A-Share Market Recap | 2026-08-10 Monday

Data-basis disclosure first (please read this section before anything else)

1. Market-data basis: index levels and turnover are taken from the Sina Finance closing snapshot; the SSE/SZSE percentage moves were independently cross-checked and found consistent via East Money's HK–Mainland Stock Connect summary API (SSE +0.67%, SZSE +0.04%). Individual-stock percentage moves, turnover rate, net main-force inflow and turnover value are taken from the Tencent Finance closing snapshot (5,539 records market-wide). Limit-up / failed-limit-up / limit-down / previous-day consecutive-limit-up pools are taken from East Money. Sector percentage moves and sector net inflows are taken from the Tonghuashun sector definition (90 sectors).
2. The two fund-flow bases must not be mixed: Tonghuashun's "sector net inflow" and Tencent's "single-stock net main-force inflow" belong to two different statistical systems; they cannot be added together and cannot substitute for each other.
3. Price limits are tiered by board, and every stock in this report is tagged with its board: SSE/SZSE main board ±10%, ChiNext (30/301) ±20%, STAR Market (688) ±20%, BSE ±30%.
4. Two data items not obtained / left blank in this report:
   • Today's Dragon-Tiger List was not obtained; the Dragon-Tiger List part of Section 5 is left blank. The exchanges only disclose the Dragon-Tiger List at about 18:00 after the close, and this report was generated at 15:30 — a scheduling problem, not a data gap. Today's "fund flows" section therefore has only two bases, net main-force inflow and sector net inflow, and lacks the exchange-level seat verdict — the rule laid down in last Friday's recap, "when institutional seats and net main-force inflow point in opposite directions, defer to the institutional seats", cannot be executed today. We suggest discounting this report's fund-flow conclusions accordingly.
   • Northbound net inflow cannot be given, and this is not a data-fetch failure. Effective 2024-08-19 the Shanghai and Shenzhen exchanges changed the Stock Connect disclosure mechanism: daily northbound net buying is no longer disclosed and has been replaced by quarterly publication of shareholding quantities (Sina Finance¹, The Beijing News²). Therefore any source claiming "northbound net inflow of RMB XX billion today" should be treated with suspicion. This report substitutes southbound flows and sector net inflows.
5. Basis for relative-strength benchmarking: sector percentage moves use the Tonghuashun sector definition, and a stock's sector is assigned by its principal business, which may differ from Tonghuashun's constituent lists; this is used only as a relative-strength reference, not as precise attribution. The sector assignment of Dongwei Technology (东威科技, 688700) could not be verified in this report, so its reconciliation is judged on absolute percentage move alone.
6. A/H premiums continue to use the pre-market fixed exchange rate of 0.8656; today's actual rate was not obtained. Therefore the absolute level of the premium carries a small error, but the direction ("narrowing/widening") is unaffected (both periods are compared at the same rate).
7. ⚠️ The Hong Kong and southbound data in this report are 15:30 intraday readings, not closing prices. Hong Kong trades until 16:00, while this report was generated at 15:30. The H-share percentage moves in Section 2.3 and the southbound flows in Section 5.1 are both taken from the intraday snapshot at that moment, and the last 30 minutes of change is not included. Directional conclusions (who rose, who fell, outperformed or underperformed the HSI) are robust at this magnitude, but please do not quote these figures as closing data.

⚠️ Data-fetch and implementation notes for this run (internal only):

