A-Share Market Recap | 2026-07-31 Friday

⚠️ Risk disclaimer (up front): This recap is post-close information review and observation only. It does not constitute investment advice and contains no ratings, target prices, position sizing or buy/sell recommendations. The terms "delivered / did not deliver / called right / called wrong" used below are strictly an after-the-fact reconciliation against this morning's pre-market list, and imply nothing about future returns.

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

1. Time window: 2026-07-31 09:15 (call auction) → 15:30 (cut-off for this report). Today is Friday and the last trading day of July.
2. Market data sources: index and single-stock closing prices use Sina full-market quotes stock_zh_a_spot (5529 valid samples, timestamp 15:30:02); limit-ups / limit-up breaks / performance of yesterday's limit-up names use East Money pools stock_zt_pool_em / stock_zt_pool_zbgc_em / stock_zt_pool_previous_em; industry moves and net inflows use THS industry aggregates stock_board_industry_summary_ths (90 industries); northbound/southbound use stock_hsgt_fund_flow_summary_em.
3. ⚠️ Data gap #1 (must be disclosed, and it is today's single most important gap) — block-trade disclosure list (Dragon-Tiger list) not obtained: stock_lhb_detail_em(20260731) returned an empty object at 15:40 (exchange Dragon-Tiger lists are usually published around 18:00 after the close). This entire report contains no 7/31 Dragon-Tiger figures whatsoever; anywhere seat-level direction is involved it is written as "cannot be confirmed at this time", with no speculation. This is logged as an outstanding item that tomorrow morning's pre-market list must fill in (see Section 6.4).
4. ⚠️ Data gap #2 — no valid basis for single-day northbound net flow: stock_hsgt_fund_flow_summary_em returns 0.000 for the "net buy turnover" of both the Shanghai Connect and Shenzhen Connect northbound records. This is not "net buying of zero"; it is a null value returned by the interface after the exchanges discontinued single-day net-buy disclosure from August 2024. This report does not use it as 0 and writes "no reliable data available"; only the still-valid parts of that interface — the advance/decline counts of northbound-eligible names and southbound net buying — are used. This is the third consecutive day with this same gap.
5. ⚠️ Data gap #3 — East Money industry/concept boards and fund-flow rankings failed across the board: stock_board_industry_name_em, stock_board_concept_name_em and stock_sector_fund_flow_rank returned ConnectionError / 502 Bad Gateway / JSONDecodeError respectively. All sector moves and net inflows in this report are on the THS basis, which is the same source as the THS basis cited in this morning's pre-market list and is therefore comparable across days; but it is not comparable across sources with the "industry" classification used by the East Money limit-up pool — every place below where counts are compared has the basis labelled.
6. ⚠️ Two different bases exist for the limit-up count; this report discloses both side by side: the East Money limit-up pool stock_zt_pool_em shows 99 names; Sina Finance's closing commentary shows 88 stocks at limit-up. The source of the difference has not been verified (it may involve ST stocks, the Beijing Stock Exchange, or whether new listings are included). The body of this report uses the East Money pool figure of 99 throughout, and hereby states that the other basis exists.
7. What this report does not do: it uses no narrative of the form "today's outcome proves the pre-market call was right all along". Section 8 of this morning's pre-market list wrote down 6 sector criteria, 5 single-stock criteria and 10 risk signals of its own accord. The reconciliation in this report is measured against those observable conditions already written down in black and white before the open, not against standards invented after the fact. Where the call was wrong, it is recorded as wrong.

---

0. One-line recap

Today's sentiment temperature: euphoric (but a "split euphoria") — 99 limit-ups, 0 limit-downs, 4690 advancers, an all-A median of +2.03%; at the index level it was the strongest day since the start of July. But there were 107 limit-up breaks, a 51.9% break rate, and almost none of the names that gapped up held their gains.

