consecutive boards grew inside branches graded B+/C.

---

3. Verifying Today's Leadership

3.1 Did the pre-market piece identify the strongest leadership? — No, and not by a wide margin

The pre-market piece gave S to optical modules and A+ to PCB; today those two came in at +0.05% and +0.20% respectively (sector caliber), both with main-force net outflow. Meanwhile agriculture and general retail, graded B+ and C, delivered the market's highest consecutive boards and its largest limit-up cluster.

⚠️ But one point must be made in the pre-market piece's defense, otherwise the reconciliation is unfair: today was not "optical modules collapsed", it was "optical modules stopped rising". Communication equipment +0.05% and components +0.20% are both small gains, not declines; what actually fell was semiconductors (−1.28%), consumer electronics (−1.08%) and electronic chemicals (−1.46%), and pre-market those three were given only a B+ and one explicit "must avoid". So the accurate statement is: the pre-market piece picked the line that went flat on the day, not the line that was worst on the day.

3.2 ⚠️ The attribution for today's strongest branch is not closed — no reasons invented

Agriculture + chemicals performed extremely strongly today (phosphate fertilizer & phosphorus chemicals +7.01%, 9 up, 0 down), but this report could not identify today's first-hand catalyst, and actively ruled out two seemingly valid explanations:

1. It is not "the US designating phosphorus and glyphosate as national security priorities". That item is the executive order Trump signed on 2026-02-18 invoking the Defense Production Act, more than six months ago. The search engine ranks it first for "phosphorus chemicals limit-up catalyst"; copying it would have manufactured a completely false same-day attribution.
2. It is not commodity-price driven — the futures give an almost opposite reading:

Conclusion: today's agriculture move has only one price leg, wheat; all the other products' futures were flat. This report therefore characterizes it as capital-rotation-driven with insufficient fundamental confirmation, not "the agricultural upcycle delivering". That characterization directly determines tomorrow's action (§6): you may follow it, but you cannot size it as a "fundamental inflection".

⚠️ Additional caveat: although wheat at 757.25 cents is +3.17% today, it is still below the 8/28 high of 784 cents cited in the pre-market piece. "Wheat at a new high" is wrong; today was only a bounce.

3.3 ⚠️ There is one genuine macro line that the pre-market piece got right and nobody acted on: crude oil

Crude rose sharply today and was cross-confirmed by two independent sources: Shanghai crude SC2610 +4.66% (720.6, position 38,743, volume 136,306, a live contract), WTI +3.00% ($94.22), Brent +1.76% ($98.71). The corresponding A-share sectors: oil & gas extraction +3.72% (5 up, 0 down), oil & petrochemicals +3.55%, petroleum processing & trade +2.92%, refining & trade +2.98%, with 中曼石油 (Zhongman Petroleum, 603619, SH main), 和顺石油 (Heshun Petroleum, 603353, SH main) and 华锦股份 (Huajin Chemical, 000059, SZ main) limiting up.

The pre-market list contained no energy leg at all — which is precisely the genuinely independent leg that should have been added when §9① admitted "there is only one variable" (energy and compute drivers are entirely uncorrelated). ⚠️ But this report could not confirm the first-hand driver of the oil rally (the OPEC+ JMMC was scheduled to meet 9/6, and the original meeting outcome was not obtained), so no attribution is written, only the facts recorded.

3.4 ⚠️ Foldables: the pre-market evidence grading was completely inverted by the market

Pre-market §6.1 did a solid piece of first-hand verification, grading five foldable names into four tiers, concluding that 东睦股份 (NBTM New Materials, 600114) had the strongest evidence (the only quantified disclosure, +219.3%) and that 科森科技 (603626) — which the company itself described as "a tiny share with no material impact" — be moved to Pass.

Today's result: 科森科技 +9.99% limit-up (the only foldable limit-up in the market), 东睦股份 −3.84%, 精研科技 −5.03%.

The evidence grading was right, the up/down ranking was backwards. This is the second recurrence of memory [[financial-quality-is-not-a-short-term-ranker]] on the same day (the first was the Pass group in §2.2). The correct use is not to abandon fundamental verification but to state its scope explicitly: first-hand evidence determines "how long you can hold", not "whether it goes up today". The pre-market piece merged those two into a single ranking.

---

4. Limit-Up Ladder / High Boards

4.1 High Boards and Promotions

12 names at 2 boards (up from 9 yesterday): 华脉科技 (Huamai Technology, 603042, SH main, communication equipment), 读者传媒 (Reader Publishing, 603999, SH main), 桂林旅游 (Guilin Tourism, 000978, SZ main), 中百集团 (Wuhan Zhongbai, 000759, SZ main), 中京电子 (Zhongjing Electronics, 002579, SZ main, components), 百合花 (Lily Group, 603823, SH main), 华体科技 (Huati Technology, 603679, SH main), 金正大 (Kingenta, 002470, SZ main, agrochemicals), 华盛昌 (CEM, 002980, SZ main), 上海电影 (Shanghai Film, 601595, SH main), 集泰股份 (Jitai, 002909, SZ main), 澳洋健康 (Aoyang Health, 002172, SZ main).

