A-Share Market Recap | 2026-08-26 Wednesday

Data read timestamp: 2026-08-26 15:30 (Beijing time). Quotes, the limit-up/limit-down pools, sectors and fund flows are all post-close snapshots. Coverage check: the full-market per-stock snapshot captured 5,904/5,904 names (bidirectional pagination by ascending + descending change, de-duplicated); the sum of per-stock turnover is RMB 1,821.8 billion vs the combined index turnover of the Shanghai/Shenzhen/Beijing exchanges of RMB 1,822.2 billion, a deviation of 0.02%, confirming there is no gap in the sample. The four pools — limit-up / limit-down / broken limit-up / consecutive limit-ups — are all taken from the official exchange-standard pools and have been validated with len(pool) == tc.

---

0. One-Sentence Recap

Sentiment warmed up moderately, but today's real leading theme was not the one identified pre-market. Shanghai Composite 3912.52 (+0.59%), CSI 300 +0.85%, STAR 50 +1.71%, 52 limit-ups and 0 limit-downs; the whole market opened flat and traded higher (average open −0.02%, average open→close +0.29%, 56% of stocks closed above their open) — this was a day of sustained intraday buying. The strongest theme was copper / industrial metals (copper sector +5.16%, industrial metals main-force net inflow +RMB 7.115 billion, first across the market), and the second was brokers / non-bank financials (securities +3.01%, net inflow +RMB 4.924 billion); pre-market these two ranked only 3rd and did not appear at all, respectively. The three pre-market recommendation slots averaged +4.89% and 3/3 beat the all-A median, with Jiangxi Copper (600362) limit-up +10.01% and WUS Printed Circuit (002463) +4.31%; the only bearish call, Luzhou Laojiao (000568) −2.37%, also played out. But the Pass list averaged +2.99% and 10 of 14 beat the all-A median — what most needs review today is the Pass list, not the recommendations. Tone for tomorrow: NVIDIA's earnings land at 05:00 Beijing time on 8/27, so tomorrow is the reaction day for the compute chain, and today the compute chain already cut exposure in advance via "flat all day + net fund outflow".

---

1. Market Overview

Turnover: Shanghai + Shenzhen came to RMB 1.81 trillion, down about 1.2% from RMB 1.83 trillion on 8/25; including the Beijing Stock Exchange it was RMB 1.82 trillion. The across-the-board index gains happened on shrinking volume — a structural rotation rather than new money coming in.

Advancers / decliners: 2,946 up / 2,448 down / 153 flat (5,547 names counted; another 357 were suspended or had no turnover today). Share of advancers 53.1%.

Follow-through quality of yesterday's limit-up names (computed in-house, the ZTPoolPrevious endpoint was not used): joining the 65 codes in the official 8/25 limit-up pool against today's full-market snapshot, all 65 were matched — 38 up / 26 down / 1 flat, 58% closed green, average +2.20%, median +1.21%, of which 13 hit the limit again today. Average opening gap +1.65%, average open→close +0.62% (they were still being bought after the gap-up, not sold).

For comparison: the pre-market report cited 8/24→8/25 follow-through of "only 43% of 46 limit-up names closed green, average +1.03%". Follow-through repairing from 43% to 58% is the hardest evidence for today's "warming" sentiment call.

Sentiment reading: warming, but the strength is in the indices and in low-level cyclicals, not in the theme ladder. Three supports: ① 0 limit-downs; ② 58% of yesterday's limit-up names closed green with a positive average open→close; ③ the whole market opened flat and traded higher. Two constraints: ① the highest streak is only 5 consecutive, and it belongs to the purely flow-driven Shenzhen China Bicycle A (000017); ② volume shrank — no new money arrived, and today's gains are existing money rotating out of themes into cyclicals and large financials.

---

2. Reconciliation of the Pre-Market List

This table gives two change columns: Chg% is based on the previous close, which is the benchmark an investor faced when the pre-market list was published (07:30, pre-market), and is what should be used to judge whether the call was right; Open→Close% is based on today's open and is the segment actually obtainable by executing the list. Where the two diverge is where the list's true quality shows. Today's benchmarks: all-A median +0.16%, average +0.26%; Shanghai Composite +0.59%. "Beat/lagged" below always refers to the all-A median of +0.16%.