• East Money's push2.eastmoney.com main API failed on a broad scale today, but the failure is at the endpoint level rather than the domain level — this is the third occurrence of the same failure pattern this month.
  • Failed on all 4 retries: stock_zh_index_spot_em (indices), stock_zh_a_spot_em (market-wide snapshot), stock_board_industry_name_em (sectors), stock_hk_spot_em (HK snapshot), all with ConnectionError / RemoteDisconnected; stock_board_concept_name_em returned HTTP 502.
  • Completely normal under the same domain: the stock_zt_pool_em family (limit-up / failed limit-up / limit-down / previous-day consecutive limit-up), stock_hsgt_fund_flow_summary_em, stock_hsgt_fund_min_em.
  • Every fallback route hit, and not a single figure in this report is missing as a result: indices → Sina stock_zh_index_spot_sina; market-wide snapshot → Tencent stock_zh_a_spot_tx; sectors → Tonghuashun stock_board_industry_summary_ths; HK shares → direct connection to the raw qt.gtimg.cn quote endpoint (bypassing the akshare wrapper).
  • → Process recommendation: design for "East Money main snapshot endpoints fail" as the normal case, not the exception. Indices, market-wide snapshot and sectors — the three data classes needed every single day — should be permanently configured with a primary + backup route and automatic degradation, instead of retrying 4 times a day and then hand-hunting for a substitute. The time spent on fallback routes today exceeded all the time spent on analysis.
• The Dragon-Tiger List again could not be obtained today because of scheduling; this is the fourth occurrence of the same root cause. stock_lhb_detail_em and stock_lhb_jgmmtj_em both returned TypeError: 'NoneType' object is not subscriptable (i.e. the endpoint returned an empty body), and stock_lhb_detail_daily_sina returned KeyError: '股票代码' — all three routes fail in a shape that looks like "the API is down", when in fact the data simply had not been published at 15:30. → This morning's pre-market recommendation should be upgraded to a hard requirement: the post-close recap should permanently re-run the Dragon-Tiger List fetch after 18:30 and write it back into that day's report, or at the very least fetch the "previous trading day's Dragon-Tiger List" in the pre-market flow. Otherwise every recap generated at 15:30 will permanently be missing the exchange-level fund verdict — which happens to be our only hard evidence for judging "hot-money ramp vs institutional accumulation". Today's verification of Accelink Technologies (光迅科技, 002281) depended directly on last Friday's Dragon-Tiger List.
• The northbound item was nearly written up wrong today, and deserves a note of its own. stock_hsgt_fund_flow_summary_em returned a northbound "net buy turnover" of 0.0 today, while southbound had normal values (+RMB 3.176 billion / +RMB 40 million). The first reaction was "the API is broken", but all 241 minute-level data points were 0 while southbound was normal — the shape does not look like a fault. A check confirmed it is the disclosure-mechanism reform of 2024-08-19: daily northbound net buying is no longer disclosed. → The lesson shares a root with [[listing-status-must-be-queried]]: this is a "status-type fact" — the rule changed two years ago, while my default workflow still contains the step "pull northbound net inflow". It does not throw an error, it just returns 0, and if you don't go check the rule, the most likely outcome is writing a completely fabricated conclusion like "northbound net inflow was RMB 0 billion today, a neutral stance". Process assertion: for any fund-flow data that returns "exactly 0", first check whether that data is still being disclosed, then check whether it really was 0 that day.
• The unit of Tencent's turnover field was explicitly validated rather than recalled from memory. Cross-validated with Shengyi Technology (生益科技, 600183): the limit-up pool (East Money, unit: yuan) gives RMB 13,037,738,752 = RMB 13.04 billion, and Tencent's turnover/1e4 = RMB 13.038 billion — the two sources agree, confirming the unit is ten-thousands of yuan. The market-wide total of RMB 2,538.7 billion cross-confirms with Sina's SSE RMB 1,166.9 billion + SZSE RMB 1,356.2 billion = RMB 2,523.1 billion (the difference being the BSE). This is the same class of application as the remembered rule "run a ratio sanity check on multi-day fields first", applied today to units.
• The format of the 涨停统计 field was again confirmed today as x/y = y limit-ups within x days (this morning's internal pre-market note had the subject and object of this sentence reversed, though the reading at the time was correct). Cross-validation: Baihua Pharmaceutical (百花医药, 600721) shows 5/5 today with 连板数=5, and Baoding Technology (宝鼎科技, 002552) shows 7/6 today (6/5 last Friday). This report uses the 连板数 field directly throughout and no longer parses 涨停统计 itself.
• Today's reconciliation is negative for the pre-market scoring model, and the mechanism — not just the result — needs to be recorded faithfully. Pre-market downgraded ACM Research (Shanghai) (盛美上海, 688082) from A to B and put it on the "watch only" list, on the single ground that ex-non-recurring profit was −15.31% and net profit attributable to parent grew entirely on RMB 473 million of unrealised securities gains — that financial judgement was not overturned today (the interim-report figures have not changed). But it was +12.16% today, the single best performer among all 41 tracked names. The real reason lay outside pre-market's field of view: semiconductor equipment took 6 of the top 20 net main-force inflows today (AMEC (中微公司, 688012) at +RMB 890 million, first market-wide; Naura Technology (北方华创, 002371) +RMB 339 million; ACM Research (Shanghai) +RMB 305 million; Changchuan Technology (长川科技, 300604) +RMB 302 million; Jingce Electronic (精测电子, 300567) +RMB 302 million; Piotech (拓荆科技, 688072) +RMB 301 million), while over the same period the semiconductor sector as a whole saw a net outflow of RMB 14.351 billion, the largest outflow market-wide. In other words, money rotated within the sector from optical modules / PCB / memory into equipment, and pre-market had these three equipment names scattered across three different branches, each screened one by one on financial quality — so no step anywhere would ever have seen that "equipment is a set being bought as a whole". → Two concrete model defects: ① The single-stock financial-quality screen (ex-non-recurring / attributable profit, OCF / attributable profit) answers "is this worth holding long term", not "who goes up tomorrow". Using it as the main weight for short-term ranking will systematically screen out the direction being bought on rotation days. This is not an argument for abandoning the screen, but it should not on its own decide the trade-level label "watch only".
• That A/H table was nearly published as a close-to-close comparison. At 15:30 I pulled from qt.gtimg.cn WuXi AppTec (药明康德, 603259) H shares +3.02%, Pharmaron (康龙化成) −0.58%, JOINN Laboratories (昭衍新药) −2.33% — a conclusion so clean it was tempting to write it straight up as "today's H-share close" — but Hong Kong does not close until 16:00, so these are intraday readings. What caught it was not remembering the trading hours, but seeing the API tag "trading status = trading" while checking the southbound data, which sent me back to realise the H-share snapshot was likewise not a close. → This is the mirror image of the same class of fault as [[stale-source-silent-lag]]: that time the data was a day behind reality, this time the data was half an hour ahead of the close. What they share is that neither throws an error, both are internally consistent, and both can be written straight into a conclusion. This report flags it twice, in disclosure item 7 and in the Section 2.3 table header. Process assertion: in a recap generated at 15:30 A-share time, any cross-market data (Hong Kong, overseas, FX) must be tagged with its reading time, because the A-share close is not their close.
• Southbound flows were not used today, and the reason is worth recording. Stock Connect (Shanghai) southbound +RMB 3.176 billion vs Stock Connect (Shenzhen) southbound +RMB 40 million — a gap of nearly 80×. Either figure alone looks entirely unremarkable; it is the ratio when they are placed side by side that exposes the problem — exactly the same discovery method as this morning's pre-market zllr_d5 finding (both are ratio checks, not recall of field semantics). The same endpoint's "net fund inflow 420.0", identical across both channels, is the daily quota ceiling, not an inflow amount; quoting it directly would have produced the utterly fabricated figure "RMB 42 billion flowed in northbound and southbound each". ② Pre-market lacks a "fund-set perspective" step: first look at which set the money is going into, then rank within that set by financial quality — rather than slicing branches first and scoring stock by stock. Today's mistake was the latter, and the errors line up very neatly — all three names judged wrongly (ACM Research (Shanghai) +12.16%, Longsys +6.70%, Kinwong Electronic +1.11%, outperforming its sector by 2.17pct) sit inside "the set the money is flowing into".
• There is one falsifiable point that pre-market wrote down itself and that must be acknowledged as falsified today. Pre-market Section 8.1, verbatim: Zijin Mining (紫金矿业, 601899) "if it opens up by less than 1%, it means money does not accept this expectation gap, and the strength of this branch needs re-rating". Zijin was in fact +0.85% for the whole day, and the closing gain never even reached 1%, while the precious-metals sector was +4.54% over the same period (2nd market-wide). The branch was called right, the vehicle was called wrong, and pre-market had already written down the criterion that would identify it. → This is both the most creditable point in this report and the one most in need of reflection: the criterion was written correctly, but it triggers intraday while the recap happens after the close — writing down a falsifiable point that nobody executes intraday is the same as not writing it. Process recommendation: turn the verification signals of pre-market 8.1/8.2 into a standalone, tickable intraday checklist rather than burying them in the body text.