4.2 ⚠️ Which industry holds the market's highest consecutive board — the mandatory item of this report

Memory lesson: the limit-up pool's industry distribution must not be used only to count the branch you already care about. This section is required to answer "which industry does the highest consecutive-board stock belong to".

Today's three 4-board names belong to: small home appliances (ASD), general retail (Baida Group), and crop farming (Gansu Yasheng). Not one of the three is in the pre-market piece's branches 1–7.

• Crop farming: pre-market branch 8 (B+), and Gansu Yasheng was explicitly written into the Pass list.
• General retail: pre-market branch 9 (grade C, "catalyst unidentified"). Still unidentified today, now two days running.
• Small home appliances: not mentioned at all pre-market.

Taken together the three say one thing: the pre-market branch framework covered the market's strength but not the market's height. High boards are the anchor of sentiment, and today all of them grew where the pre-market grade was lowest or absent.

4.3 ⚠️ The liquidity trap in the high boards (the key constraint on tomorrow's execution)

The two 4-board names locked at the limit (Baida Group 59 million, Gansu Yasheng 89 million) have combined turnover of under 150 million. Their seal ratios are 3.995 and 6.244 — which looks like "extremely strong seal quality", but that strength is a product of liquidity drying up, not proof of demand: the seal orders sit there with almost no trading, and any decent-sized sell can break it open.

So tomorrow "4 boards locked at the limit + a high seal ratio" must not be read as "chaseable". The control case is 爱仕达 (002403): also 4 boards, seal ratio of only 0.141, but turnover of 592 million — it is the only one of the three that is actually tradable.

4.4 Broken Boards and Failed Limits

• Yesterday's highest consecutive board, 龙版传媒 (605577, SH main board, 6 boards), rose +9.12% today without limiting up — the streak broke. ⚠️ Two things must be said at once: listing it as "do not chase" pre-market was right (it indeed failed to extend), but it still rose 9.12% today, so executing "avoid" would have missed a leg. A broken streak ≠ a decline.
• 37 failed limits today, a failed-limit rate of 33.6% (17.0% yesterday). The three most brutal: 华盛昌 (002980, SZ main) broke 27 times before finally sealing a 2nd board, 集泰股份 (002909, SZ main) broke 9 times, 盈新发展 (Yingxin Development, 000620, SZ main) broke 7 times. Breaking 27 times and still closing at the limit is the best footnote to today's back-and-forth sentiment.
• Yesterday's 93 limit-up stocks today: average +2.89%, median +2.24%, 60 up / 33 down, 19 limiting up again, average gap-up +3.14%. ⚠️ Note "average gap-up +3.14% against an average close of +2.89%" — the group as a whole gapped up then eased slightly, the same shape as the Top5.

---

5. Capital Flows

5.1 ⚠️ First, three things that could not be obtained

1. Today's LHB was not obtained. The LHB publishes after 18:00 and this report was generated at 15:34. The hardest evidence in the pre-market piece was "the top 7 LHB net buys are all inside the compute-hardware complex, most with 3–4 institutional seats", and that cannot be verified today — whether institutions kept buying PCB or turned to agriculture-chemicals has to wait for tomorrow morning's pre-market piece. This is a structural information gap of the recap slot, not a fault of this run.
2. No northbound flow data (daily net buy disclosure was discontinued as of 2024-08-19).
3. "Main-force net inflow" below is East Money's large-order caliber, which is not the same ruler as Tonghuashun's "net inflow (inflow − outflow)"; the two cannot be subtracted.

5.2 Sector Main-Force Net Inflow (East Money caliber)

⚠️ Key reading: the components sector was up a marginal +0.20% while main force took out 1.304 billion. Price did not fall but money is leaving — the classic distribution pattern, more worrying than an outright decline. Communication equipment is the same (+0.05% / −2.388 billion).

5.3 Top Single-Stock Main-Force Net Inflows / Outflows

⚠️ Two contradictory signals that need to be called out separately:

1. 中际旭创 (300308) was up only +0.44% all day, yet took in 1.349 billion of main-force net inflow, second in the market. That is the opposite of the apparent "optical modules ebbing". The reasonable reading: large orders are absorbing the selling pressure, money has not left the leader; what is leaving is the second tier (Accelink −467 million, TFC −3.18%). Whether this line is "the leader consolidating on shrinking volume" or "the second tier dragging the leader down" is tomorrow's key point of disagreement.
2. 摩尔线程 (688795) fell 3.70% yet was the market's largest main-force net inflow (1.404 billion). The stock limited down on 9/7 on lockup expiry, and two consecutive days of large net inflow show money is actively absorbing the lockup selling pressure. Listing it pre-market as a "reverse indicator for compute sentiment" was reasonable, but the judgment that "the selling pressure is not yet digested" needs to be revised in light of the flows.