Today's strongest theme was neither "AI hardware oversold bounce" — ranked #1 on this morning's pre-market list — nor "memory price hikes" ranked #2, but "domestic AI applications / software / media", which appeared in the pre-market list only as a single C-rated ChinaJoy headline. Of THS's 90 industries, the top 8 gainers were, in order, IT services +6.29%, software development +6.15%, automation equipment +5.34%, cultural media +5.28%, film & cinema +4.97%, computer equipment +4.60%, gaming +4.50%, communication services +4.47% — semiconductors at +4.23% ranked only 10th, and components at +4.33% ranked 9th.

And the answer from money flows is sharper than the price moves: software development net inflow +RMB 9.029 billion, IT services +RMB 5.465 billion, cultural media +RMB 4.795 billion — the top three of the entire market; semiconductors net outflow −RMB 20.502 billion and components net outflow −RMB 7.289 billion — the bottom two of the entire market (THS industry net-inflow basis). In other words, the two sectors that rose most violently today, semiconductors and components, were the two most heavily sold — "price up, money out". This is the sector-level amplification of the same pattern logged in yesterday's recap: "德明利 (Demingli Electronics) hit limit-up but showed RMB 706 million of net Dragon-Tiger selling".

Pre-market list hit summary: of the 30 main-board names on the main list, 12 hit on the strict basis (40.0%), and 15 on the loose basis (partial delivery counted as half) (50.0%). By tier the differences are stark: "watch only" was right on 6 of 7 (85.7%), the most accurate tier on the whole list; "priority deep-dive" delivered 2 of 4; "watch closely" delivered strictly on only 4 of 18; and the 2 "dual-dimension Pass" names — 太极实业 (Taiji Industry, 600667) and 江海股份 (Nantong Jianghai Capacitor, 002484) — were 0/2, both called wrong, with 太极实业 closing at a locked limit-up of +10.03%.

Tone for tomorrow (Monday 8/3): the theme rotation has already happened, but the new theme (AI applications) is itself in the high-risk position of "day 1 of launch + a 51.9% limit-up break rate at elevated levels", while the old theme (AI hardware / memory) is in a "index up, money out" divergence. The tone is set as a divergence day: neither continuation nor washout, but a day that must be characterised by two things — "can AI applications produce a second consecutive limit-up" and "can semiconductor money flow turn from negative to positive".

---

1. Market overview

1.1 Indices and turnover (Sina basis, timestamp 15:30:02)

• Turnover: Shanghai RMB 1.19 trillion, Shenzhen RMB 1.35 trillion, combined roughly RMB 2.54 trillion (RMB 2.56 trillion on the all-A basis including the Beijing Stock Exchange). A clear expansion versus yesterday, and the one uncontested piece of positive evidence in this bounce.
• ⚠️ The single most important fact at the index level: all three major indices traced a "gap up → fade", not a "gap down → rally". The ChiNext Index opened +5.15% and closed +3.06%; the STAR Market Composite was up over 6% intraday and the STAR 50 closed +2.99%. Today's high came near the open, and that single fact determines the character assessment of every stock below.

1.2 Sentiment indicators

Sentiment temperature over the last 6 trading days (East Money pool basis)

1.3 Sentiment call: euphoric, but lower quality than the 7/27 climax

Three pieces of evidence, side by side, none dispensable:

1. Positive: limit-downs to zero, 4690 advancers, turnover expanding to RMB 2.54 trillion, 99 limit-ups — every breadth indicator is the best of the week.
2. Negative: a 51.9% limit-up break rate, the highest of the week and almost 5 times that of the 7/27 climax day (10.5%). On two days that both had "about 100 limit-ups", 7/27 saw only 13 breaks while today saw 107 — the two days are completely different in character: 7/27 was a climax with ample follow-through; today was a broad rally with insufficient follow-through.
3. Structurally negative: by industry, the limit-up-break pool was led by components with 15 names and semiconductors with 8 — the heaviest breakage happened precisely in the chain this morning's pre-market list pushed hardest.

Conclusion: today was a "broad rally with limit-downs at zero", not a "theme confirmation day". Summed up in one figure: 879 stocks rose more than 5%, but only 99 held a limit-up, while 107 reached limit-up and failed to hold it.

