A-Share Market Recap | 2026-08-28 Friday

Data read timestamp: index/single-stock quotes are a post-close snapshot taken 2026-08-28 15:00 (Tencent qt.gtimg.cn, each stock carrying its own trade timestamp, verified as the post-15:34 settlement basis); the limit-up / limit-down / failed-limit pools are the East Money push2ex official pools on the 2026-08-28 basis, with tc == len(pool) verified (limit-up 82/82, limit-down 1/1, failed-limit 16/16); single-stock and sector net fund flows are on the THS (Tonghuashun) basis, and are an intraday snapshot taken before the close (see the next item); the interim-report disclosure calendar is on the CNINFO scheduled-disclosure basis.

⚠️ The single most important methodological caveat in this report — read it first: every "net fund flow" figure in this report comes from the THS "real-time" fund flow, a snapshot whose whole-market turnover totals RMB 1,944.7 billion, while actual turnover across the two exchanges that day was RMB 2,101.7 billion — 92.5% coverage, a gap of RMB 157.0 billion, and what is missing is the tail-end session and the closing call auction. There are two pieces of evidence: ① 星网锐捷 (Star-net Ruijie Networks, 002396) shows turnover of RMB 3.039 billion in that snapshot, versus RMB 4.141 billion actually at the close — the missing RMB 1.102 billion is 26.6% of the full-day turnover; ② 誉衡药业 (Yuheng Pharmaceutical, 002437) shows +0.68% in the snapshot, while it closed limit-up. Therefore every net fund flow figure cited in this report is a "reading taken some minutes before the close", not a full-day settled value; it will be systematically distorted for stocks that changed direction into the close. This report labels "intraday snapshot" at every point of citation; please do not directly subtract or compare it against full-day fund flows from other sources. The pre-market report (8/28) cited "yesterday's net fund flow" from the same interface, so the two reports are comparable with each other, but neither is comparable with the full-day basis.

Today's dragon-tiger list (LHB) was not obtained for this report and is left blank: the exchanges publish the dragon-tiger list only after 18:00, while this report goes out at 15:30. §5 therefore has no substantive dragon-tiger content; this is a structural gap determined by publication timing, not an absence of a dragon-tiger list today. The related seat data will be added in the next report (Monday pre-market).

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

• East Money push2/clist returns 502 Bad Gateway, now the third consecutive full trading day (four live tests: 8/27 pre-market, 8/27 post-close, 8/28 pre-market, 8/28 post-close). Capabilities affected: sector quotes, sector fund flows, whole-market paginated snapshots, ulist.np batch fund flows.
• New failure: East Money push2his single-stock fund flow (ak.stock_individual_fund_flow) returned RemoteDisconnected this time, with all four test tickers failing. This cluster had not yet been listed as down at the time of the pre-market report, which shows the East Money fund-flow lines have degraded from a single point of failure at clist to multiple clusters. There is currently no usable East Money fund-flow channel.
• This report's fund data therefore rests on a single THS line (ak.stock_fund_flow_individual(symbol="即时") + ak.stock_board_industry_summary_ths()), with no second independent basis for cross-validation. This is the root reason the fund-flow conclusions in this report carry lower confidence than the price conclusions, as flagged in the body.
• The THS "real-time" fund flow has been confirmed by live test to be an intraday snapshot: two pulls at 15:35 and 15:47 returned exactly identical values (frozen), and whole-market turnover totals only 92.5% of the actual. Recommend hard-coding this into the scripts: this interface must not be used as a full-day settled fund flow, and its snapshot nature must be labelled whenever cited. Also measured: its "inflow + outflow ≈ turnover" (median 0.9864), i.e. a single-classification basis, so net / turnover ∈ [−1,1] and can be used as a normalized indicator.
• Dragon-tiger list ak.stock_lhb_detail_em(20260828) returned TypeError: 'NoneType' object is not subscriptable — this is the scheduling problem of publishing at 15:30, earlier than the 18:00 disclosure, not an interface failure, and is the same pit hit repeatedly in the past. Do not raise an alert over this.
• ⚠️ WebSearch returned an "8/28 closing wrap" that contradicts today's facts; it has been discarded. That summary claimed "SSE Composite +0.07%, SZSE Component +0.56%, ChiNext +1.26%, 中芯国际 (SMIC) up over 10% to an all-time high, CPO and semiconductors leading", the exact opposite of the timestamped primary quotes fetched locally (actual: SSE Composite −0.11%, SZSE Component −0.68%, ChiNext −1.41%, 中芯国际 (688981) −0.36%, semiconductor sector −1.61% with a net outflow of RMB 16.470 billion). The search results also contained an item about "over 4,300 stocks rising on 17 August 2026", leading to the judgement that the summary stitched a closing wrap from a trading day around 8/17 into today's content. All price conclusions in this report follow the timestamped primary interfaces; no price figure from any search summary was used.
• The THS concept-sector interface stock_board_concept_summary_ths() returns "driving event / leading stock / constituent count" and contains no percentage-change column, so it cannot be used to rank concept sectors; all sector conclusions in this report are based on the 90 THS industry classifications.