2.1 Recommendation Slots (3 names)

Recommendation-slot summary: average +4.89%, 3/3 beat the all-A median, 0 names down. More worth recording: all three "verification points for today" were triggered or falsified exactly the way they were written pre-market — Jiangxi Copper's opening band, WUS's turnover rate and +6% cap, Hengtong's fund follow-through. The verification point carries more information than the conclusion itself.

2.2 Bearish Call (1 name)

This one has to be booked separately: "Luzhou Laojiao's interim report is an un-priced major negative" — right; "it will be read as an industry read-across, avoid consumer staples as a whole today" — wrong. The market treated −43.37% as a company-specific case. Pre-market risk signal ① "baijiu gaps down, trades lower and spreads to food & beverage as a whole" was not triggered today, even though pre-market listed it as "the one thing most in need of guarding against".

2.3 Pass List (14 names) — today's biggest miss

The Pass list averaged +2.99% with a median of +2.63%, and 10 of 14 beat the all-A median. As a "do not buy" list, it failed today.

2.4 Hit Rate and Review

Overall hit rate: of 18 directional calls (3 recommendations + 1 avoid + 14 Passes), 9 were right, 50%. Within that, recommendations and the avoid were 4/4, while Passes were 5/14.

Review: the Pass failure was not random — split the Pass rationales into two types and the outcomes separate completely

All 7 Type B names beat the median, without exception; Type A averaged negative.

• Type A stood up to scrutiny in full today, and did so on a broadly rising day: 百傲化学 (603360) −1.02% — its "agrochemicals" sector was +1.68% today and produced 2 limit-ups, yet it fell against the trend, validating the "classification label ≠ actual business" call in the strongest possible way (sector up, mismatched stock down); 欧科亿 (688308) −1.36%; 新华百货 (600785) −3.06%.
• Type B failed in full today. 白银有色 (601212) and 湖南白银 (002716) had only just validated the Pass on 8/25 (gap-up then fade, −1.79% / −6.02%), and today reversed straight into a limit-up and +5.11%; 深中华 A (000017) sealed its 5th consecutive limit-up on a day when the precious metals sector fell again (gold industry −1.62%, main-force net outflow RMB 1.216 bn).

Conclusions (to be carried into tomorrow morning's pre-market list):

1. "Extended position" is not a reason to Pass — it is a reason to size. Writing "+69% over 20 days", "seal ratio 0.011", "5 consecutive limit-ups", "top-buyers-list net selling" into the Pass column amounts to using chip structure to predict next-day direction, and chip structure has almost no predictive power over next-day direction. These two types should be split into separate columns: the Pass column should hold only "the logic does not hold", while position issues move to "half position / do not chase / only trade the disagreement".
2. This is a repeat error. A same-type failure was recorded once before — using single-day extremes in main-force net inflow to rank branches, which left the recommended group at 21.4% lagging the Pass group at 72.2%. Today is its mirror image: using single-day chip metrics (consecutive limit-up count / turnover rate / seal ratio) to rank for the next day, with the Pass group at +6.18% beating everything outside the recommendation slots.
3. "Pure flow behaviour" is an accurate description, but it is not an expiry date. The pre-market call on 深中华 A — "sector down while the leader keeps limiting up = pure flow behaviour" — was completely correct as a description and held again today (the gold industry was still −1.62% today); what was wrong was the implicit assumption that pure flow behaviour cannot continue. It did continue, all the way to the 5th consecutive limit-up.
4. The Open→Close column rescued two names: 英维克 (002837) closed +3.05%, which looks like a wrong Pass, but its intraday high was the very first second of trading (+4.68%) and it fell one-way afterwards, so executing at the open gives −1.55%; 汉森制药 (002412) closed +6.16% but broke the limit with open→close of −2.68%. Looking only at the close-to-close change systematically overstates the Pass failure rate, and also overstates the success rate of the recommendations.

---

3. Verification of Today's Leading Themes

3.1 Did pre-market identify the strongest theme — no, and the error was in the ranking, not in the stock picking

Pre-market ranked the three main branches: ① PCB/high-speed switches ② optical fibre & cable ③ copper. Today's actual ranking was: ① copper ② brokers ③ PCB, while optical fibre & cable was one of the largest outflow directions of the day.