1.4 One macro data point unconfirmed pre-market that landed today

The July official manufacturing PMI was published today at 09:30: 49.2%, down 1.1 percentage points month-on-month (non-manufacturing business activity index 49.0%, −1.2 m/m; composite PMI output index 49.3%, −1.3; production index 49.9% and new orders index 48.5%, −1.5 and −2.7 respectively).

• Item 8 of Section 8.4 of this morning's pre-market list explicitly wrote: "this report could not verify today's exact release schedule or market expectations, so it falls under 'no reliable data available' — please confirm it yourself before the open, as it could change the tape the instant trading starts". The outcome: the data was released as usual and was clearly weaker (the Nth consecutive month below the boom/bust line, with a large m/m drop), and the market ignored it entirely, with indices gapping up and running higher.
• But there is one structure within it that is highly consistent with today's tape: equipment manufacturing PMI 51.4% and high-tech manufacturing PMI 53.3%, clearly above the manufacturing total and still expanding briskly; consumer goods 47.8% and energy-intensive industries 47.0%. The divergence of "high-tech expanding, consumption and energy-intensive contracting" points in the same direction as today's tape structure — technology/AI applications sharply higher, baijiu +0.63%, banks −0.77%, coal +0.82%.
• Lesson for the recap (usable tomorrow morning): flagging the PMI as an "unconfirmed risk" pre-market was correct, but it only said "could change the tape" and did not spell out the branch: "if the data is weak but internally divergent (high-tech strong, consumption weak), the tape may rise rather than fall and style divergence may widen". Next time an unconfirmed macro release comes up, give a two-sided scenario rather than a one-directional risk warning.

---

2. Reconciliation against the pre-market list

Judging benchmark (standard set first, reconciliation second, to avoid hindsight): today's all-A median change was +2.03% and the SZSE Component was +2.21%. On that basis, for the "priority deep-dive / watch closely" tiers: a close of ≥ +5.00% counts as "delivered"; +2.03%~+5.00% counts as "partially delivered"; < +2.03% counts as "did not deliver". For the "watch only" tier: a gain below the all-A median counts as "called right", ≥ +5% counts as "called wrong". For the "Pass" tier: a limit-up or a large gain counts as "called wrong". Open % and high % are listed alongside, because "the open was the high" was today's dominant pattern across the market, and looking only at the close would badly misread it.

2.1 Priority deep-dive (4 names)

Priority deep-dive delivery rate: 2/4 (50%).

2.2 Watch closely (18 names)

Watch closely tier: strictly delivered 4/18 (22.2%), including partial delivery 10/18 (55.6%).

2.3 Watch only (7 names) — the most accurate tier on the whole list

Watch only tier accuracy: 6/7 (85.7%).

2.4 Pass-list reconciliation — today's biggest misjudgement is here

⭐ The single most important conclusion from the Pass tier: the two-level classification of "earnings-dimension Pass / dual-dimension Pass" introduced pre-market produced an extremely clean test result today — all 9 earnings-dimension Passes were classified correctly (they did indeed rise, and pre-market explicitly said "not claiming it will not rise"); both dual-dimension Passes (太极实业, 江海股份) were called wrong, and that is precisely the only category that argues for "do not participate". This classification framework was introduced in yesterday's recap to correct the error of "Passed it and it hit limit-up"; today proves the classification itself is sound, and the problem lies in the far-too-loose test for "which names qualify for a dual-dimension Pass".

2.5 Hit-rate summary

Branch-ranking hits: 1/3 partially delivered (see Section 3), and the strongest theme was missed entirely.

2.6 Item-by-item reconciliation of Section 8 "observable verification conditions" from the pre-market list (the highest-information section here)

Section 8 of this morning's pre-market list said verbatim: "the observable conditions in this section take priority over the branch ranking in Section 2 and the single-stock master list in Section 3. Where the two conflict, this section governs." So this is the real report card.

8.2 Sector signals (6 items)

Sector signals: 3 held / 3 failed. But among the 3 that held, items 1 and 5 are loose conditions that hold automatically on a broad rally day; the genuinely discriminating items 2, 3, 4 and 6 were 1 held / 3 failed.

