A-Share Market Recap | Thursday, 2026-08-27

Data read timestamp: all quotes, fund flows and limit-up/limit-down pools are a snapshot taken after the 2026-08-27 close (15:00); the report was generated at 15:30 Beijing time. Data coverage boundary of this issue (judge the strength of the conclusions accordingly): ① Dragon-Tiger List (LHB) data is not included — the exchanges disclose the LHB after 18:00, and this issue is generated at 15:30, so it cannot possibly be retrieved; §5 therefore has no seat analysis. This is a scheduling constraint, not a data gap; ② daily northbound net buying has not been disclosed since 2024-08-19; the 0.0 returned by the interface means "no longer disclosed", not "net inflow was zero", so this issue makes no judgment on northbound direction (see §5.1); ③ stock-level "main-force capital" was not obtained; the per-stock capital column in §2/§5 uses the Tonghuashun "net fund flow" definition instead (inflow − outflow, across all 5,209 listed stocks), which is a different definition from the Eastmoney "main-force net inflow" cited in the pre-market issue and cannot be compared across issues directly; ④ the denominator of the failed-seal rate uses "official limit-up pool 77 + official failed-seal pool 17", and excludes stocks that touched the limit intraday but were not included in the pool. Definition checks: limit-up pool tc = 77 and len(pool) = 77 agree; limit-down pool tc = 3 and len = 3 agree; both are mutually consistent with Tonghuashun's "genuine limit-ups 69 + ST limit-ups 1" and "genuine limit-downs 3". Total turnover of the two markets was cross-verified through three independent routes (Tencent index basis RMB 2,125.9 billion / Tonghuashun per-stock sum RMB 2,125.6 billion / Tonghuashun industry sum RMB 2,141.0 billion).

⚠️ Data-retrieval failure log (this section is not sent to clients):

• The Eastmoney clist interface returned 502/RemoteDisconnected all day today — the same failure as in this morning's pre-market issue, now lasting a full trading day; all three domains push2/82.push2/48.push2 failed. Affected capabilities: sector quotes, sector fund flows, full-market paginated snapshots, ulist.np batch per-stock fund flows (intermittently available this morning, completely unavailable today), and every akshare interface that depends on clist (stock_zt_pool_zbgc_em timed out outright after 2 minutes).
• Workarounds (verified to work, recommend making them permanent): ① index and single-stock quotes via Tencent qt.gtimg.cn (stable throughout, and its turnover reconciles with the other two routes); ② limit-up/limit-down/failed-seal pools via push2ex (a different cluster from clist, unaffected); ③ sector quotes + sector fund flows switched to Tonghuashun stock_board_industry_summary_ths(), which returns the change, turnover, net inflow and advance/decline counts for 90 industries in one call — every sector-capital conclusion in this issue rests on it; this route was not used in this morning's pre-market issue, which is why that issue had no sector-level capital data at all; ④ per-stock fund flows via stock_fund_flow_individual(symbol="即时"), covering 5,209 stocks; ⑤ advance/decline counts via stock_market_activity_legu().
• Two silent traps in push2ex, both of which we hit this time: ① the sort parameter must not be URL-encoded in advance — passing fbt%3Aasc returns tc=77 but an empty pool array (the number is right, the list is empty, with no error raised); passing fbt:asc works. ② the limit-down pool's dpt is not wz.dtzt; both wz.dtzt and wzdtzt returned tc=None, and you actually have to pass wz.ztzt; the same goes for the failed-seal pool — wz.zbpool and wz.zbgc both fail, only wz.ztzt works. The correct parameters for these three pools are now pinned in /tmp/pools2.py.
• getTopicZTPoolPrevious was again confirmed today to be a same-day trap: passing date=20260827 returns 77 stocks that are exactly identical to the same day's limit-up pool (intersection 77/77), not "yesterday's limit-up stocks' performance today". The promotion rate in this issue instead uses getTopicZTPool with date=20260826 to fetch yesterday's pool (52 stocks, matching the tc=52 recorded in the pre-market issue) and then intersects it with today's pool.
• One discrepancy booked: a financial media closing commentary claimed "more than 3,300 stocks advanced"; two independent local routes (Tonghuashun's 5,209 stock-by-stock data, and the legu aggregate) both give 3,191. This issue uses 3,191 and labels the source; the media figure was not adopted.