---

0. One-Sentence Recap

Today was a day of "right branch, wrong vehicle" — sentiment clearly warmed while the structure was extremely torn. The SSE Composite was +0.67% at 3966.59, 4,068 stocks market-wide advanced (73.4%), 99 limit-ups (74 last Friday), the failed-limit-up rate fell to 12.4% (26.0% last Friday), and the highest streak rose from 4 to 5 consecutive limit-ups, 2 names — the sentiment temperature moved from "cautious" back to "warming". Yet on the very same day, the ChiNext Index was −0.73% and the STAR 50 −0.36%, with Components −1.06% (89th of 90 market-wide), Communication Equipment −0.84% (88th of 90) and semiconductors seeing a net outflow of RMB 14.351 billion (the largest market-wide).

Pre-market got two branches right and the third wrong — and that third one was precisely the branch pre-market called "the only one with earnings already delivered". The innovative-drug/CXO branch ranked 1st pre-market → Healthcare Services +4.59%, 1st market-wide; precious metals ranked 2nd → Precious Metals +4.54%, 2nd market-wide; AI computing power/PCB ranked 3rd → Components −1.06%, 2nd from the bottom market-wide. The direction was not off by a margin, it was inverted.

An even harsher number: 73.4% of all stocks market-wide rose today, while only 6 of the 12 pre-market recommendations rose (50%), and only 2 outperformed their sector. On a broad-advance day, the recommendation book's up-ratio was significantly below the market-wide baseline — that says more than the "50% hit rate" figure by itself.

All three falsifiable observation points left by pre-market received a clear verdict today — two proving pre-market right and one proving it wrong; see Sections 2 and 3.