8.4 Risk signals (10 items)

⭐ The Section 8.1 signal that "takes priority over the A-share auction" (look at Korean and Taiwanese equities first): pre-market wrote "if Korean/Taiwanese semiconductors gap up and fade, then the A-share gap-up at 09:25 is very likely the high of the day". This report did not obtain today's intraday data for Korean/Taiwanese semiconductors and cannot verify whether that signal gave an advance warning — but the A-share outcome was indeed "the gap-up open was the high". This is logged as an outstanding item.

2.7 One-line self-review: which categories were called right, which wrong, and what to change tomorrow morning

Three categories called right (reusable):

1. The separation of "the news is real but the price has already run ahead" — 永鼎股份 (600105) closed −2.79% on a market-wide broad rally day; this methodology was the sharpest single item on the whole list today.
2. "Resilience on the way down = no oversold-bounce room" — 紫光国微 (002049) at +0.50% was last within the entire chain; a precise micro call.
3. The seesaw structure — banks −0.77%, insurance −0.70%, baijiu +0.63% were the worst in the market, strictly inverse to the technology bounce.

Three categories called wrong (must be changed):

1. ⭐ The most serious: treating "the overnight US-equity mapping" as the sole stock-selection entry point, and therefore being systematically blind to the "domestic AI application line that has nothing to do with overnight US equities". Of the 26 pre-market headlines, only item 23 (ChinaJoy) was AI-application-related, rated C ("weak persistence of sentiment stimulus"). Yet on ChinaJoy's opening day, gaming +4.50%, cultural media +5.28%, film & cinema +4.97%, and 4 limit-ups in the digital media direction, together with a combined net inflow of +RMB 14.494 billion into software development / IT services — this was the real theme of the day, and it did not even have a branch in the pre-market list.
   • What to change tomorrow morning: in the news overview, any exhibition, launch event or opening ceremony actually taking place today/tomorrow must not be given a C rating outright. ChinaJoy and the solar Yancheng meeting were two events held on the same day; pre-market gave solar an A and ChinaJoy a C, and the result was solar +1.83% underperforming the market while media and gaming led the whole market — the ratings were exactly inverted. The criterion should be changed to: "whether the event actually occurs on the day" takes priority over "the industrial hardness of the event", because A-shares price the event window, not the industrial hardness.
2. ⭐ The second most serious: treating "a gap-up open" as evidence that the news has been digested, rather than as the risk itself. Section 8.1 pre-market turned the size of the gap into a "3%–5% is most balanced" range criterion, but today's fact is that the size of the gap was significantly negatively correlated with the full-day return — 兆易/德明利/风华高科, which all opened +10%, closed +2.02 / −1.06 / −2.28 respectively; 工业富联, which opened +6.21%, closed +5.39%; 紫光股份, which opened +3.87%, closed +3.84%.
   • What to change tomorrow morning: write the empirical rule "for stocks gapping up ≥ 8%, the expected value of the day's close should be revised down, not up" directly into the pre-market list, and make "is the opening price the high of the day" the number-one intraday observation item.
3. The test for "dual-dimension Pass" is too loose — 太极实业 (600667) and 江海股份 (002484) were 0/2 today. The fundamental falsification (at company-announcement grade) is solid, but it was wrongly extrapolated into "the sentiment dimension is unsupportive too".
   • What to change tomorrow morning: the sentiment-dimension test for a "dual-dimension Pass" must add a hard threshold — any name with a maximum drawdown of more than 40% over the last 30 days may never be judged "sentiment dimension unsupportive" (太极实业 drew down −51.2% from 7/1 to 7/30, 德明利 drew down −60.2%; both are natural fuel for an oversold bounce). A sharp collapse is itself sentiment ammunition, and this is the most expensive lesson of this session.