---

0. One-Sentence Recap

Sentiment is euphoric, but today was a session of "mild index, extreme structure" — combined turnover of the two markets expanded to RMB 2.13 trillion (up 17.4% from RMB 1.81 trillion yesterday), 77 limit-ups against only 3 limit-downs, an 18.1% failed-seal rate, and yet the full-market median gain was just +0.54%, with almost the entire advance concentrated in a single line, the AI hardware chain: the STAR 50 rose 3.77%, 3.3x the SSE Composite (+1.13%). This morning's pre-market issue ranked "memory/storage" first and "PCB/copper-clad laminate" second, and those two turned out to be exactly the strongest two lines of the whole session (memory group equal-weighted +6.25%, PCB group +5.29%), while the third-ranked "copper/industrial metals" was the weakest of the five branches (+0.80%) — the pre-market top two were right, the third was wrong. Of the 26 names on the pre-market master list, all 26 rose today, 24 beat the SSE Composite, equal-weighted +4.87%; more worth booking is that the three names moved out of the recommendation slots by the pre-publication quality check — WUS Printed Circuit (沪电股份, 002463), Zhongji Innolight (中际旭创, 300308) and HGTech (华工科技, 000988) — returned only +1.73% equal-weighted today and all landed in the bottom one-sixth of the 26, while the 4 that stayed in the recommendation slots returned +6.46% equal-weighted — that quality check produced 4.73 percentage points of discrimination on the very same day. Today's genuine incremental catalyst was not inside the pre-market issue's scan window: the National Bureau of Statistics released January–July industrial enterprise profits on 8/27, with profits in computers, communications and other electronic equipment manufacturing up 105.0% year on year — official data that landed intraday. Tone for tomorrow: the main line continues but enters an "earnings verification" window — Montage Technology (澜起科技, 688008) holds a board meeting tomorrow (8/28) to review its interim report, and the ex-non-recurring range in its 7/17 pre-announcement (+14.5%+32.9%) is far weaker than the attributable-net-profit range (+63.9%+81.2%); that is tomorrow's single largest event risk.

---

1. Market Overview

1.1 Indices and Turnover

Combined turnover of the two markets was RMB 2.13 trillion, roughly 17.4% larger than the RMB 1.81 trillion of 8/26.

This single line directly answers risk signal No. 5 set in §8.3 of this morning's pre-market issue. The pre-market text read: "If even with overnight news this good the market still cannot expand above RMB 1.9 trillion today, it means no incremental money came in and the compute rally is merely a rotation of existing capital." The actual RMB 2.13 trillion cleared that threshold by 12%, so the risk signal did not trigger — this is one of the few judgments today where the pre-market issue wrote down a quantitative threshold in advance and had it falsified/confirmed the same day, and it was right.

1.2 Limit-Ups/Downs and Sentiment Temperature

Sentiment temperature: euphoric, but a "narrow euphoria".

Three pieces of evidence point to the same conclusion: ① only 3 limit-downs and an 18.1% failed-seal rate is a textbook strong environment; ② but the full-market median is only +0.54% while the SSE Composite rose 1.13% and the STAR 50 rose 3.77% — half of all stocks underperformed the index by more than a factor of two, meaning the index gain was contributed by a handful of sectors, not a broad rally; ③ within the 77 limit-ups, the AI hardware chain accounts for 22 (28.6%), while the rest of the limit-ups are highly scattered across sectors (43 SW tier-3 industries with 1–5 names each), showing that apart from AI hardware, the remaining limit-ups lack any sector cohesion.

One definition that must be stated clearly: the official limit-up pool of 77 is the full-scope count including ST names and one-way boards; Tonghuashun simultaneously reports "genuine limit-ups 69, ST limit-ups 1", and the gap mainly comes from statistical differences over stocks that hit the limit, opened up again and did not re-seal. This issue uses the exchange-basis official pool figure of 77 throughout, and has verified tc == len(pool) == 77.