Tomorrow's tone: the main lines are pharma and precious metals, but both already show "high disagreement" signals; the bleeding in AI hardware has not stopped; the genuinely clean new direction is semiconductor equipment.

---

1. Market Overview

1.1 Indices and Turnover

• Turnover shrank about 4.7% from last Friday's RMB 2,664.4 billion but still held above RMB 2.5 trillion, the 11th consecutive session above RMB 2 trillion. The demerit line pre-market had set was "shrinking below RMB 2.4 trillion" — that was not triggered today, so this is healthy contraction.
• Index divergence is today's single most important structural feature: the SSE up, ChiNext and STAR 50 down. The main-board median gain was +1.52%, while the STAR Market median was only +0.27% and the BSE +0.34%. This was a day of broad main-board advance and growth being sold off.

1.2 Sentiment Indicators

• Limit-ups by board: 94 on the main board, 4 on ChiNext, 1 on the STAR Market. Limit-ups were almost entirely concentrated on the main board, fully consistent with "growth being sold off".
• Two signals contradict each other and must be read together: the failed-limit-up rate of 12.4% is very low (new names can hold the board), but the promotion rate was only 16.2%, and 33 of yesterday's 74 limit-up stocks fell today (44.6%). This is the classic structure of "good low-level rotation, poor high-level continuation" — money is willing to open new boards but unwilling to take over old ones.
• 5 limit-downs: Jinfa Labi (金发拉比, 002762, SZSE main board), Gansu Engineering Consulting (甘咨询, 000779, SZSE main board), Biolight (宝莱特, 300246, ChiNext), Ningxia Building Materials (宁夏建材, 600449, SSE main board), Lianchuang Optoelectronics (联创光电, 600363, SSE main board). An extremely low count, with no systemic-risk signal.

Sentiment temperature verdict: warming (neutral-to-warm). Limit-up count, failed-limit-up rate, limit-down count and advancing share all point to warming; but the promotion rate and index divergence point to a "structural, unsustainable" warming. This is not euphoria — on a euphoric day two major growth indices would not close lower.

---

2. Pre-Market List Reconciliation

Judging standard (using the condition pre-market set for itself, not changed after the fact): pre-market Section 6 states explicitly — "if indices broadly advance today and limit-ups exceed 70, please use 'did it underperform its sector' rather than 'did it fall' as the judging standard". With 99 limit-ups and 4,068 advancers today, the condition is triggered, so the table below uses excess return versus sector as the primary criterion, with absolute percentage move given alongside.

2.1 Recommended Names (3 priority deep-dives + 9 watch closely)

Recommendation-book hit rate

The one line most worth remembering: on a day when 73.4% of stocks rose, only 50% of the recommendation book rose. This is not "failing to outperform", this is significantly underperforming random stock selection.

2.2 Watch-Only / Pass List Reconciliation

Pass-list hit rate: 12 / 19 = 63.2% (9/19 = 47.4% last Friday, an improvement of 15.8pct).

2.3 All Three Falsifiable Points Left by Pre-Market Were Adjudicated Today

Pre-market wrote three "verifiable today" observation points in Section 8. They are the most valuable part of this report — because they were written down before the outcome happened.

① The A/H natural experiment — pre-market was right, and right with precision.

Pre-market verbatim: "If the HK CXO names rise clearly more than the A shares, the market is genuinely re-rating the geopolitical risk premium; if the A shares surge while the H shares are flat, it is an A-share sentiment trade." It also gave an L1/L2/L3 three-tier classification, stating that the ban legally reaches only WuXi AppTec.

Basis reminder: the H-share data in the table below are 15:30 intraday readings (Hong Kong closes at 16:00), not closing prices; the HSI on the same basis was +1.09%. The last 30 minutes of change is not included.

The conclusion is extremely clean: on a day when the HSI rose 1.09%, WuXi AppTec — the only company named in 1260H and the only one to win its case — saw its H share outperform the HSI, while the H shares of the two unnamed companies fell outright. Foreign investors cast a clear vote against the narrative that "unnamed companies should be re-rated too", while the A-share market gave those two almost the same gain as WuXi AppTec (+3.86%, +3.75% vs +4.21%).

This precisely validates pre-market's L1/L2/L3 tiering, and also validates pre-market's line that "any account that writes this up as an improvement in all three companies' order-taking capacity is treating an inference as a fact". What A shares bought today was sentiment diffusion; what Hong Kong bought was the legal fact.

② "Who is leading the rally" — pre-market was right.

Pre-market verbatim: "If the leaders are unnamed companies, this is sentiment diffusion rather than logical pricing, and the durability should be discounted."