One more ranking failure to log: the fine-grained rankings made pre-market within branches (packaging & testing: 长电 > 通富 > 华天; AI PCB: 沪电 > 生益 > 深南 > 东山) were almost entirely inverted today — the actual closes were 华天 +3.76 > 长电 +1.86 > 通富 +1.50, and 深南 +8.26 > 沪电 +7.46 > 东山 +5.98 > 生益 +5.56. Conclusion: on an oversold-bounce day, ranking within a branch on one-point differences in "fundamentals / money flow" is useless; pre-market should give only "branch membership" and no within-branch ordering.

---

3. Verification of today's leading theme

3.1 Did pre-market identify the strongest theme: no

The pre-market ranking of the top 3 branches was: ① AI hardware oversold bounce ② memory price hikes ③ solar anti-involution.

Today's actual top 10 industries by gain (THS, 90 industries):

Conclusion: the pre-market list covered none of the top 8 gainers; the branches it pushed hardest ranked 9th and 10th, and were the two sectors with the largest net outflows in the market.

3.2 Unexpected themes missed pre-market (3 items)

1. ⭐ Domestic AI applications / large models / media and gaming — the real theme of the day. At least three attributable public catalysts exist: ① ChinaJoy 2026 opened in Shanghai today (7/31) under the theme "Journey with AI", with more than 500 exhibitors and over a thousand titles (this was written pre-market, but rated C); ② Hong Kong-listed large-model names rose in tandem, with 智谱 (Zhipu AI) +20.53% and 迈富时 (Marketingforce) up more than 20%; ③ compute-leasing names rallied collectively, with 利通电子 (Litong Electronic, 603629), 美利云 (000815), 盈峰环境 (Infore Environment, 000967) and 证通电子 (Zhengtong Electronics, 002197) at limit-up, and 深信服 (300454) +14.78%. ⚠️ On the specific attribution of 蓝色光标 (300058)'s limit-up, this report has obtained only after-the-fact analysis from brokers/media (AI + overseas expansion strategy, AI-related revenue of over RMB 3.7 billion in 2025, overseas revenue of over RMB 56.4 billion), and has not obtained the company's original announcement today, so none of those figures is cited as factual evidence.
2. Automation equipment +5.34% (net inflow +RMB 1.453 billion) — 绿的谐波 (Leaderdrive, 688017) +16.34%, 雷赛智能 (Leadshine Technology, 002979) at limit-up. Directionally consistent with the event window of "宇树科技 (Unitree Robotics) price consultation on 8/5 and subscription on 8/10", but this report has obtained no evidence directly linking today's rise to the Unitree IPO, so no causal attribution is made; it is logged only as a direction to verify tomorrow morning.
3. Precious metals +4.00% (14 up, 0 down) — see the side-by-side note in Section 2.4.

---

4. Limit-up tier structure / high-streak names

4.1 High streaks and consecutive limit-up structure

⚠️ The core feature of the consecutive-limit-up structure: 3 of the 6 names in the 2-consecutive tier (普联软件, 泛微网络, 税友股份) belong to software / IT services — the new theme has already secured position at the most critical relay slot, "the 2nd consecutive limit-up", and this is the first thing to watch in judging whether AI applications can continue tomorrow. Meanwhile the high-streak names at 3–9 consecutive (高争民爆, 一鸣食品, 传智教育, 爱丽家居) have nothing to do with today's theme and are independent sentiment high-streak names.

4.2 Follow-through rate (performance today of the 52 limit-up names from 7/30)

• 31 up / 20 down / 1 flat, average gain +2.85% (slightly above the all-A median of 2.03%).
• 10 hit limit-up again today, a follow-through rate of 19.2%: 普联软件 (300996) +19.97, 中国宝安 (000009) +10.07, 香江控股 (600162) +10.03, 高争民爆 (002827) +10.02, 税友股份 (603171) +10.01, 爱丽家居 (603221) +10.00, 神雾节能 (000820) +10.00, 一鸣食品 (605179) +9.99, 泛微网络 (603039) +9.99, 传智教育 (003032) +9.99.
• ⚠️ On a huge broad-rally day with 99 limit-ups, 0 limit-downs and 4690 advancers, the follow-through rate of yesterday's limit-up names was only 19.2% with 20 of them falling — which says today's money was "rotating into a new direction" rather than "relaying yesterday". This point matters a great deal for tomorrow's judgement.

