A-Share Market Recap | 2026-07-28 Tuesday

Data basis: index and stock quotes are as at the 2026-07-28 close (Tencent quotes qt.gtimg.cn, scraped at 15:31); the limit-up / limit-down / broken-seal pools are Eastmoney's push2ex same-day pools (date=20260728); sector moves and large-order net inflows come from the Tencent quotes sector interface (proxy.finance.qq.com).

Three data gaps, disclosed up front (iron rule: if it cannot be found, write "no reliable data"):

1. Today's northbound net inflow: no reliable data. Eastmoney's Stock Connect history interface (RPT_MUTUAL_DEAL_HISTORY) returns null for all of the NET_DEAL_AMT / FUND_INFLOW / ACCUM_DEAL_AMT fields for every trading day from 07-22 to 07-27. This report does not fabricate northbound figures; Section 5 substitutes industry / concept large-order net inflows instead.
2. Today's (07-28) top-trader disclosure list: not yet published. The interface returns "returned data is empty" (the list is usually disclosed after 18:00). Section 5 instead cross-verifies using the 07-27 list plus the actual next-day (i.e. today's) performance, a basis that is in fact more useful for a recap.
3. Full-day advancers/decliners: not obtained. The full-market detail interface push2.eastmoney.com kept returning 502 today. The text only cites Eastmoney's midday basis (more than 2800 advancers, more than 2500 decliners); that figure does not represent the close, so please do not judge full-day breadth from it. The three figures of 61 limit-ups / 49 limit-downs / 20 broken seals are exact closing figures.

---

0. Recap in One Sentence

Sentiment temperature: a tech freeze and thematic exuberance, coexisting on the same day. Today's strongest mainline was not any line on the pre-market list, but lithography (triggered by a foreign-media report on domestic lithography machines) and software / IT localisation (11 limit-ups), plus defensive absorption into consumer staples and banks; of this morning's pre-market list, all 6 positive recommendations closed down, 4 of them at limit-down, an absolute hit rate of 0%, while the avoid / Pass list hit 14/20 (70%) — this is a report where "everything bearish was right and everything bullish was wrong". Tomorrow's tone: day one of a retreat across the whole AI hardware chain, do not catch the falling knife; the thematic ladder actually thickened (consecutive-limit-up names went from 8 to 20), but the FOMC decision lands at 02:00 tomorrow, so position size and overnight risk are mismatched.

---

1. Market Overview

Indices and turnover

Combined turnover across the two markets was about RMB 2.03 trillion (949.683 + 1,076.098 = RMB 2,025.781 billion), down about RMB 60 billion from yesterday's RMB 2.09 trillion.

A structural fact that is easily overlooked: of yesterday's RMB 2.09 trillion, 长鑫科技 (ChangXin Technology, 688825) alone contributed RMB 141.1 billion, and today that stock's turnover fell to RMB 44.428 billion. Stripping out ChangXin, yesterday's base was about RMB 1.95 trillion and today's about RMB 1.98 trillion — the existing capital did not leave, it actually rose slightly. Today is not capital exiting; it is capital rotating within the market from tech heavyweights into thematic stocks and defensive sectors. This point determines every judgement that follows.

Limit-ups, limit-downs and the ladder

Sentiment reading: split

Today cannot be described with a single temperature. It must be seen in two layers:

• The institutional-crowding layer: freezing. Limit-downs exploded from 5 to 49, concentrated in CPO / memory / CCL / PCB; large-order net outflows were RMB 77.920 billion from the chip concept, RMB 53.545 billion from the electronics industry and RMB 44.607 billion from CPO. The ChiNext Index at −7.35% and the STAR 50 at −6.33% are the result of a single-day de-rating of heavyweight names.
• The retail-speculator thematic layer: exuberant. The concept index "yesterday's consecutive limit-ups" was +4.81% today, "yesterday's limit-ups" +1.00%, and "yesterday's first boards" +1.02% — yesterday's limit-up names were collectively green today. The consecutive-limit-up ladder expanded from 8 to 20 names, so the ladder thickened rather than collapsed. The broken-seal rate of 24.7% is moderate-to-low.

Conclusion: today was a structural switch, not a broad retreat. The pre-market report's assumed risk scenario was "a one-day oversold bounce with the ladder collapsing" — the ladder did not collapse; what collapsed was the layer the pre-market report itself recommended.

---

2. Reconciliation Against the Pre-Market List

2.1 Name-by-name reconciliation of the 17 on the overall ranking

The pre-market conclusions are taken from the conclusion column of Section 3, "Overall Stock-Level Catalyst-Strength Ranking", in reports/a-share/2026-07-28.md.