But it should be booked fairly: the ranking was wrong, the names were not. The sole representative stock pre-market gave to copper, 江西铜业 (600362), limited up +10.01% today, the best performer in the entire list; the treatment given to optical communications was "strong branch, weak stocks — no recommendation slot, do not chase a gap-up", and when optical communications receded today that treatment protected the position. The branch-strength ranking error did not translate into losses because stock selection and entry rules worked independently of the branch ranking.

3.2 One thing that must be spelled out: copper stocks rose far more than copper itself

Copper only rose 0.71%, copper stocks rose 5.16%, and Jiangxi Copper rose 10.01%. This is not a commodity-following move, it is a re-pricing on the equity side. Breaking it down makes it clearer:

• Jiangxi Copper (600362) beat the copper sector by about 4.9pp and beat the largest mining stock, 紫金矿业 (601899), by 7.7pp. The pre-market criterion was "watch whether Zijin Mining moves the same way — if Zijin does not rise while Jiangxi Copper rises alone, what is being bought is the news, not copper". Zijin actually came in at +2.35% (same direction but clearly lagging), so both are true: copper's modest rise (+0.71%) supplied the direction, while the 4.9pp by which Jiangxi Copper exceeded the sector came from its own interim report.
• Pre-market repeatedly stressed that "Jiangxi Copper is not a copper-price elasticity stock, it is a processing business operating in a high copper-price environment". Today the group with heavier smelting characteristics — 江西铜业 (600362) +10.01%, 云南铜业 (000878) +5.50%, 铜陵有色 (000630) +4.98% — was collectively stronger than the purest mining name, 紫金矿业 (601899), at +2.35%. With TC/RC at historically extreme negative levels, this looks more like an expected repair in smelting and processing profit than mine-end elasticity to rising copper prices. If that reading holds, its dependence on the copper price is lower than the market assumes, and its dependence on treatment charges (TC/RC) is higher than the market assumes.

The same-type phenomenon appeared in precious metals, in the opposite direction: SHFE gold −0.14%, gold industry −1.62%, while 白银有色 (601212), 金一文化 (002721) and 深中华 A (000017) hit the limit. On the same day, copper was "metal up slightly, stocks up sharply (with financial-statement support)", and gold was "metal down, concept stocks limit-up (with no financial-statement support)". The former is trackable, the latter can only be avoided.

3.3 Leading Themes Missed Pre-Market

1. Brokers / non-bank financials (the most important one). Main-force net inflow of +RMB 5.785 bn (non-bank) / +RMB 4.924 bn (securities), 锦龙股份 (000712) and 湘财股份 (600095) limit-up, with the whole sector rising. Broker-related items did appear in the §1 news table pre-market (中金公司 (601995) approved to issue up to RMB 80 billion of corporate bonds; 招商证券 (600999) H1 net profit attributable to parent +104.87%), but both were downgraded as "already pre-announced, zero magnitude expectation gap" and never promoted to a branch. Today 招商证券 was +5.21%.

   This report could not confirm whether there was a specific new in-window catalyst for the broker sector today (policy/data/event); as of the 15:30 generation time there was no verifiable primary source. Here we state only the flow and price facts, and do not invent a reason for them.

2. Extreme divergence inside telecom equipment. The telecom equipment sector was −0.09% with net outflow of RMB 1.681 bn, yet at the same time it produced 3 limit-ups, all on huge volume: 剑桥科技 (Cambridge Industries Group, 603083) +10.00% on turnover of RMB 4.996 bn (first among all limit-up names), 共进股份 (Gongjin Electronics, 603118) +10.00% on RMB 2.172 bn, 楚天龙 (Chutian Dragon, 003040) +10.01% (4th consecutive) on RMB 1.770 bn. A sector in net outflow with leaders limiting up on huge volume is the shape of money concentrating from within the sector into a handful of names.

   剑桥科技 (603083) sits in the high-speed optical module / CPO direction; this report could not obtain a specific catalyst announcement for it today. The confirmable public background is its interim pre-announcement of net profit attributable to parent up 156.65%–197.18% (that pre-announcement is out-of-window old news and should not be booked as new today). An optical module leader limiting up on huge volume on the eve of NVIDIA's earnings, coexisting with net outflow across the compute chain, is a contradiction this report will not force into a single explanation.