Healthcare Services was +4.59% today, 1st market-wide, but the leaders were neither WuXi AppTec nor any CXO, but biological reagents / research services: Bide Pharmatech (毕得医药, 688073, STAR Market) +20.00%, ACROBiosystems (百普赛斯, 301080, ChiNext) +20.00%, Novoprotein (近岸蛋白, 688137, STAR Market) +17.51% (failed limit-up), Medicilon (美迪西, 688202, STAR Market) +8.41%, Hongbo Pharmatech (泓博医药, 301230, ChiNext) +8.27%. WuXi AppTec's +4.21% ranks toward the back within the sector.

The driver of this chain is better-than-expected global pharma R&D capex (multinational pharma companies' external capex year-to-date reached US$200.3 billion, 73.5% of the full-year 2025 figure), beginning with last Friday 8-07's collective limit-ups across the CRO sector (Sina Finance³, ifeng.com⁴) — a driver independent of the 1260H ban. Pre-market did mention "at the sector level this is a recovery in global healthcare investment and financing", but placed it in a secondary position, with the main narrative still riding on the judicial injunction. Seen today, the judicial injunction is WuXi AppTec's business alone; the sector's money came from a different line.

③ Precious metals, "leader vs back row" — pre-market was right, but hit in reverse, and the cost was the recommendation book.

Pre-market verbatim: "Note whether the leader heads the move or the back row rises first. If Zijin Mining does not move while the small caps hit the board first, this is a late-stage catch-up pattern." And: "If Zijin opens up by less than 1%, money does not accept this expectation gap."

The gain is perfectly negatively correlated with "is it the leader". This is exactly the late-stage catch-up pattern pre-market described. Pre-market ranked Zijin Mining 1st in the branch and 3rd overall, on the grounds that it "rose only 1.88% on Friday with a 1.33% turnover rate, having not digested the payrolls print at all" — that "not digested" judgement was itself correct (gold prices genuinely had not been priced in); what was wrong was assuming the leader would be the one to digest it. Today the gold-price expectation gap was digested by the back-row small caps.

2.4 One-Sentence Self-Criticism

What was called right was "which branch"; what was called wrong was "which name inside the branch", plus one entire branch whose direction was inverted.

• Types of correct calls: ① tiering built on legal/institutional facts (L1/L2/L3, perfectly validated by A/H); ② the Pass built on deferring to exchange seat data when fund-flow bases conflict (Accelink Technologies at −7.64% is the strongest validation); ③ the judgement built on liquidity drain from a scheduled event (the robotics chain at −0.39%, 86/90 market-wide). What these three share is that the criterion is a hard fact, not a valuation opinion.
• Types of wrong calls: ① translating "poor financial quality" directly into "it will underperform tomorrow" — ACM Research (Shanghai) (盛美上海, 688082) +12.16% and Longsys (江波龙, 301308) +6.70% both went wrong this way, and their financial judgements were not overturned today; ② picking the weakest vehicle inside a correctly chosen branch — Zijin Mining (紫金矿业, 601899) underperformed its sector by 3.69pct; ③ assigning too much weight to the criterion "earnings already delivered" — Cambricon (寒武纪, 688256) posted an interim report of +122.61% yet was −6.33% today, and Guanghe Technology (广合科技, 001389) posted ex-non-recurring +96.71% yet was −0.92% today.

Exactly how to change pre-market next time (three items, all executable):

1. Add a "fund set" preliminary step: before slicing branches, first look at which set the top 20 net main-force inflows are concentrated in, then rank within that set by financial quality. Today semiconductor equipment took 6 of the top 20 net main-force inflows, while pre-market had those names scattered across three branches and screened each out on ex-non-recurring profit — the screen was not wrong, the order was.
2. Decouple "financial quality" from "trading conclusion": metrics like ex-non-recurring profit and OCF / attributable profit should determine position size and holding period, and should not on their own decide the trading label "watch only". We recommend narrowing the Pass threshold to only the three types backed by hard facts: a company clarification announcement, net selling by institutional seats, and scheduled liquidity drain. — all three of those were right today.
3. Turn the verification signals of pre-market Section 8 into a standalone intraday checklist. Today's criterion for Zijin Mining was written correctly pre-market, but nothing anywhere executed it intraday. A falsifiable point that is written down but not executed is the same as one never written.

---

3. Validation of Today's Main Lines

3.1 Did Pre-Market Identify the Strongest Main Line?

It got the top two right, and in the right order. The innovative-drug/CXO branch ranked 1st pre-market corresponds to Healthcare Services, 1st market-wide today; the precious-metals branch ranked 2nd corresponds to Precious Metals, 2nd market-wide. Hitting the top two in order out of 90 sectors is not luck.

But the 3rd branch came in 2nd from the bottom. And pre-market's confidence was highest in that third branch — it was the only one with "earnings already delivered", on the grounds of the formal interim report released on the evening of 08-07. Today proves: already-disclosed earnings are not a reason to buy, they are something that has already happened. The Components sector was +6.45% with 11 limit-ups last Friday — the earnings were already priced then, and the weekend's interim report merely confirmed the pre-announcement.