---

2. Pre-Market List Reconciliation

Reconciliation baseline and execution basis (the most important methodological premise in this section): The pre-market issue went out before 08:20, so the reader can buy no earlier than the 09:30 open. This table therefore gives two measures at once:

• Today's %chg = versus the 8/26 close; this is the "paper gain", and it includes the opening gap, a segment the reader cannot capture;
• Open→Close % = versus today's opening price; this is the real P&L of executing the list at the open on the day. Reconciling with the first column alone systematically overstates the hit rate. The delivery test uses a dual standard: today's %chg > SSE Composite (+1.13%) AND open→close % > 0; only names satisfying both are booked as "delivered".

2.1 The Five Key Names in §9.1

The five names overall: closing equal-weighted +5.53%, open→close equal-weighted +2.74%, 5/5 beat the SSE Composite. Of these, the 4 recommendation slots (1 priority deep-dive + 3 watch closely): closing equal-weighted +6.46%, open→close equal-weighted +3.74%, 4/4 beat the SSE Composite.

2.2 Full Reconciliation of the 26 Names on the §3 Master List

Master-list statistics: all 26 rose (26/26), 24 beat the SSE Composite (+1.13%), closing equal-weighted +4.87%.

2.3 Pass List Reconciliation

Group statistics:

• 6.1 "logic/facts do not hold", 7 names: closing equal-weighted +4.06%, 5/7 beat the SSE Composite.
• 6.2 "position too high / valuation too expensive", 9 names: closing equal-weighted +5.17%, 8/9 beat the SSE Composite.

2.4 Hit Rate and Self-Critique

One-line self-critique: the stock picking was right, the ranking tiers were wrong, and the avoidance calls were wrong across the board.

① What was right — ranking ability, and it came from the quality check. All 26 pre-market names rose, equal-weighted +4.87%, far above the full-market median of +0.54%, meaning the selection pool itself caught today's only main line. But what really deserves booking is the internal ranking: the 4 names that stayed in the recommendation slots returned +6.46% equal-weighted, while the three moved out of the slots by the pre-publication quality check — WUS Printed Circuit (002463) +2.35%, Zhongji Innolight (300308) +1.79%, HGTech (000988) +1.06% — returned only +1.73% equal-weighted, and were respectively the 6th-weakest, 4th-weakest and 2nd-weakest of the 26, all in the bottom one-sixth. The spread between the two groups was 4.73 percentage points. That quality check was not a formality; it cashed out into discrimination the same day; WUS Printed Circuit was moreover the only net outflow on the whole list (−RMB 0.183 bn) with an open→close of essentially zero (+0.02%), exactly matching the quality check's characterization of "a day-after reaction to the interim report, not a pre-positioned setup". Next time the pre-market issue should write "removed by quality check" into the list as an independent negative signal, rather than simply removing the name and leaving it at that.

② What was wrong — treating "position/valuation/timeliness" as reasons to avoid. The 9 names in the 6.2 "do not chase a high open" bucket returned +5.17% equal-weighted with 8/9 beating the SSE Composite, stronger than every group other than the recommendation slots. Among them Shengyi Technology (600183) was waved off on the timeliness argument that "the interim report was disclosed on 8/14 and the information has been priced for 13 days", and it rose +8.43% and touched the limit intraday, the 4th-strongest name on the whole list. This is consistent with the observation direction from the 8/26 recap (the high-position group outperformed); today is the second independent confirmation, so n has gone from 1 day to 2, and both point to the same conclusion: chip position and valuation percentile determine position size, not direction. On a main-line breakout day, "position too high" is precisely a synonym for "the money is already inside". Next time the pre-market issue should change the 6.2 bucket from "do not chase" to "halve the position but do not exclude", and stop using "the news has been priced for N days" as a reason to avoid — in a trending market, the pricing of old news is continuous, not one-off.