4.3 Core names (large turnover + high recognisability)

4.4 Broken streaks and limit-up breaks

• 107 limit-up breaks, a break rate of 51.9%, the highest of the week (the 7/27 climax day was only 10.5%).
• Leading industries in the limit-up-break pool: components 15, semiconductors 8, auto parts 7, plastics 5, automation equipment 5, software development 5 — the heaviest breakage was precisely in the chain pushed hardest pre-market.
• The 5 largest-turnover names in the limit-up-break pool are all names called out pre-market: 兆易创新 (603986) RMB 34.043 billion, 东山精密 (002384) RMB 21.468 billion, 德明利 (001309) RMB 17.355 billion, 中国巨石 (China Jushi, 600176) RMB 15.803 billion, 风华高科 (000636) RMB 13.582 billion.
• Broken streaks (limit-up yesterday, down today): 20 names; the worst 5: 恒尚节能 (Hengshang Energy Saving, 603137) −4.62%, 广电电气 (601616) −3.68%, 新芝生物 (Xinzhi Biotechnology, 920685) −3.44%, 陆家嘴 (600663) −3.02%, 和泰机电 (Hetai Electromechanical, 001225) −2.82%. Among them 广电电气 and 陆家嘴 were pre-market Pass / watch-only names, so those calls were right.
• Distribution of first limit-up seal times: call auction–9:30 6 names, 9:30–10:00 45 names, 10:00–11:00 29 names, 11:00–13:00 7 names, 13:00–14:30 11 names, 14:30–close 1 name. 74.7% of limit-ups occurred before 11am, with almost no new seals in the afternoon — completely consistent with the indices' "gap up then fade" pattern, further showing that today was a "single morning wave".

---

5. Fund flows

5.1 ⭐ Today's most important fund-flow fact: the theme rotation is more complete in money terms than in price terms

Top 6 net inflows (THS industry basis, RMB 100 million)

Top 6 net outflows (same basis)

⭐ This comparison table is the single most valuable item in everything today:

• Semiconductors and components ranked 9th and 10th on the gainers board and simultaneously 1st and 2nd on the net-outflow board. Combined net outflow of RMB 27.791 billion.
• Software development, IT services and cultural media ranked 1st, 2nd and 4th on the gainers board and simultaneously 1st, 2nd and 3rd on the net-inflow board. Combined net inflow of RMB 19.289 billion.
• One side is "rising while money leaves", the other is "rising while money arrives". Judged only on price change these two lines both look strong (both up 4%–6%), but on the money dimension they are exact opposites.
• ⚠️ Basis reminder: the THS "net inflow" is a main-force fund basis (a net figure computed by stratifying individual trade sizes) and is not equal to true market-wide fund flow, nor can it be mixed with the Dragon-Tiger or northbound bases; semiconductor turnover for the day was RMB 458.373 billion, so the RMB 20.502 billion net outflow is about 4.5% of it — a medium-strength signal, not decisive evidence, but it points the same way as three independent pieces of evidence (zero limit-ups + a 51.9% break rate + huge volume with a stalled price), and it is the four together that constitute this report's judgement.

5.2 Northbound flows: no reliable data available (third consecutive day)

• stock_hsgt_fund_flow_summary_em returns 0.000 for the "net buy turnover" of both the Shanghai Connect and Shenzhen Connect northbound records. As stated in disclosure item 4, this is a null value from the interface after the exchanges discontinued single-day net-buy disclosure from August 2024; it is not "net buying of zero". This report does not use that figure, writes "no reliable data available", and makes no speculation.
• The part of that interface that is still valid and usable — the advance/decline counts of northbound-eligible names: Shanghai Connect 1245 up / 52 flat / 343 down; Shenzhen Connect 1582 up / 36 flat / 257 down. This can only describe breadth within the northbound-eligible pool and cannot be used to infer money direction.
• Southbound (usable): Stock Connect (Shanghai) net buying +RMB 1.297 billion, Stock Connect (Shenzhen) −RMB 185 million, with the day's fund balance at RMB 42 billion for both; the Hang Seng Index −0.03% (same interface basis).