3. STAR 50 +1.71% led the whole market, with semiconductor equipment +1.71% and net inflow of RMB 1.279 bn (中微公司 (AMEC, 688012) +3.97%, 长川科技 (Changchuan Technology, 300604) +4.77%; 北方华创 (Naura, 002371) gapped down −4.15% and closed flat at +0.04%, open→close +4.38%). Pre-market had only one item in this direction, 金辰股份 (Jinchen Machinery, 603396), judged "long-dated, watch only" — that stock limit-locked at the open today, +10.02% (zero intraday range, turnover of only RMB 115 million, an untradeable limit-locked board).

---

4. Limit-Up Ladder / Height Boards

Ladder structure: 1 name at 5 consecutive, 1 at 4, 1 at 3, 10 at 2, and 39 first boards. The ladder is complete but lacks height and has obvious gaps (only 1 name each at 5→4→3).

Seal quality on the height boards was generally very poor. The 5-consecutive and 4-consecutive names had seal order / turnover ratios of only 0.064 and 0.020 (empirically >1 is a genuinely locked limit-from-open board, <0.06 is a high-churn poor board), turnover rates above 23% for both, and 10 broken limits combined. These two were pushed up by repeated churn, not locked up.

Core names (recognisable advancers on large turnover):

Distribution of sealing times: of the 52 limit-ups, 37 sealed before 10:00 (6 of them at the 09:25 call auction), and only 3 sealed in the afternoon (联合精密 (United Precision, 001268) at 13:00, 立新能源 (001258) at 13:14, 茶花股份 (Chahua, 603615) at 13:18). Limit-ups happened mainly in the morning session; there was no fresh attack in the afternoon.

Broken limits (20 names):

• 汉森制药 (002412): failed to make a 5th consecutive limit-up, spiking to +9.08% at the open before breaking, closing +6.16%, turnover rate 37.1%, turnover RMB 2.432 bn. The pre-market Pass was based on "on the 8/25 limit-up day the top-buyers list showed net selling of RMB 102 million with institutions as net sellers", and today's broken limit delivered the core of that call — but it still closed up 6.16%.
• 中南文化 (Zhongnan Culture, 002445): after breaking the limit it turned red-to-green and closed −2.96% on turnover of RMB 1.533 bn, the only name in the broken-limit pool to close sharply lower.
• Most of the other broken-limit names still closed higher (国仪公司-U (Guoyi Quantum-U, 688828) +15.09%, 兆日科技 (Zhaori Technology, 300333) +16.75%, 太阳纸业 (Sun Paper, 002078) +8.79%), so "a broken limit means weakness" did not hold today, consistent with the whole market opening flat and trading higher.

---

5. Fund Flows

5.1 Northbound Flow — disclosure has been discontinued (a definitional note, not a data-fetch failure)

The Shanghai and Shenzhen exchanges stopped disclosing daily northbound net buying from 2024-08-19, replacing it with quarterly publication of shareholdings in individual securities. Any source claiming "northbound net inflow of RMB XX today" should not be trusted. This section therefore provides no northbound data. Source: Sina Finance¹

5.2 Top-Buyers List (Longhu Bang) — not included in this report

Top-buyers-list data is published after 18:00 on the trading day, and this report was generated at 15:30, so the 8/26 list is not included. This is a scheduling gap, not missing data: the same endpoint returns 8/25 data normally, with 65 records. Several conclusions in the pre-market list depend on the quality of the seats on that list (亨通光电 (600487) net buying of RMB 1.850 bn, 汉森制药 (002412) institutional net selling of RMB 102 mn, 通鼎互联 (002491) net selling of RMB 455 mn); the corresponding seat verification for today has to be checked after 18:00.

5.3 Main-Force Net Inflow / Outflow (industry basis)

Two key readings on a concept basis: the compute concept −RMB 2.056 bn (sector +0.02%), memory chips −RMB 1.669 bn (−0.15%), the CPO concept −RMB 624 mn, the consumer electronics concept −RMB 1.178 bn; while the PCB concept was +RMB 2.792 bn and the gold concept +RMB 4.744 bn.