3.2 Unexpected Main Lines Pre-Market Missed

① Semiconductor equipment — the cleanest line today, and the one that most should have been caught.

On the very same day the semiconductor sector as a whole saw a net outflow of RMB 14.351 billion (the largest outflow market-wide), the equipment segment saw net inflows across the board and took 6 of the top 20 net main-force inflows on its own. This is a directional switch inside the sector, not a broad advance.

Two things must be stated clearly: ① this did not start today — equipment names generally show 5-day gains of +23% to +41%, meaning it has already been running for a week, so what pre-market missed was not a starting signal but a main line already in motion; ② pre-market did not fail to see these names, it actively Passed one of them — ACM Research (Shanghai) was put on "watch only" because of ex-non-recurring −15.31%, and it was the best performer of all tracked names today. See Section 2.4 for the detailed mechanism review.

② Defence equipment — pre-market gave it a B, while it was in fact a top-6 sector by net inflow. Defence Equipment was +2.60% with a net inflow of +RMB 1.055 billion and 72 up / 10 down. Pre-market hung defence under the "oil / tanker shipping / defence" geopolitical branch and gave it a B, underestimating its independence.

③ The broad advance in low-priced cyclicals and consumer stocks — the other half of the market, unrelated to themes. Animal Husbandry +3.81% (36 up / 0 down), Beverage Manufacturing +3.46%, Agricultural Product Processing +3.11%, Food Processing & Manufacturing +2.89%, Baijiu +2.43% (net inflow +RMB 1.939 billion), Banks +0.57% (net inflow +RMB 2.410 billion, 2nd market-wide). Among today's 99 limit-ups, Power Grid Equipment had 6, Chemical Pharmaceuticals 5, Chemical Products 5, Power 4 and Auto Parts 4 — the sector distribution of limit-ups was extremely dispersed, with the top 5 sectors together accounting for only 24 names. This shows today's limit-up wave was not driven by any single main line, but by broad rotation among low-level main-board stocks. Pre-market's framework is "find the strongest branch", while the way the market paid out today was "the further from AI hardware the better" — a dimension pre-market did not cover at all.

---

4. Limit-Up Tiers / Highest Streaks

4.1 Tier Structure

• The highest streak rose from last Friday's 4 to 5, so height attrition eased somewhat.
• 2 names went from 4 to 5, 3 broke their streak: Baoding Technology (宝鼎科技, 002552) and Baihua Pharmaceutical (百花医药, 600721) were promoted; Huilv Ecology (汇绿生态, 001267) +0.66%, Yunnan Germanium (云南锗业, 002428) +2.50% and Woge Optoelectronics (沃格光电, 603773) +0.59% broke their streaks. Note that all three broken-streak names still closed higher — this is a "mild break" rather than a sell-off, a contrast with last Friday's 26% failed-limit-up rate.
• The 3-board tier is almost a gap: only Kaikai Industrial (开开实业, 600272) at 3 boards, with only one step between 2 boards and 5 boards. The tier structure is "barbell-shaped" (87 first boards + 2 five-boards, thin in the middle), which is structurally unhealthy: new sentiment is strong but relay support is lacking.

4.2 Promotions and Broken Streaks

• Of yesterday's 74 limit-up stocks, only 12 hit the limit again today, a promotion rate of 16.2%; 33 fell (44.6%), with an average move of +2.22% and a median of +1.05%.
• The harshest broken-streak sell-offs (limit-up yesterday → today): Ningxia Building Materials (宁夏建材, 600449, SSE main board) −10.00% limit-down, Jinfa Labi (金发拉比, 002762, SZSE main board) −9.99% limit-down, Wuhan Fingu Electronic (武汉凡谷, 002194, SZSE main board) −9.91% (2-day streak broken), Kaiweite (锴威特, 688693, STAR Market) −8.76% (2-day streak broken), Tongyu Communication (通宇通讯, 002792, SZSE main board) −8.31% (2-day streak broken), Accelink Technologies (光迅科技, 002281, SZSE main board) −7.64%.
• One clear directional signal: 3 of the 6 worst decliners are communication equipment / semiconductors (Wuhan Fingu, Tongyu Communication, Accelink Technologies). Yesterday's limit-ups were cut down today, and what was cut was precisely the AI hardware chain.

4.3 Core-Leader Identification (large turnover + rising + recognisable)

The name most deserving of the label "core leader" today is AMEC (中微公司, 688012): turnover of RMB 10.93 billion, up 6.08%, net main-force inflow 1st market-wide, and its set (semiconductor equipment) rose across the board. WuXi AppTec had larger turnover but a net main-force outflow — a pattern of retail investors and sentiment traders taking the other side.