③ The bucket where avoidance was wrong across the board — at which layer was it wrong? The direction was wrong, but the quality judgment was not. Two stocks in the 6.1 bucket that had been directly falsified by company announcements went limit-up the same day: Kangsheng (002418) +10.09% to a 3rd consecutive limit-up, and Chutian Dragon (003040) +9.98% to a 5th. On the surface this is the most serious misjudgment of all — the companies themselves issued documents falsifying the story, yet the shares sealed the limit. But pull out the capital flows and the seal quality, and the pre-market judgment was only half wrong:

All three high-rung limit-up names had net capital outflows, and their seal quality was collectively below the 0.15 genuine-seal threshold set pre-market (0.022–0.075); Chutian Dragon broke its seal 19 times over the day before barely holding it. In other words: the pre-market judgment that "these three have no fundamentals and their seals are not genuine" was accurate; what was wrong was equating "poor quality" directly with "it will fall today". In a phase of euphoric sentiment, the decline of low-quality high-rung names is lagged, not synchronous. Next time the correct way to write up such names is not "Pass · avoid" but "do not participate, but do not predict a same-day decline" — separate the quality judgment from the direction forecast.

④ A structural gap that must be admitted: one of today's strongest main lines was entirely outside the pre-market field of view. See §3.2.

---

3. Verifying Today's Main Line

3.1 Measured Ranking of the Five Pre-Market Branches

Benchmarks: SSE Composite +1.13% / SZSE Component +1.50% / ChiNext +1.71% / STAR 50 +3.77% / full-market median +0.54%

Did the pre-market issue call the strongest main line? It got the top two right and the third wrong. The "three strongest branches today" given in pre-market §9.2 were [memory → PCB → copper]; the measured order was [memory → PCB → power/liquid cooling]. The order of the top two was exactly right, and the equal-weighted gains of those two (+6.25%, +5.29%) were indeed markedly higher than the other three — the pre-market line "today is not a question of whether the compute chain rises, but of which segment rises" holds, and it pointed at the right segment.

The error was the third name, and it was an informative error. Pre-market put copper third on the grounds that "the LME squeeze continues and the supply-side story is real"; yet it had already written down the correct counter-evidence itself: "at the 07:19 read, COMEX copper was up just +0.21%, while Jiangxi Copper had already gone limit-up in the A-share market on 8/26 — A-shares moved ahead of the metal, so today is a hand-off, not an ignition." That counter-evidence was right, but the conclusion did not follow it — it was filed under the risk section instead of being used to strike copper from the third slot. Today the copper group returned +0.80% equal-weighted, the only one of the five branches to underperform the SSE Composite, with Jiangxi Copper (600362) and Zijin Mining (601899) both opening lower (−1.89%, −1.94%). The improvement for next time is very concrete: when "the branch's overseas driving price barely moved on the day" and "the corresponding A-shares already went limit-up yesterday" hold simultaneously, that branch should be downgraded outright, not listed as the third-strongest with a footnote in the risk section.

One branch that was wrongly killed: power supply/liquid cooling/thermal management. Pre-market cut it from A− to B and wrote it into item 3 of "directions not recommended to chase today", on three qualifiers — NVIDIA was talking about the US domestic market, A-share companies have no order evidence, and the first commissioning node is FY2029. All three qualifiers hold at the factual level, and nothing today overturned them, yet the group still returned +4.09% equal-weighted and ranked 3rd, with Envicool (002837) +6.28% and Shenling Environmental (301018) +5.15%. This shows the pre-market error was not factual but a classification error — "mistaking the hardness of the long-term logic for the trading strength of the day" — on a main-line breakout day, adjacent branches with weak logic still eat the beta. Booked: logic hardness should determine holding duration, not whether to participate on the day.

3.2 A Main Line the Pre-Market Issue Missed Entirely: Agriculture/Seeds

The 5th-best-performing sector today was "crop farming and forestry (+4.91%)", and the agriculture chain accounts for 9 of the market's 77 limit-ups (11.7%), second only to the AI hardware chain's 22 — while this morning's pre-market issue gave this line zero coverage.

Why it was missed: this is a structural blind spot of the "scan window" mechanism, not a one-off oversight. The original catalyst for this theme was a J.P. Morgan report on the global food crisis, published on 8/16–8/17, far outside the scan window set by the pre-market issue (8/26 15:00 → 8/27 08:20). The sector has been fermenting continuously since 8/18 — 15 limit-ups in a single day on 8/18; by 8/25 Golden Health Rice had 5 limit-ups in 7 days and Denghai Seeds 3 in 6. Today is the 8th trading day of this line being priced, and it still contributes 11.7% of the market's limit-ups, yet the whole line was discarded because "the catalyst did not fall within the past 17 hours".