Two definitional conflicts that must be spelled out: ① The "gold concept" saw net inflow of +RMB 4.744 bn, while the "gold" industry saw net outflow of RMB 1.216 bn and the "precious metals" industry net outflow of RMB 1.385 bn. This is not a contradiction, it is a difference in constituents: what flowed in were low-priced concept stocks such as 白银有色 (601212), 金一文化 (002721) and 深中华 A (000017), and what flowed out were genuine gold miners such as 山东黄金 (600547) and 山金国际 (000975). Looking only at "gold concept net inflow of RMB 4.7 bn" would produce exactly the opposite conclusion. ② The printed circuit board sector was −0.32% yet took in RMB 1.865 bn of net inflow, whereas yesterday it was "sector +1.32%, net inflow RMB 2.365 bn, and the leader WUS actually −1.11%". Today it is reversed: money kept coming in, the leader rose +4.31%, and the sector did not follow. The direction of net inflow was consistent across the two days; this time it just went into the leader.

Fund flows in one sentence: money moved from "telecom / compute / semiconductor materials / precious-metal mining" into "industrial metals (copper) / non-bank brokers / the PCB leader". The former is active de-risking on the eve of NVIDIA; the latter is catch-up buying in low-level cyclicals and large financials.

5.4 Data Integrity Note

Summing main-force net inflow stock by stock across the 58 constituents of the industrial metals sector gives +RMB 7.115 bn, exactly consistent with the sector-level figure, so there is no contamination of sector fund flow by a same-day new listing. (The only recent new listing today, C 高凯 (Gaokai, 688835), was +27.44% on turnover of RMB 2.529 bn and belongs to the "specialized equipment" sector; its main-force net inflow of +RMB 181 mn accounts for the entirety of that sector's +RMB 109 mn and is what turns it positive — the stock must be stripped out when citing "specialized equipment" fund flow, and this report does not use that sector's data.)

---

6. Outlook for Tomorrow (2026-08-27 Thursday)

① Judgement on the Continuation of Tomorrow's Leading Themes

Tomorrow is a day governed by one external event. NVIDIA's FY2027Q2 results are released at 05:00 Beijing time on 8/27 (call at 17:00 ET), and the A-share reaction day is tomorrow. Today the compute chain already completed its pre-event de-risking through "flat prices + net fund outflow" (compute concept −RMB 2.056 bn, memory chips −RMB 1.669 bn, communication network equipment & components −RMB 2.205 bn), which means tomorrow's swings in the compute chain will be amplified: low exposure = plenty of room to add on good news, but also = nobody was willing to take the other side today.

② Key Focus Tomorrow

③ Avoid Tomorrow

1. The high-level consecutive-limit-up ladder (highest risk of the tide going out): 深中华 A (000017) at 5 consecutive, 楚天龙 (003040) at 4. Their seal order / turnover ratios are only 0.064 and 0.020, they broke the limit 10 times combined, and both had turnover rates above 23%. This is the classic "sealed but not locked".
2. The entire precious-metals concept: 白银有色 (601212), 金一文化 (002721), 湖南白银 (002716). SHFE gold −0.14%, the gold industry −1.62% with net outflow of RMB 1.216 bn — the metal side offers no support at all; and this group has already completed one round of "limit-up → sharp drop → limit-up" oscillation across 8/24→8/25→8/26, leaving it at the upper edge of the oscillation range.
3. Chasing optical fibre & cable: 亨通光电 (600487), 长飞光纤 (601869), 中天科技 (600522). Money reversed by RMB 5.3 bn in a single day (+4.389 bn → −951 mn), so do not participate until there is fresh price-increase or centralised-procurement news. 杭电股份 (603618) limited up today but its four issues — low base + newly recognised impairment provisions + dilution from the private placement + share sales by directors and senior management — are unchanged.
4. Limit-locked boards and untradeable limit-ups: 金辰股份 (603396) limit-locked at +10.02% today with zero intraday range and turnover of only RMB 115 mn; 海鸥住工 (002084) at 3 consecutive with turnover of only RMB 16 mn. Statistically these count as hits, but in execution you cannot buy them, and they should not go into any list.
5. Any heavy position anywhere in the compute chain before the NVIDIA result lands — including today's huge-volume limit-ups 剑桥科技 (603083) and 共进股份 (603118). They were pushed up on the eve of the event against a backdrop of sector net outflow; the risk-reward is asymmetric.