5.3 Dragon-Tiger list: not obtained today, logged as an outstanding item

• stock_lhb_detail_em(start_date=20260731, end_date=20260731) returned an empty object at 15:40 (TypeError: 'NoneType' object is not subscriptable). Exchange Dragon-Tiger lists are usually published around 18:00 after the close, so this report's cut-off is earlier than the publication time.
• Therefore this report cites no 7/31 Dragon-Tiger figures anywhere. In particular, the following directions — the most worth checking today — are all marked "cannot be confirmed at this time":
  1. The nature of the buy/sell seats behind 太极实业 (600667)'s limit-up (retail speculators, institutions, or Shanghai Connect) — this is the single most critical piece of information for judging tomorrow morning whether it can post a 2nd consecutive limit-up;
  2. The seat structure behind the two 20cm limit-up leaders 蓝色光标 (300058) and 昆仑万维 (300418);
  3. The net buy/sell direction of 兆易创新 (603986) and 德明利 (001309) during their "one-line limit-up opened → break" process — both were net-sell samples on the Dragon-Tiger list yesterday, and today's direction could directly confirm or refute the judgement that "money is leaving the memory chain";
  4. Whether the money behind 长电科技 (600584)'s +RMB 317 million of net buying from the fifth-largest buy seat yesterday added or exited today.
• ⚠️ This is the same kind of outstanding item as yesterday's recap logging "the 7/30 Dragon-Tiger list was not obtained" (which was filled in and cleared in this morning's pre-market list). It is logged again this time, and tomorrow morning's pre-market list must fill it in.

5.4 Summary of sectors leading main-force net inflows

Net inflows are highly concentrated in two lines, "domestic AI applications + broad manufacturing equipment": software development +9.029, IT services +5.465, cultural media +4.795, auto parts +2.710, computer equipment +2.615, general equipment +2.413, gaming +2.342, industrial metals +2.306, communication services +1.901, white goods +1.537, precious metals +1.453, automation equipment +1.453 (RMB billion).

Net outflows are highly concentrated in "AI hardware + procyclical financials": semiconductors −20.502, components −7.289, construction & decoration −3.754, batteries −1.973, electronic chemicals −1.610, brokers −1.367, energy metals −1.097, consumer electronics −1.083 (RMB billion).

---

6. Outlook for the next session (Monday, August 3)

⚠️ Qualifier up front: today is the last trading day of July, so month-end ranking and portfolio-rebalancing behaviour is present; the next trading day is the Monday after the month turn, separated by two calendar days and a full weekend (overseas markets — US equities on 7/31 and Asia-Pacific on 8/1 — will by then constitute new external variables; none of these had happened as of this report's cut-off, and no prediction is made). Today's single-day conclusions are therefore less extrapolable than usual, and the confidence of the judgements below should be discounted accordingly.

6.1 Theme-continuation judgements

6.2 Key focus for tomorrow (branches + main-board names + verification points)

Stated again: the following is a ranking of research attention and observable verification conditions, and contains no buy/sell recommendations, position sizing, target prices or ratings.