---

0. One-Line Recap

Today was a textbook high-to-low rotation: the indices fell but the money-making effect rose — SSE Composite −0.11%, ChiNext −1.41%, STAR 50 −1.85%, while 2,957 stocks rose and 2,110 fell, with 82 limit-ups (77 yesterday), only 1 limit-down, and a 16.3% failed-limit rate. What fell was the crowded AI hardware chain; what rose was low-positioned resources, chemicals and agriculture: semiconductors saw a net outflow of RMB 16.470 billion, communications equipment −RMB 9.313 billion, components −RMB 3.927 billion, consumer electronics −RMB 3.588 billion, optics & optoelectronics −RMB 2.726 billion, electronic chemicals −RMB 2.606 billion, six items totalling about −RMB 38.6 billion (intraday snapshot basis); while the top net inflows were agrochemicals +RMB 1.742 billion, precious metals +RMB 1.729 billion, industrial metals +RMB 1.572 billion, chemical raw materials +RMB 1.217 billion. The industry composition of the limit-up pool says it even more bluntly: yesterday's limit-up pool held 5 communications equipment, 5 semiconductors, 4 components and 3 electronic chemicals, 17 in total; today those six hardware industries have only 4 left between them.

The pre-market list scored "both ends right, the middle wrong", and the errors follow a pattern. Judging the 26 listed stocks by "beat / missed the whole-market median (+0.30%)", the overall hit rate is 17/26 = 65.4%; broken out, it is three completely different numbers: the 14 "watch only — logic is negative" names hit 12 (85.7%), averaging −2.14%, with the 10-name negative-logic subgroup at 0/10 beating the median, the hardest single judgement in this report; the 5 recommendation slots in §9.1 hit 3/5, averaging +3.66% with a median of +3.32%, clearly beating the whole-market median of +0.30%; while the 6 Pass names hit only 2 (33.3%), averaging +2.90%; of those, the 4 passed on the reasoning "limit-up + net fund outflow" all rose, averaged +3.98%, 3/4 beat the median, and only 1 was a hit (25.0%) — using "limit-up accompanied by net fund outflow" pre-market to exclude them was systematically proven wrong by the market today.

But there is a conclusion more important than the hit rate: all 5 verification points set pre-market triggered exactly as designed today, and 4 of them delivered negative signals — including the one that hit limit-up. 星网锐捷 (002396) closed +10.01% limit-up, first on the whole list, yet the answer to its verification point ("can it post a 2nd consecutive day of net inflow?") was a net outflow of RMB 1.110 billion (intraday snapshot), the largest outflow among today's limit-up stocks; 深南电路 (Shennan Circuits, 002916) had a verification point reading "if turnover is still below 1.5% today and the price does not move, this stock should be dropped from the list", and today turnover was 1.36% with a close of −0.74%, precisely meeting that exit condition; 工业富联 (Foxconn Industrial Internet, 601138)'s "can turnover break 1%?" answered 0.80%, no break; 金安国纪 (Jin'an Guoji Technology, 002636)'s "can fund flow turn positive with turnover no higher than yesterday's 7.15%?" answered still a net outflow of RMB 483 million with turnover of 13.13% (doubled), failing both conditions at once. The headline percentage changes say the recommendations were right; the verification points say the execution was fragile, and today the verification points should govern.

No trading tomorrow (8/29 is a Saturday). The next trading day is 2026-08-31 (Monday), and it will open into an unprecedented density of earnings events: verified against the CNINFO scheduled-disclosure calendar, 4,598 of the market's 5,550 companies have already published interim reports, leaving 952 undisclosed, of which 706 are scheduled for 8/29 (tomorrow, Saturday) and 151 for 8/31 (Monday, the statutory deadline). This means roughly 700 interim reports will land over the weekend, with all of their share-price reactions piling into Monday's open, so any position held into today's close is an exposure that cannot be adjusted within the week. 澜起科技 (Montage Technology, 688008) and 长鑫科技 (ChangXin Memory Technologies, 688825) are in this batch; today they were −3.72% (net outflow of RMB 1.285 billion) and −0.88% (net inflow of RMB 420 million) respectively.