④ Input Notes for Tomorrow Morning's Pre-Market List

1. Split the Pass column into two columns. In today's Pass list the 7 names whose "logic does not hold" averaged −0.21%, while the 7 whose "position is too extended" averaged +6.18% with 7/7 beating the median. From tomorrow: Pass carries only logic/fact failures (classification mismatch, direction read backwards, old news, landing at the lower bound of guidance); position and chip issues all move into "do not chase / only trade the disagreement / half position" and are no longer written as Pass. This is the most expensive lesson of the day.
2. Branch-strength ranking and stock selection must be done separately. Today the branch ranking was wrong (copper, ranked 3rd, was placed behind PCB at 1st and optical communications at 2nd), yet the recommendation slots went 3/3 beating the median, because stock selection and entry rules worked independently. Tomorrow morning, do not feel obliged to pick names from a branch just because it ranks 1st, and do not shrink a branch's position size just because it ranks 3rd.
3. Every recommendation must carry a falsifiable verification point, with the read timestamp written out. All three verification points on today's recommendation slots were triggered or falsified exactly as written (Jiangxi Copper's opening band at +1.83%, WUS's turnover rate of 3.1% and intraday high of +6.65%, Hengtong's money failing to take the other side). The verification point carries more information than the conclusion itself, and this has held for several days running.
4. Tomorrow morning must add a call on a "brokers / non-bank financials" branch, and go back and check whether a specific catalyst existed today (8/26) — this report could not verify it as of 15:30 today, which does not mean it does not exist. In addition, the top-buyers list available after 18:00 can be used to verify today's seat quality in copper and brokers.
5. Add a permanent "Open→Close" column to the reconciliation table. Today 英维克 (002837) closed +3.05% but was −1.55% open→close (the intraday high was the very first second of trading), and 汉森制药 (002412) closed +6.16% but was −2.68% open→close and broke the limit. Looking only at the close-to-close change overstates both the success rate of the recommendations and the failure rate of the Passes.

---

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

---

Data this report could not obtain (gaps declared):

• The 8/26 top-buyers list (published after 18:00; this report was generated at 15:30). Three calls in the pre-market list depend on seat quality — 亨通光电 (600487) net buying of RMB 1.850 bn, 汉森制药 (002412) institutional net selling of RMB 102 mn, 通鼎互联 (002491) net selling of RMB 455 mn — and the corresponding verification for today has to wait until after 18:00.
• Northbound flow: daily disclosure was institutionally discontinued from 2024-08-19; this is not a data-fetch problem on our side.
• A specific catalyst for the broker / non-bank financials sector today: this report can only confirm price and flow facts (securities +3.01%, net inflow RMB 4.924 bn), and could not verify whether there was a specific policy or event driver in the window.
• A specific catalyst announcement for 剑桥科技 (603083)'s limit-up today could not be obtained; its interim pre-announcement of +156.65%–197.18% is out-of-window old news and should not be booked as new today.
• LME copper and COMEX copper closing prices for today could not be obtained (the endpoint returned empty), so the copper price judgement rests solely on the SHFE copper front month CU2610 (RMB 108,750/tonne, up 0.71% from the 8/25 settlement of 107,980).
• 杭电股份 (603618): the specific amount in its 8/25 "Interim Provision for Asset Impairment" announcement is still unavailable (left blank pre-market, and not filled in today).

Internal Notes (not sent to clients)

Data-fetch failures and channels

• hq.sinajs.cn returns HTTP 403 on this machine across all Referer variants (it worked yesterday); the Sina quote channel is dead.
• push2.eastmoney.com returns HTTP 502 (same as yesterday); full-market / sector / K-line / intraday all go through push2delay.eastmoney.com, which is empirically not lagged (the modal f124 timestamp is 2026-08-26 15:34).
• The limit-up / limit-down / broken-limit / consecutive-limit pools go through push2ex.eastmoney.com; all four pools passed len(pool)==tc (52/0/20/168).
• Background processes get reclaimed by the Bash tool: a fetch started with nohup ... & dies immediately with exit 1 and empty stderr, the same type as the previously recorded "the Bash tool takes away Chrome started in the background". Wrapping it as setsid ./run.sh works, and switching per-page requests to an 8-thread pool cut full-market fetching from >2min to about 50s.
• A trap in the top-buyers-list endpoint: sortColumns=NET_BUY_AMT returns {"success":false,"message":"NET_BUY_AMT排序列不存在","code":9501}; the correct column name is BILLBOARD_NET_AMT. Back-testing with 8/25 confirmed the endpoint itself works (65 records returned), which pins "cannot fetch 8/26" as a scheduling issue (published at 18:00) rather than an endpoint failure — this is the second time the "the top-buyers list is only published after 18:00" lesson has paid off.