2.2 Reconciliation of the observation zone and Pass list (key items selected)

2.3 Hit rate

Two separate bases are required, because the two halves of this report are of completely opposite quality:

Overall: this pre-market report's "landmine detection" passes (70%), while its "stock selection" scores zero (0%).

2.4 Review: what kind of call was right, what kind was wrong, and how to change tomorrow

The kind that was right: falsification using primary financial verification. 景旺电子 (603228), 创新医疗 (002173), 天赐材料 (002709), 亨通光电 (600487), 长芯博创 (300548), 立新能源 (001258) — of these six acts of "overturning the popular narrative", five delivered directly today as limit-downs or heavy falls. The most expensive part of the pre-market report (primary verification plus the risk-auditor rework) generated all of the positive return.

The kind that was wrong: treating "hard earnings evidence" as "it will rise tomorrow". This is today's core lesson, in four parts:

1. An earnings pre-announcement is hard evidence about the past, not a directional signal. The main reason the top five made the list was four H1 pre-announcements (Shengyi +117131%, Goldenmax +9361063%, Honghe +280364%, Jushi +6585%). But those were released on 7/13–7/15, making today T+13~15, long since priced. The pre-market report itself recorded "Shengyi is −6.3% since the pre-announcement" and "Goldenmax is −44.6% over the 20 days since the pre-announcement" — the evidence was written into the report, yet it did not change the ordering.

2. The report identified every risk signal but never gave the risk signals veto power. The pre-market text wrote: every mainline name's volume ratio below 1 (Shengyi 0.74 / WUS 0.64 / Shennan 0.82), the SOX down more than 2% overnight, a hawkish-leaning FOMC, a consecutive-limit-up ladder of only 8, and "waiting for the core leaders to confirm on volume is safer than chasing yesterday's limit-up boards". Every one of those judgements was correct today. The problem is that they were placed in the "risk warnings" while the "5 most worth watching" was output as usual. The perception was right; the output structure buried the perception.

3. The overnight overseas read-across was only half used. Row 1 of the pre-market news table explicitly wrote "sell chips, buy software and Chinese assets — precisely hitting yesterday's strongest A-share line, semiconductors/memory" and recorded Google +3% and Microsoft +2%. The directional judgement was entirely correct. But the report treated it only as risk-column text for Branches 1 and 2, neither downgrading the ordering on that basis nor creating a new "software" branch — while today software / IT services had 11 limit-ups and was the second-strongest mainline on the board. It read the mapping correctly, yet only picked stocks on the losing half.

4. "Oversold" was treated as a plus. One of the reasons 金安国纪 (002636) made the list was "−44.6% over 20 days, −48.8% from the high, the deepest oversold". Today it fell −10.00% to limit-down. In a downtrend, "oversold" only means "it has not finished falling".

Four specific changes for tomorrow's pre-market report (expanded in Section 6 ④): promote the overnight overseas read-across to an ordering variable rather than risk text; use earnings pre-announcements only to "exclude", not to "rank"; treat a deep drawdown as a deduction only, never a plus, until a volume-backed bottoming appears; and allow "no positive list today" as a legitimate output.

---

3. Verifying Today's Mainlines

3.1 The three pre-market mainlines: all falsified

Did the pre-market report identify the strongest mainline? — No, it got it backwards. The two branches ranked 1 and 2 pre-market were the two worst performers in the whole market today. The communications industry was −10.88% and electronics −7.02%, the two weakest primary industries.

3.2 The three real mainlines the pre-market report missed entirely

④ Brain-computer interface (rated "C, explicitly not worth chasing" pre-market) kept strengthening today: 创新医疗 (002173) +10.00% (3 boards in 4 days), 南京熊猫 (600775) +9.99% (2 consecutive boards), 爱朋医疗 (Aipeng Medical, 300753) touched the board intraday and closed +18.58%. Not one item of the pre-market fundamental falsification has been overturned, yet the price has moved against the fundamentals for three straight days.

3.3 Why the market fell today (an external trigger, not endogenous to A-shares)

Two independent negatives stacked:

1. The overnight US memory / semiconductor chain kept selling off. Per CNR and NBD, SanDisk, SK Hynix ADRs, Western Digital, Seagate and Micron fell about 11%, 7.5%, 4.2%, 4.1% and 2.3% respectively, with ASML down 6% and Nvidia down nearly 5% (CNR³, NBD⁴, JRJ¹).