This is a recurrence of a known failure mode. The fix is not to lengthen the window (that would import a great deal of stale-news noise) but to add a second dimension orthogonal to the window — "pricing progress": before each open, first take yesterday's limit-up pool and aggregate it by sector, and whenever a sector produces ≥3 limit-ups on ≥3 consecutive days, it is forced into the day's candidate pool regardless of whether its original catalyst falls inside or outside the window, and its stage (ignition/fermentation/climax/ebb) is assessed separately. Today's agriculture chain fully satisfies that condition (fermenting since 8/18, 9 limit-ups today, highest rung 3), and the rule would have prevented the miss had it been in force. This has been written into §6④ and handed to tomorrow morning's pre-market issue.

3.3 A Catalyst the Pre-Market Issue Could Not Have Had, but Which Explains Today's Strength

The National Bureau of Statistics released "Profits of National Industrial Enterprises above Designated Size, January–July 2026" today (8/27): nationwide profits of industrial enterprises above designated size rose 17.6% year on year; within that, profits in computers, communications and other electronic equipment manufacturing rose 105.0% year on year (media rendered this as "1.1x").

This is official data that landed intraday; the pre-market issue could not possibly have obtained it when it went to press at 08:20, so it does not count against the pre-market score. It provides the AI hardware chain's strength today with a domestic fundamental support independent of NVIDIA's earnings: the doubling of electronics-industry profits is measured on the full "industrial enterprises above designated size" basis, not the listed-company basis, proving the boom has already spread to the non-listed middle of the supply chain.

One citation basis that must be clarified: several media closing commentaries described the data as being "led by compute chips and memory chips". On checking the NBS release of 8/27, the words "compute chips" and "memory chips" do not appear anywhere in that document; it gives +105.0% only for the broad industry category "computers, communications and other electronic equipment manufacturing". The "compute/memory chip" attribution comes from an earlier (January–May) statistician's commentary article and has been carried over by the media to today's data. This issue uses the verifiable +105.0% industry-level figure and does not adopt the chip-segment attribution.

---

4. Limit-Up Echelon and High-Rung Names

4.1 Full View of the Consecutive-Limit-Up Echelon (2 or more, 16 names)

How to read it: the echelon is complete but the top is extremely fragile.

• The height moved up one notch: the top rung rose from 5 yesterday to 6 (China Bicycle A, 000017), and the echelon runs 6-5-4-3-3-3-3-3-2×8 with no gaps — structural completeness is evidence that sentiment is not yet exhausted.
• But the seal quality of the top three names collectively fails. The genuine-seal threshold set in the pre-market issue is a seal/turnover ratio >0.15: the 6-rung China Bicycle A is only 0.075, the 5-rung Chutian Dragon 0.022 with 19 broken seals over the day, and the 3-rung Kangsheng 0.070. All three high-rung names show the combination of "low ratio + high turnover (13.5%–22.5%) + net capital outflow", the classic signature of relay money trading against itself, in sharp contrast with genuine seals: the 4-rung Seagull Sanitary Ware (002084) has a ratio of 11.91, turnover of only 0.8%, and zero broken seals.
• One definition trap that must be called out: Seagull Sanitary Ware (002084), Xinnong Chemical (002942), Jierong Technology (002855) and Jiheng Pharmaceutical (002742) show seal/turnover ratios as high as 2.98–11.91, which looks like a "super-genuine seal", but their turnover is only RMB 19–56 million — the denominator of the ratio is turnover, and the extreme shrinkage of turnover on a one-way board mechanically inflates the ratio. These are "shrunken-volume one-way boards", not "deep-pocketed capital". When judging seal quality, the ratio must always be read together with the absolute turnover; reading the ratio alone conflates one-way boards with genuine large-capital seals.