⚠️ WebSearch produced three mutually contradictory closing wraps today, all conflicting with the hard data

• One claimed "the Shanghai Composite fell 0.39% to 3868.38", another claimed "the Shanghai index rose 0.73% and the Shenzhen Component rose 1.00%", while the API actually shows Shanghai Composite 3912.52 / +0.59% and Shenzhen Component +0.69%.
• What exposed them was the previous-close cross-check: the Shanghai Composite previous close of 3889.44 returned by push2delay matches the 8/25 close recorded in the pre-market report exactly, and the sum of per-stock turnover differs from the combined three-exchange index turnover by only 0.02%. Both independent checks point to the API data being real.
• One of them also mixed in the 8/24 session (Shanghai index −0.59%, Shenzhen Component −2.13%) as recent data. This is the same family of failure as "recurring column headlines hide the publication date" and "a search summary can conjure an earnings report out of nothing": the summary is self-consistent, the numbers look plausible, and the dates are shifted.
• Handling: every number in this report comes from the API; search results were used only as corroboration for qualitative directions such as "large financials led", and not a single number from them was accepted.

Methodology notes (mapped to past lessons)

• The most valuable item today: the A/B split of the Pass list. The "logic does not hold" group of 7 averaged −0.21% (3/7 beat the median), while the "position too extended" group of 7 averaged +6.18% (7/7 beat the median). This is the second time "a single-day extreme in main-force fund flow is a contrarian indicator" has paid off, except this time the contrarian indicator is a chip metric (consecutive limit-up count / turnover rate / seal ratio / top-buyers-list net selling). Suggest adding a new memory: chip metrics can grade position size, they cannot characterise direction.
• "The verification point beats the headline conclusion" paid off for the third time today, and it was three-for-three: Jiangxi Copper's "flat to +2% and holding" (actual open +1.83% → limit-up), WUS's "3%–5% turnover rate is optimal, abandon above +6%" (turnover rate 3.1%, intraday high +6.65% then a fade), Hengtong's "watch whether the money keeps taking the other side" (it did not, closed +0.36%). The "headline conclusion" of downgrading Jiangxi Copper from 1st to 2nd pre-market was wrong (it limited up), but the verification point written for it in the same report was right.
• The correct A-share form of "reconciliation must add a column relative to the pre-market price" is the "Open→Close" column, which rescued two names today (英维克 +3.05% but −1.55% open→close, with the intraday high in the very first second of trading; 汉森制药 +6.16% but −2.68% open→close and a broken limit).
• ZTPoolPrevious is still useless; today's performance of yesterday's limit-up names was all computed in-house by "joining the 8/25 limit-up pool codes against today's snapshot". Cross-checking the 8/25 pool against the 7 names from yesterday's limit-ups that were called out pre-market (金健米业 / 汉森制药 / 深中华 A / 凯莱英 (Asymchem) / 百傲化学 / 金螳螂 (Gold Mantis) / 登海种业) gave 7/7 matches, confirming that what was fetched really is the 8/25 pool.
• "A same-day new listing contaminates sector fund flow" was actively checked today: summing net inflow stock by stock across the 58 industrial metals constituents = +RMB 7.115 bn = the sector-level figure, so no contamination; but the +RMB 181 mn from the recent listing C 高凯 (688835) is the entire source of the "specialized equipment" sector's +RMB 109 mn and is what flips it from negative to positive, so this report does not use that sector's data.
• "Cannot find news ≠ there is no news" was triggered twice today: the catalyst for the broker sector, and the reason for 剑桥科技's limit-up, neither of which could be verified. Both were written as "this report could not confirm", not as "no news driver, pure flow behaviour" — the latter would have been a self-consistent but possibly completely wrong conclusion.
• One positive case worth remembering: 百傲化学 (603360) was −1.02% today, while its "agrochemicals" sector was +1.68% and produced 2 limit-ups. The pre-market call that "the classification label ≠ the actual business" was validated in reverse against a rising sector — far stronger evidence than it falling along with a falling sector.