   ⚠️ Basis to be verified: this morning's pre-market report recorded, for the same trading day, "SanDisk −8%, SK Hynix −5%, ASML −4%, Micron −3%, Nvidia turning from gain to loss". The two sets differ significantly for SanDisk, SK Hynix, ASML and Nvidia. This report cannot determine which are the final closing prices and simply presents both faithfully. If the pre-market figures were an intraday snapshot rather than closing prices, then the pre-market report understated the scale of the overseas sell-off that day — and that was precisely the direct trigger for today's A-share fall, making it a data-process problem requiring immediate repair.

2. A negative specific to the CPO sector. Per an Eastmoney Caifuhao post, two negatives fermented at midday on 7/28: an industry-survey rumour that overseas cloud providers' capex pace is slowing with order divergence (fermenting since 7/26), and concerns about CPO system yield bottlenecks and a delay in large-scale commercial deployment (Eastmoney Caifuhao⁵).

   ⚠️ Both are rumour / survey grade, not filing grade, and under the pre-market report's scoring model they are low-weight sources. But the market priced them as A-grade negatives.

3. Eastmoney's midday commentary offers one more attribution: "affected by concerns related to a major Nvidia AI supply agreement and by intensifying market competition" (Eastmoney⁶). This report could not trace that agreement to a primary source, so it is relayed only and not used as a basis for rating.

---

4. The Limit-Up Ladder / Highest Boards

4.1 The consecutive-limit-up ladder (20 names, markedly thicker than yesterday's 8)

4.2 Three key readings

1. The ladder thickened; it did not collapse. The pre-market assumed risk was "only 8 consecutive-limit-up names; if second and third boards fall rather than rise it points to a retreat". The actual result was the opposite: consecutive names went from 8 to 20. The pre-market indicator was pointed in the right direction but read to the wrong conclusion — because it assumed the ladder resonates with the mainline, whereas today the ladder decoupled from it.

2. No core leader. Among today's limit-ups the largest turnovers were 顺钠股份 (000533) at RMB 2.148 billion, 岩山科技 (002195) at RMB 1.571 billion and 东方新能 (002310) at RMB 1.064 billion. There was no core leader anywhere with "large turnover + high recognisability + fundamentals". More than half of the 61 names in the limit-up pool traded less than RMB 300 million. This shows today's limit-up structure was small-money guerrilla action, incapable of carrying a mainline.

3. Two extreme samples worth remembering:

   • 爱丽家居 (603221), 6 boards in 6 days, traded only RMB 13 million today with a 0.33% turnover ratio and a RMB 354 million bid queue — an extremely thin one-word board; no volume means no absorption, so once the seal breaks there is nobody to take it.
   • 顺钠股份 (000533), 4 boards in 4 days, broke its seal 12 times today with a 28.08% turnover ratio and RMB 2.148 billion of turnover — the most fiercely contested relay position, and a direct weather vane for tomorrow's sentiment.

4.3 Broken boards and limit-downs

• The 49 limit-downs are themselves "yesterday's mainline breaking": 生益科技 (600183), 沪电股份 (002463), 深南电路 (002916), 金安国纪 (002636), 兴森科技 (002436), 景旺电子 (603228), 中材科技 (002080), 鹏鼎控股 (002938), 广合科技 (001389), 江南新材 (603124), 华正新材 (603186), 光迅科技 (Accelink Technologies, 002281), 长飞光纤 (YOFC, 601869), 永鼎股份 (Yongding, 600105), 华工科技 (HGTech, 000988), 深科技 (000021), 德明利 (Demingli, 001309), 新洁能 (NCE Power, 605111), 金海通 (Jinhaitong, 603061), 云南锗业 (Yunnan Germanium, 002428), 博迁新材 (Boqian New Materials, 605376), 江海股份 (Jianghai Capacitor, 002484), 英维克 (Envicool, 002837), 杰瑞股份 (Jereh Group, 002353), 利通电子 (Litong Electronic, 603629), XD 东材科 (Sichuan EM Technology, 601208), 立新能源 (001258), 孚日股份 (002083).
• Names that did not hit limit-down but fell further (ChiNext / STAR Market 20CM): 新易盛 (Eoptolink, 300502) −17.13%, 中际旭创 (Innolight, 300308) −15.69%, 天孚通信 (T&S Communications, 300394) −13.21%, 生益电子 (688183) −13.09%, 胜宏科技 (300476) −11.93%, 长芯博创 (300548) −11.94%, 太辰光 (Taichen Optical, 300570) −11.34%, 华盛锂电 (688353) −10.88%, 南亚新材 (688519) −10.69%.
• Among the 20 broken seals, worth noting: 珍宝岛 (603567) broke its seal 10 times intraday and closed +6.83% (the 4-boards-in-8-days high name has entered extreme divergence), 贤丰控股 (002141) broke its seal and closed flat, 国新能源 (Guoxin Energy, 600617) broke its seal and closed −1.88% (the next-day pattern for a name in the top tier of net buying on yesterday's top-trader list).