4.2 Promotion Rate and Broken Rungs

• Of the 52 limit-ups yesterday (8/26), 16 went limit-up again today, a promotion rate of 30.8%.
• 36 broke their rung (69.2%), among which the closely watched Cambridge Industries (剑桥科技, 603083, Shanghai main board) closed +3.86% today after yesterday's limit-up, breaking its rung, consistent with the pre-market characterization of "stale news recycled repeatedly, extended position + weak seal orders".
• 17 failed their seals, a failed-seal rate of 18.1%, and the most watched of these was Shengyi Technology (600183, Shanghai main board) — it sealed for the first time at 11:27, then opened up again and closed +8.43%, the only main-line name in today's AI-PCB chain to attack the limit; Zhaojin Gold (招金黄金, 000506) broke its seal 17 times over the day and closed +9.25%, the most repetitive single name of the session.

4.3 Identifying the Generals

By the three criteria of "large turnover + significant gain + sector representativeness", today's generals of the AI hardware chain are:

Zhongji Innolight (300308) needs a separate note: it tops the entire market with RMB 22.78 bn of turnover, 1.2x the second-placed name, yet gained only +1.79% with an open→close of −1.24%. Enormous volume + a stalled price + a high open fading lower is the classic pattern of divergence-driven churn — it is not that nobody is buying; it is that buyers and sellers are exchanging hands on a large scale at a high level. It neither voided itself nor confirmed itself today; it pushed the decision to tomorrow.

---

5. Fund Flows

5.1 Northbound Capital: No Longer Disclosed, No Judgment in This Issue

Daily net buying via the Shanghai and Shenzhen Stock Connects has not been disclosed publicly since 2024-08-19. Today the interface returned "net turnover 0.000000, net capital inflow 0.0, trading status 3"; the meaning of that 0 is "this field is no longer provided", not "foreign investors bought net zero today" — any conclusion drawn from it that "northbound is neutral / on the sidelines" is a misreading. This issue makes no judgment whatsoever on northbound direction.

The "420.0" returned by the same interface is the daily quota ceiling for southbound trading (RMB 42.0 billion), not the day's net inflow, and cannot be cited as evidence of fund flows either.

What remains valid and available in the interface today is the advance/decline count of Stock Connect eligible names:

The Shenzhen Connect advance-decline ratio of 2.76:1 is markedly stronger than the Shanghai Connect's 1.77:1, consistent with the day's structure of "growth over value, Shenzhen over Shanghai".

5.2 Sector Fund Flows (Tonghuashun basis, 90 industries)

All 90 industries together saw a net inflow of RMB 77.6 billion; capital was in net inflow overall.

Top 12 by net inflow:

Top 8 by net outflow:

Three readings:

1. Capital concentration is extreme. Semiconductors alone absorbed RMB 25.193 bn of net inflow, 32.5% of the market-wide net inflow total (RMB 77.6 bn); semiconductors + communications equipment + components together took RMB 44.889 bn, or 57.8%. This is not the capital distribution of a broad rally; it is single-main-line siphoning.
2. Advance/decline counts corroborate the "wholeness" of the main line. Components 63 up, 0 down, electronic chemicals 43 up, 0 down, semiconductors 184 up, 3 down — these three lines were essentially "zero decliners", a full-roster advance, in extreme contrast with the market-wide 3,191 advancers / 1,821 decliners (only 63.7% advancing). This is the mark of no internal disagreement within the main line, and the strongest evidence that it can continue tomorrow.
3. Capital rotated out of dividend and new-energy names into AI hardware. Banks saw RMB 5.392 bn of net outflow and fell 1.05% against the market, the only industry down more than 1% today; power, grid equipment, batteries and photovoltaics together saw RMB 7.331 bn of net outflow. The pre-market issue listed power supply/electrical equipment as a beneficiary branch, but today "grid equipment −0.01%, power 0.00%, combined net outflow RMB 3.559 bn" — A-share electrical equipment was not driven by NVIDIA's "power bottleneck" narrative; today's gainers Envicool (002837) and Shenling Environmental (301018) belong to the "thermal management/liquid cooling" sub-segment, which is a different line from grid equipment. This calls for a correction to the pre-market branch definitions: "power supply/liquid cooling/electrical equipment" should not be scored as a single merged branch — today the internal divergence within it exceeded 6 percentage points.