---

5. Fund Flows

5.1 Northbound Flows

Cannot be given, and this is not a data-fetch failure. Effective 2024-08-19 the Shanghai and Shenzhen exchanges changed the Stock Connect disclosure mechanism, and daily northbound net buying is no longer disclosed, replaced by quarterly publication of single-security shareholding quantities (Sina Finance¹, The Beijing News²).

This section was originally to substitute southbound flows, but that data is likewise not adopted as a conclusion in this report: the southbound net buy turnover read at 15:30 was Stock Connect (Shanghai) +RMB 3.176 billion and Stock Connect (Shenzhen) +RMB 40 million, and the API simultaneously tagged the trading status as "trading". The two channels differ by nearly 80×, and Hong Kong had not yet closed at that point, so this looks more like an unfinished intraday snapshot than full-day data; this report therefore only discloses it and draws no "southbound stance" judgement from it. The "net fund inflow of RMB 42 billion" given by the same endpoint is the daily quota ceiling rather than an actual inflow, and must not be quoted. The Hang Seng Index at that moment was +1.07% to +1.09%.

5.2 Dragon-Tiger List

Not obtained in this report; left blank. The exchanges only disclose the day's Dragon-Tiger List at about 18:00 after the close, and this report was generated at 15:30. This means today lacks exchange-level seat data, so it is impossible to judge whether a stock's move was hot-money behaviour or institutional behaviour — which is exactly the data the rule established in last Friday's recap ("when the two fund-flow bases point in opposite directions, defer to the institutional seats") depends on. Please discount this section's fund-flow conclusions accordingly.

5.3 Sector Net Inflows (Tonghuashun basis)

The conclusion is plain: money is retreating wholesale from "tech hardware" into "resources + consumer + financials + pharma". Seven of the top 8 net outflows are electronics/tech sectors, with combined outflows of more than RMB 47 billion; not one of the top 8 net inflows is an electronics sector.

One detail that must be pointed out: Healthcare Services was +4.59%, 1st market-wide, yet its net inflow was only RMB 750 million, behind precious metals; while Chemical Pharmaceuticals was +1.23% yet saw a net outflow of RMB 3.395 billion. The pharma line's gains far exceed the money it received — the classic signature of a sentiment-driven rather than money-driven move.

5.4 Single-Stock Main-Force Flows

Top 8 net inflows: AMEC (中微公司, 688012, STAR Market) +RMB 890 million / +6.08%; Longsys (江波龙, 301308, ChiNext) +RMB 863 million / +6.70%; Kweichow Moutai (贵州茅台, 600519, SSE main board) +RMB 706 million / +3.03%; Shijia Photons (仕佳光子, 688313, STAR Market) +RMB 686 million / +4.94%; CATL (宁德时代, 300750, ChiNext) +RMB 546 million / +1.86%; Tengjing Technology (腾景科技, 688195, STAR Market) +RMB 449 million / −1.09%; Jingzhida (精智达, 688627, STAR Market) +RMB 400 million / +0.43%; Weichai Power (潍柴动力, 000338, SZSE main board) +RMB 394 million / +6.40%.

Top 8 net outflows: Innolight (中际旭创, 300308, ChiNext) −RMB 7.242 billion / −6.01%; ChangXin Memory (长鑫科技, 688825, STAR Market) −RMB 3.255 billion / −2.25%; Eoptolink (新易盛, 300502, ChiNext) −RMB 3.240 billion / −5.07%; GigaDevice (兆易创新, 603986, SSE main board) −RMB 3.171 billion / −3.36%; Dongshan Precision (东山精密, 002384, SZSE main board) −RMB 2.963 billion / −4.78%; Accelink Technologies (光迅科技, 002281, SZSE main board) −RMB 2.288 billion / −7.64%; BOE Technology (京东方Ａ, 000725, SZSE main board) −RMB 2.102 billion / −2.97%; Luxshare Precision (立讯精密, 002475, SZSE main board) −RMB 1.339 billion / −2.11%.

Innolight (中际旭创, 300308) alone saw a net outflow of RMB 7.242 billion, close to 2.2× the second-largest, and also 2.9× the entire precious-metals sector's total net inflow (RMB 2.493 billion). This is the largest single fund event market-wide today.

---

6. Outlook for Tomorrow

6.1 Main-Line Continuity Assessment

6.2 Key Names to Watch Tomorrow

The following are objects of observation, not buy recommendations. And one thing must be stated first: this report's recommendation book had a hit rate of 2/12 today, underperforming random stock selection on a day when 73.4% of stocks rose. The credibility of this section should therefore be discounted accordingly — please rely on your own judgement.