---

5. Capital Flows

5.1 Northbound flows: no reliable data

Eastmoney's Stock Connect history interface returns null for the net-buy, fund-inflow and turnover fields for every trading day from 07-22 to 07-27. This report provides no northbound net inflow figure for today and uses no substitute figure passed off as northbound. Large-order flows are used below instead.

5.2 Large-order net inflow / outflow (Tencent quotes sector basis, unit: RMB 100 million)

Outflows (all concentrated in tech hardware)

Inflows (only two, both defensive)

Reading: only two primary industries in the whole market, banks and food and beverage, recorded significant large-order net inflows, totalling about RMB 2.4 billion, while the chip concept alone saw RMB 77.9 billion of outflows. This is a textbook "risk-averse reallocation", not "incremental money entering".

5.3 Top-trader disclosure list

Today's (07-28) list has not yet been published (the interface returns empty; it is usually disclosed after 18:00). Instead, the 07-27 list plus actual next-day (i.e. today's) performance is used for a genuine test of capital effectiveness:

Reading on seat preference (important, and counter-intuitive):

• What determined next-day P&L was not the seat type but the industry. Among the 7/27 names on the list, every one belonging to the semiconductor / new-materials chain (International Composite Materials −6.38%, Jianghai Capacitor −9.99% limit-down, Binhua Group −7.93%, Jiangsu Yoke −1.72%, Fenghua Advanced −2.37%) fell the next day; every one belonging to pharma / themes / low positions (Innovative Medical Management +9.99%, Zhenbao Island +6.83%, Longshen Rongfa +6.18%, Zhichun Technology +4.50%) rose the next day.
• "Institutional buying" offered no protection today: International Composite Materials had 3 institutions net-buying RMB 720 million, 46.5% of turnover, and still fell −6.38% the next day.
• This cross-check independently confirms the Section 1 conclusion: today's divergence is capital leaving at the industry level, not a stock-level game.

---

6. Outlook for Tomorrow

① Judgement on mainline continuation tomorrow

② Key focus tomorrow (branch + main-board stock + verification point)

The following are observation items, not trading instructions. Today's lesson is precisely that "producing a list itself creates pressure to act".

Two macro verification points tomorrow:

• The FOMC decision, 02:00 Beijing time on 7/30 (i.e. after tomorrow's close and before the following day's open). The pre-market report recorded CME pricing: 65.3% for no change, 34.7% for a 25bp hike, and about 82% for a September hike. The risk-reward of holding overnight tomorrow is asymmetric.
• The July Politburo meeting still had not convened as of today (this report re-checked and found no report of it convening on 7/28); it could land at any time, with two-way direction.

③ What to avoid tomorrow

1. The entire AI hardware chain (CPO / optical modules / memory / CCL / PCB) — a single day's large-order net outflows of RMB 77.9 billion from chips, RMB 53.5 billion from electronics and RMB 44.6 billion from CPO. A bounce the day after limit-down is not a buy point; it is a liquidity window.
2. High-level, low-volume lone names:
   • 爱丽家居 (603221), 6 boards in 6 days, traded only RMB 13 million today with a 0.33% turnover ratio — a volumeless one-word board, with no absorption once the seal breaks.
   • 艾艾精工 (603580), 9 boards in 14 days, 5 broken seals today.
   • 海南海药 (000566), 5 boards in 10 days, PE (TTM) −17.27 (loss-making).
3. 创新医疗 (002173) — 3 boards in 4 days; the company has filed that "AM5 has not yet generated revenue" and pre-announced an H1 loss of RMB 48–62 million (widening 322~446% YoY), with a volume ratio of only 0.64 today (a limit-up on shrinking volume). The fundamental falsification holds while the price runs the other way — precisely the most dangerous combination, because there is no fundamental support to catch it when it reverses.
4. 顺钠股份 (000533) — 12 broken seals today with a 28.08% turnover ratio, so the relay cost is extremely high.
5. 长鑫科技 (688825) — −4.08% today with turnover of RMB 44.428 billion and a 20.66% turnover ratio; on its second day of listing it is still the largest single drain on liquidity in the whole market.
6. The semiconductor / new-materials names on yesterday's top-trader list — 国际复材 (301526), 江海股份 (002484), 雅克科技 (002409) and 风华高科 (000636) all fell today, and institutional seats offered no protection.