5.3 Dragon-Tiger List: Not Included in This Issue

Exchange Dragon-Tiger List data is disclosed after 18:00 each day, and this report is generated at 15:30, so it cannot be obtained at that point in time; §5 therefore has no seat-preference analysis. This is a scheduling constraint, not a data gap or an interface failure.

Impact on tomorrow's pre-market issue: among today's 77 limit-ups, the 3 high-rung names (China Bicycle A 000017, Chutian Dragon 003040, Kangsheng 002418) all showed net capital outflow plus weak seals, and the composition of their Dragon-Tiger seats (hot-money relay vs institutional distribution) is the key evidence for judging whether the high rungs can continue tomorrow — but that evidence does not yet exist as this issue goes to press. Recommendation: pull the 8/27 Dragon-Tiger List once again before tomorrow's open, and specifically check whether the top-five buying seats of these three are well-known hot-money proprietary seats.

---

6. Outlook for Tomorrow

6.1 Judgment on the Main Line's Continuation Tomorrow

Core conclusion: the AI hardware chain most likely continues tomorrow, but shifts from "everything rises together" into "divergence under earnings verification", with the trigger for that divergence being the interim reports landing en masse tomorrow.

Four pieces of evidence supporting continuation:

1. Zero disagreement inside the main line — components 63 up 0 down, electronic chemicals 43 up 0 down, semiconductors 184 up 3 down, with no sell-off of high-flying names inside any of these sectors;
2. Driven by capital rather than sentiment — semiconductors alone drew RMB 25.193 bn of net inflow (32.5% of the whole market), and the leaders (CXMT +RMB 3.230 bn, Hengtong Optic-Electric +RMB 2.519 bn, Montage Technology +RMB 2.095 bn) all had large net inflows, in inverse contrast with the "limit-up + net outflow" of the high-rung theme stocks;
3. Turnover confirms — the two markets' RMB 2.13 trillion is a 17.4% expansion; incremental money is entering, this is not a closed-pool game;
4. Independent domestic fundamental confirmation — the NBS January–July electronics-industry profit figure of +105.0% provides a support that does not depend on NVIDIA.

Two constraints on continuation:

1. Main-line names generally already carry single-day gains of 4%–10% today, and Montage Technology (688008), Hengtong Optic-Electric (600487) and others have entered short-term overbought territory; the size of tomorrow's opening gap will determine the risk/reward;
2. 8/31 is the statutory deadline for interim-report disclosure, so tomorrow through Monday is the densest earnings-landing window of the year — today's "everything rises together" contains a great deal of expectation not yet verified by earnings, and this week has already given us the cautionary case of Zhongji Innolight (300308), "excellent results but sold anyway".

6.2 Key Focus for Tomorrow

6.3 What to Avoid Tomorrow

1. The three high-rung relay names — China Bicycle A (000017, Shenzhen main board, 6 consecutive), Chutian Dragon (003040, Shenzhen main board, 5 consecutive), Kangsheng (002418, Shenzhen main board, 3 consecutive). The reason is not "the position is high" (today proved a high position does not mean a same-day decline) but that three hard pieces of evidence hold simultaneously: ① all three had net capital outflows (−RMB 0.174 bn / −RMB 0.481 bn / −RMB 0.159 bn); ② seal/turnover ratios of 0.075 / 0.022 / 0.070, all below the 0.15 genuine-seal threshold; ③ turnover rates of 13.5% / 20.8% / 22.5%, all in the extreme-turnover zone. Chutian Dragon only held its seal after breaking it 19 times over the day, and the company itself announced on 8/26 that digital-currency revenue is below 5% of the total. The signature of such names is "they may not fall on the day, but once the rung breaks they fall limit-down repeatedly" — an extremely poor risk/reward.
2. Copper/industrial metals — Jiangxi Copper (600362, Shanghai main board), Zijin Mining (601899, Shanghai main board). Equal-weighted +0.80% today, last of the five branches, and both opened lower (−1.89%, −1.94%). Pre-market ranking it the third-strongest branch has been falsified; this branch should drop out of the main-line sequence, downgraded to "reassess only if copper prices post a fresh large overnight gain".
3. Banks and dividend sectors. Banks fell 1.05% today (the only industry down more than 1%) with RMB 5.392 bn of net outflow, and under a main-line siphoning regime they are the market's cash machine; if the AI chain keeps strengthening tomorrow, the sector will most likely keep bleeding.
4. Zhongji Innolight (300308, ChiNext ±20%) — downgraded to "wait for the divergence to resolve". RMB 22.78 bn of record turnover today for a gain of only 1.79% and an open→close of −1.24%, the only huge-volume laggard inside the main line. It can neither be chased (a high open fading lower has now happened twice in a row) nor declared to be ebbing (net fund flow is still positive at RMB 0.342 bn). The correct handling tomorrow is to observe rather than participate: if it stalls again on huge volume, divergence within the optical module line is established; if it breaks above today's high on expanding volume, the disagreement has been digested.
5. Shrunken-volume one-way boards — Seagull Sanitary Ware (002084), Jierong Technology (002855), Jiheng Pharmaceutical (002742), Xinnong Chemical (002942). Seal/turnover ratios of 2.98–11.91 look extremely strong, but turnover is only RMB 19–56 million, which is a shrunken-volume one-way board rather than a genuine capital seal, and there will be no bid to absorb supply once the board opens.