6.3 Avoid / downgrade for tomorrow

6.4 Input prompts for tomorrow morning's pre-market list (each item must be responded to)

1. ⭐ The outstanding item that must be filled in (highest priority) — the 7/31 Dragon-Tiger list. Published around 18:00 after the close; it must be obtained tomorrow morning with stock_lhb_detail_em(20260731), focusing on the four directions listed in Section 5.3, especially the seat nature behind 太极实业 (600667) and the net buy/sell direction of 兆易创新/德明利 during their limit-up breaks. The same outstanding item from yesterday was cleared this morning; this one must not be deferred again.
2. ⭐ The branch that must be added — domestic AI applications / software / IT services / media and gaming. Today it took 7 of the top 8 places on the gainers board and all of the top 3 on the net-inflow board, while the pre-market list did not even have a branch for it. Tomorrow morning it must be given its own branch with main-board name coverage, and note one structural difficulty: the highly recognisable names in this theme (蓝色光标 300058, 昆仑万维 300418, 中文在线 300364, 深信服 300454) are almost all on ChiNext, so under the "main board first" rule the main-board candidates are inherently weaker. This must be explained to readers up front rather than padding out a main-board list.
3. ⭐ The rating rule that must be changed — any exhibition, launch event or opening ceremony "actually taking place today/tomorrow" must not be given a C rating outright. Today ChinaJoy (rated C) and the solar Yancheng meeting (rated A) were held on the same day, and the result was media and gaming leading the whole market while solar underperformed it — the ratings were exactly inverted. New criterion: "whether the event actually occurs on the day" should carry more weight than "the industrial hardness of the event".
4. ⭐ The empirical rule that must be written in — the size of the gap-up is negatively correlated with the day's return. Today's samples: 兆易创新/德明利/风华高科, which opened +10.00%, closed +2.02 / −1.06 / −2.28; 普冉股份 (688766), which opened +20.00%, closed +0.50; 工业富联, which opened +6.21%, closed +5.39%; 紫光股份, which opened +3.87%, closed +3.84%. Tomorrow morning's pre-market list should write in directly that "for stocks gapping up ≥ 8%, the expected value of the day's close is revised down", and make "is the opening price the high of the day" the number-one intraday observation item.
5. ⭐ The test that must be tightened — the sentiment-dimension threshold for a "dual-dimension Pass". Both dual-dimension Passes today (太极实业 600667, 江海股份 002484) were called wrong. New hard threshold: any name with a maximum drawdown of more than 40% over the last 30 days may never be judged "sentiment dimension unsupportive"; at most it can be judged an "earnings-dimension Pass". (太极实业 drew down −51.2% from 7/1 to 7/30, 德明利 drew down −60.2%.)
6. The practice that must be dropped — fine-grained ordering within a branch. Today packaging & testing (pre-market 长电>通富>华天, actual 华天>长电>通富) and AI PCB (pre-market 沪电>生益>深南>东山, actual 深南>沪电>东山>生益) were almost entirely inverted. On an oversold-bounce day, give "branch membership" only, not within-branch ordering on one-point differences.
7. The three effective methodologies that must be retained: ① the price separation of "the news is real ≠ you should chase it" (validated by 永鼎股份 today); ② "resilience on the way down = no oversold-bounce room" (validated by 紫光国微 today); ③ the technology/defensive seesaw (validated by banks and insurance being the worst today).
8. The fund-flow criterion that must continue to be tracked — the divergence between "sector price change vs sector net inflow". Today's semiconductors +4.23% / −RMB 20.502 billion and components +4.33% / −RMB 7.289 billion were the most important signals in the market. Recommend upgrading this into a fixed section of the pre-market list: for every sector admitted as a branch, the previous day's price change and net inflow must both be listed, and a divergence in direction forces a downgrade.
9. Two external variables that need to be verified pre-market: ① US equities on 7/31 and Asia-Pacific markets on 8/1 (neither had happened as of this report's cut-off); ② the actual performance of Korean/Taiwanese semiconductors today (7/31) — Section 8.1 pre-market set this as the "signal zero" with "priority over the A-share auction", but this report obtained no data and cannot verify whether that signal gave an advance warning; this is likewise logged as an outstanding item.
10. Calendar events to be aware of in advance: 宇树科技 (Unitree Robotics) preliminary price consultation on 8/5, online and offline subscription on 8/10, payment on 8/12 (today automation equipment rose +5.34% with net inflow of +RMB 1.453 billion, directionally consistent with that event window, but this report obtained no direct causal evidence, so tomorrow morning it must be handled under the existing rule that "any supply relationship without an announcement behind it is treated as a pure concept mapping"); ChinaJoy runs through 8/3, with Monday the last day.

---

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