④ Input notes for tomorrow morning's pre-market list

1. Promote the "overnight overseas read-across" from the news table to a weighting/veto variable in branch ordering. Today's evidence: the overnight US "sell chips, buy software" delivered 100% in A-shares the same day — the pre-market report wrote that sentence correctly but only put it in row 1 of the news table and in the risk column, neither downgrading the ordering of Branches 1 and 2 nor creating a new "software" branch. Tomorrow's pre-market report should enforce: when the overnight SOX / memory chain falls more than 2%, the corresponding A-share branch ordering is cut one level directly; when overnight software / internet leaders rise more than 2%, the corresponding A-share mapped branch must be searched and candidates given.
2. Use earnings pre-announcements only to "exclude", not to "rank". Tomorrow's pre-market report must tag every H1 pre-announcement released in July with "T+N days", and those with N>7 may not be the primary reason for a stock's rank. Today's evidence: four pre-announcements were released on 7/13–7/15 (T+13~15), and 4 of the corresponding stocks hit limit-down today.
3. A deep drawdown is a deduction only, never a plus, until stabilisation is confirmed. Delete the use of "the deepest oversold" as a positive rationale (today 金安国纪 002636 made the list on exactly that basis and hit limit-down). Add a hard constraint: any stock more than 30% off its 120-day high without a volume-backed stabilisation candle is capped at 3 pts on the "technical sentiment" dimension.
4. Allow "no positive list today" to be a legitimate output. Today's report identified every risk signal (all volume ratios <1, the overnight SOX −2%, a hawkish-leaning FOMC, only 8 consecutive limit-ups, every mainline in deep drawdown), yet the fixed structure of "the 5 most worth watching" still forced out 5 recommendations. It is recommended that a-share-premarket's SKILL.md explicitly permit this section to output "no candidates meet the conditions today; standing aside is suggested".
5. Add a standalone "sentiment ladder" subsection, ranked alongside the fundamentals list rather than subordinate to it. Today the "yesterday's consecutive limit-ups" index was +4.81% and the "yesterday's limit-ups" index +1.00%, while the institutionally crowded CPO / memory / PCB names all hit limit-down. The pre-market list currently covers only the latter and Passes the former across the board — which is why the Pass list got the direction wrong on 6 stocks (创新医疗 +10.00%, 南京熊猫 +9.99%, 海南海药 +10.00%, 珍宝岛 +6.83%, 哈三联 +3.51%, 东方精工 +1.34%). Distinguish the two conclusions clearly: "should not be bought" (a fundamental judgement, entirely correct today) versus "will fall" (a price judgement, mostly wrong today); the Pass list should assert only the former.
6. Three primary sources that must be verified tomorrow: ① the original trigger for the lithography move (today every outlet described it uniformly as "a foreign-media report" or "a rumour note fermenting", and not one gave a citable primary source); ② the original filing for 张江高科 (600895)'s 10.78% stake in Shanghai Micro Electronics Equipment; ③ the primary source for the "concerns over a major Nvidia AI supply agreement" (Eastmoney's midday attribution). All three exist today only as secondary relays, which is not enough to serve as a basis for rating tomorrow.
7. Fix the scrape timing for overnight US data. Today's pre-market record (SanDisk −8%, SK Hynix −5%, ASML −4%) differs significantly from the intraday media basis (−11%, −7.5%, −6%), suggesting the pre-market report used an intraday snapshot rather than closing prices. This was the direct trigger variable for today's fall, so an error in this basis carries the highest cost.

---

⚠️ Risk warning: this recap is only a post-close information review and observation and does not constitute investment advice. Data may differ in timeliness or basis; the exchange's and companies' disclosures prevail, and this must not be used directly as a basis for trading.

---

Data sources: Tencent quotes qt.gtimg.cn (index and stock closing prices, turnover, turnover ratio, volume ratio, PE), Eastmoney push2ex (the limit-up / limit-down / broken-seal pools), the Tencent quotes sector interface proxy.finance.qq.com (industry and concept sector moves, large-order net inflows), Eastmoney datacenter-web (the top-trader list, Stock Connect). News sources: CNR³, National Business Daily⁴, Eastmoney⁶, JRJ¹, Sina Finance², Eastmoney Caifuhao⁵. Reconciliation baseline: reports/a-share/2026-07-28.md.