5.4 Inside Compute Hardware: the ranking is almost completely inverted

The pre-market piece switched the primary ranking variable from "valuation level" to "relative strength (distance from the 60-day high) + institutional money", and today that change contributed negatively: Kinwong, with the shallowest drawdown (−0.0%), ranked last, while the chain's most expensive name Fastprint (PE 288) and the structurally worst name Suntak (seal ratio 0.15%) took the top two spots.

⚠️ But do not draw the reverse conclusion that "relative strength is a wrong criterion". Today was a mean-reversion day (high-flying tech sold off, laggards caught up), and on such days any "chase strength" variable underperforms; on continuation days it usually works. The correct lesson is: the ranking variable must be paired with the prior judgment of "is today a continuation day or a rotation day", and the pre-market list never made that prior judgment — it simply treated the variable as a universal truth.

---

6. Tomorrow's Outlook

6.1 Leadership Continuation Assessment

6.2 Tomorrow's Focus (branch + stocks + verification points)

⚠️ This section deliberately adds two legs uncorrelated with compute (energy, agriculture), a direct correction of today's biggest lesson — the "only one variable" defect of the pre-market list has now been called out two days running and can no longer be identified without being corrected.

6.3 Avoid Tomorrow

1. Two of the three 4-board names locked at the limit — 百大集团 (Baida Group, 600865, SH main) and 亚盛集团 (Gansu Yasheng, 600108, SH main). Turnover of only 59 million and 89 million respectively; the "high seal ratio" is a product of liquidity exhaustion, not proof of demand; and general retail, Baida Group's sector, has had an unidentified catalyst for two days running.
2. The whole semiconductor and electronic chemicals complex. Semiconductors 32 up / 152 down with main-force net outflow of 1.389 billion; electronic chemicals −1.46%, the weakest in the market, and 中巨芯 (688549, STAR) is still releasing lockup selling pressure.
3. Yesterday's 7 auto-parts limit-ups (the robotics line). They averaged −0.91% today, the worst group, and the catalyst (Nvidia acquiring Hugging Face) is already 9/3 old news; a one-day wonder confirmed.
4. Chasing the foldables chain. Apple's autumn event is at 01:00 Beijing time on 9/10 (i.e. the outcome is known before the open the day after tomorrow); today 科森科技 (603626) already limited up +9.99% while 东睦股份 (600114), which has the strongest first-hand evidence, fell −3.84% — pure sentiment, and directionally opposite to the evidence; tomorrow remains a pure expectations game, and the 9/10 open faces "sell the news" immediately.
5. 2nd-board names that barely sealed after breaking more than 5 times — 华盛昌 (CEM, 002980, SZ main, 27 breaks) and 集泰股份 (Jitai, 002909, SZ main, 9 breaks). Today's seals came from repeated contesting, not from absorption.

6.4 Input Notes for Tomorrow Morning's Pre-Market List

1. ⚠️ You must first answer "is tomorrow a continuation day or a rotation day", then choose the ranking variable. Today proves it: the same "relative strength + institutional money" ranking sorts systematically backwards on a rotation day (§5.4). Suggestion: add a fixed prior judgment pre-market — yesterday's limit-up stocks' average today vs the full-A median, plus whether the industry holding the highest consecutive board matches your own first branch.
2. ⚠️ The LHB is the pre-market slot's unique informational advantage, and tomorrow morning it must be used to answer the question today could not: are institutional seats still buying PCB / optical modules, or have they turned to agriculture-chemicals? That is the key evidence for the first two rows of §6.1, missing here because of the 18:00 publication time.
3. ⚠️ The verification point for the agriculture line is already set; execute it directly tomorrow morning: do urea (UR) and white sugar (SR) futures follow. Stop grading agriculture by "is the catalyst inside the scan window" — it has been running 6 days and the window cannot see it. Switch to the "pricing progress" dimension: which day it is, high-board turnover, and whether futures confirm.
4. ⚠️ The catalysts for general retail (Baida Group at 4 boards) and publishing (China Publishing & Media at 3 boards, publishing sector 27 up 0 down) have been unidentified for two days running; tomorrow morning must either close the loop or explicitly declare it abandoned. They currently account for one of the market's highest consecutive boards and the strongest group return (+9.68%), so the cost of the attribution blank is rising.
5. ⚠️ Add an independent leg unrelated to compute. Today's energy line (WTI +3.00%, SC +4.66%) is the only direction with independent commodity-price confirmation, and the pre-market list did not cover it at all. "Admitting concentration" has now been written two days in a row; tomorrow it should become "cap the number of recommendation slots on any single variable at 3".
6. Sample-size requirement for testing criteria: today proves that using 4 names against 2 to test the "seal quality" criterion produced the opposite conclusion from the full sample (n=93, corr=+0.221). Whenever a criterion is to be tested, the full-sample correlation must be run; naming a handful of stocks is not enough.

---

⚠️ 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 rely on exchange/company disclosures, and do not use this directly as a basis for trading.