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

1. [Top priority] Add a "pricing progress" dimension to repair the scan-window blind spot. The root cause of missing the agriculture chain today (9 limit-ups, 5th-best sector) is that the catalyst fell outside the window. Tomorrow morning's pre-market issue must first execute: take the 8/27 limit-up pool and aggregate it by sector; whenever a sector produces ≥3 limit-ups on ≥3 consecutive days, force it into the candidate pool regardless of the original catalyst date and judge its stage separately. Had that rule been in force today, the agriculture chain would not have been missed.
2. Stop using "position too high / valuation too expensive / news already priced for N days" as reasons to avoid. Today's 9 names in the 6.2 bucket returned +5.17% equal-weighted with 8/9 beating the SSE Composite, the strongest of all groups other than the recommendation slots; this is now the second independent confirmation after 8/26. From tomorrow the 6.2 bucket should read "reduce the position but keep participating", and "chip position" should be explicitly labelled a position-sizing parameter, not a direction parameter.
3. Promote "removed by quality check" into an independent negative signal. The 3 names the quality check removed from the recommendation slots today returned +1.73% equal-weighted while the 4 that stayed returned +6.46%, a spread of 4.73pct, and all 3 removed names landed in the bottom one-sixth of the 26. Tomorrow morning's pre-market issue should keep that mechanism and explicitly flag "downgraded by quality check" names in the list, rather than merely dropping them from it.
4. Split the composite branch "power supply/liquid cooling/electrical equipment". Internal divergence within that branch exceeded 6 percentage points today: thermal management/liquid cooling (Envicool +6.28%, Shenling Environmental +5.15%) rose markedly, while grid equipment (−0.01%) and power (0.00%) barely moved and together saw RMB 3.559 bn of net outflow. From tomorrow morning it should be split into two independent branches, "data-centre thermal management/liquid cooling" and "grid/electrical equipment", scored separately.
5. Pull the 8/27 Dragon-Tiger List (disclosed after 18:00), focusing on the buy/sell seat composition of the three high-rung names China Bicycle A (000017), Chutian Dragon (003040) and Kangsheng (002418), in order to judge the nature of the capital relaying the high rungs.
6. Earnings-window warning: 8/31 is the statutory deadline for interim-report disclosure. Tomorrow (8/28) through Monday is the densest earnings-landing window of the year, and the pre-market issue must explicitly flag for every recommended name "whether the interim report is already disclosed / the expected disclosure date", marking those not yet disclosed as two-way variables. In particular Montage Technology (688008) has a board meeting tomorrow to review its interim report, and its ex-non-recurring basis must be checked in advance rather than only the attributable-net-profit basis.
7. Definition hand-off reminder: the per-stock capital column in this issue uses the Tonghuashun "net fund flow" definition (because the Eastmoney clist/ulist interfaces failed all day), which differs from the Eastmoney "main-force net inflow" definition used in the pre-market issue; the capital figures in the two issues cannot be directly subtracted or compared. If the Eastmoney interfaces recover tomorrow morning, the pre-market issue should explicitly note that the definition has been switched back.

---

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