6.3 What to Avoid Tomorrow

1. The entire optical module / CPO / PCB / CCL chain — Innolight (中际旭创, 300308) with a net main-force outflow of RMB 7.242 billion (the largest market-wide, 2.2× the second-largest); Eoptolink (新易盛, 300502) −RMB 3.240 billion; Dongshan Precision (东山精密, 002384) −RMB 2.963 billion; T&S Communications (天孚通信, 300394) −RMB 1.065 billion; Shengyi Technology (生益科技, 600183) −RMB 1.129 billion; WUS Printed Circuit (沪电股份, 002463) −RMB 1.099 billion; Kinwong Electronic (景旺电子, 603228) −RMB 951 million. Outflows of this magnitude do not end in a day.
2. Memory proxy stocks — ChangXin Memory (长鑫科技, 688825), the only direct proxy, was −2.25% with a net outflow of RMB 3.255 billion, so the pricing verdict on the "Apple tests ChangXin" news is negative. GigaDevice (兆易创新, 603986) was −3.36% with an outflow of RMB 3.171 billion. The sole exception, Longsys (江波龙, 301308) at +6.70% with a net inflow of RMB 863 million, is a lone-survivor pattern and should not be treated as evidence that the branch is restarting.
3. The 5-board highs — Baoding Technology (宝鼎科技, 002552, SZSE main board), Baihua Pharmaceutical (百花医药, 600721, SSE main board). Baoding Technology already has a company clarification announcement (electronic copper foil cannot be used in AI servers), and its Components sector was 2nd from the bottom market-wide today; Baihua Pharmaceutical has a market cap of only RMB 4.46 billion and a large 20-day gain after 5 consecutive limit-ups. The gap at the 3-board tier means they have no relay support.
4. Communication-equipment stocks that hit the limit yesterday — 3 of today's 6 worst decliners are of this type (Wuhan Fingu Electronic (武汉凡谷, 002194) −9.91%, Tongyu Communication (通宇通讯, 002792) −8.31%, Accelink Technologies (光迅科技, 002281) −7.64%).
5. The humanoid-robot concept — Automation Equipment at 86/90 market-wide with a net outflow of RMB 2.845 billion; the drain is not over.
6. Bandwagon names in the broad chemical-pharma advance — Chemical Pharmaceuticals was +1.23% but saw a net outflow of RMB 3.395 billion, a divergence between price and money; the money in the pharma move is in biological reagents and healthcare services, not in chemical pharmaceuticals.

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

1. The Dragon-Tiger List for today must be fetched (after 18:00) before writing tomorrow morning's pre-market. Today's fund-flow conclusions lack exchange seat data, and today happens to have two open questions that require a seat-level verdict: ACM Research (Shanghai) (盛美上海, 688082) +12.16% (institutional accumulation, or a one-day hot-money trade?) and Longsys (江波龙, 301308) with a net inflow of RMB 863 million (a memory reversal, or single-stock money?).
2. The first step of tomorrow morning's pre-market should not be "find the news" but "find the fund set". Today's biggest lesson is: two branches were called right and in the right order, yet still only 2/12 outperformed their sector, because the stock-selection step used financial quality rather than the direction of money. We suggest first listing the set-level classification of today's top 30 net main-force inflows tomorrow morning, then working backwards to the news.
3. Semiconductor equipment needs a formal branch write-up, covering: its relationship to the already-ebbing optical module / PCB complex (a switch, or a lagged co-move), where it stands after a 5-day gain of 23% to 41%, and exactly how the fact of ACM Research (Shanghai)'s ex-non-recurring −15.31% should be priced within the equipment main line — today's handling (a straight Pass) has been proven wrong, but simply reversing it (ignoring ex-non-recurring profit) is equally dangerous.
4. The bonus for "earnings already delivered" needs its weight recalibrated. Cambricon (寒武纪, 688256) posted an interim report of +122.61% and was −6.33% today; Guanghe Technology (广合科技, 001389) posted ex-non-recurring +96.71% and was −0.92% today; ACM Research (Shanghai) posted ex-non-recurring −15.31% and was +12.16% today. All three samples point in exactly the same direction: disclosed earnings have no explanatory power for the next day's move, and may even be inverted (good news realised).
5. Defence equipment was underrated pre-market, and should be assessed tomorrow morning as an independent branch rather than hung under the geopolitics branch.
6. Retain the three types of Pass criteria that were validated today (company clarification announcement, net selling by institutional seats, scheduled liquidity drain), and narrow the Passes based on valuation / financial quality. Today's improvement in the Pass-list hit rate to 63.2% (47.4% last Friday) came mainly from the former.

---

⚠️ Risk warning: this recap is only a post-close review of information and observations and does not constitute investment advice. Data may differ in timeliness or definition — please defer to exchange/company disclosures, and do not use this directly as a basis for trading.