---

1. Market Overview

1.1 Indices and Turnover

Combined turnover across the two exchanges was RMB 2,101.7 billion. The index declines form a clean monotonic gradient: STAR 50 (−1.85%) > ChiNext (−1.41%) > SZSE Component (−0.68%) > CSI 300 (−0.46%) > CSI 1000 / CNI 2000 (−0.36%) > SSE Composite (−0.11%) > CSI Dividend (+0.27%). This gradient is not a "large-cap vs small-cap" ordering (CSI 1000 fell far less than ChiNext), it is an ordering by "AI hardware weight" — STAR 50 and the ChiNext Index are the two indices with the highest weights in semiconductors, optical modules and PCB, while the dividend index contains almost none of these constituents and was the only broad index to close green. This shows today was not a broad sell-off but a directed sector de-grossing.

1.2 Sentiment Indicators

Sentiment temperature: warming, on the strong side, but the structure is being replaced. The basis is four mutually independent readings all positive at once: only 1 limit-down, a 16.3% failed-limit rate in the low range, 82 limit-ups increasing against the tape, and advancers close to 1.4× decliners. The combination of "index down + strong money-making effect" corresponds in A-shares to essentially one state: money has not left, it has only changed direction. Today should therefore not be read as a contraction in risk appetite.

⚠️ But one self-imposed limit is required here: the "more up than down" reading above comes from THS's full table of 5,208 stocks, and that table is a pre-close snapshot (see the header), so it will be biased for stocks that changed direction into the close. That bias is, however, two-sided and has no systematic direction (there are both names like 誉衡药业 (002437) pulled to limit-up at the tail and names that dived into the close), unlike the one-sided truncation hit in the past where "East Money clist pagination gave only the top of the gainers list and everything missing was the tail of the losers". This table is the whole market, unsorted and complete, so no pagination bias exists. The qualitative conclusion on advancers/decliners is usable; precision to the single digit is not.

---

2. Reconciliation of the Pre-Market List

Judgement basis: "delivered or not" is uniformly judged by "did today's percentage change beat the whole-market median of +0.30%", not by "did it rise" — 2,957 stocks rose today, so "it went up, therefore it was right" would systematically inflate the hit rate. A "watch closely" name beating the median scores ✅; a "watch only / Pass" name lagging the median scores ✅.

"Open→Close %" is the column to read first in this table: the pre-market list was published at 08:05, so the actual entry price for executing the list is today's open, not yesterday's close. Looking only at the closing percentage change systematically overstates the list's executable return — across the 26 names, the average closing change is +0.72%, while the average open→close is −0.28%, with 16 names negative.

2.1 The Five §9.1 Recommendation Slots (all "watch closely" pre-market, no "priority deep-dive")

Recommendation-slot summary: the 5 names averaged +3.66%, median +3.32%, average open→close +2.68%, with 3/5 beating the whole-market median (+0.30%). Against the SSE Composite's −0.11% and ChiNext's −1.41% that day, this group clearly outperformed. But 4 of the 5 verification points delivered negative or falsifying signals, and only 生益科技 (600183) was a case of "the verification point was not met while fund flow gave a stronger positive answer".

2.2 The 14 "Watch Only / Do Not Chase a Gap-Up" Names

The table below has 16 rows: the 14 whose pre-market conclusion was "watch only / do not chase a gap-up", plus two that belong to other groups but sit on the same chain, listed for comparison — 澜起科技 (688008, pre-market §3 5th place, "watch closely", did not make the §9.1 recommendation slots) and 深南电路 (002916, already counted in the recommendation slots in §2.1). In the §2.4 hit-rate statistics these two are not counted in this group.

Summary: 12 hits out of 14 (85.7%), averaging −0.95%. Of these, the 10 "logic is negative" names (memory / PCB / wafer) averaged −2.14% with 0/10 beating the median, the cleanest group in this reconciliation; the other 4 "position too high" names averaged +2.02% with 2/4 beating, lifting the group average. Both misses (联瑞新材 (688300), 赛微电子 (300456)) belong to the "position too high" type, and both were rescued by the "do not chase a gap-up" qualifier.

2.3 The 6 Pass Names — the Most Systematically Wrong Group in This Reconciliation

⚠️ Summary and self-criticism: the Pass group hit only 2 of 6 (33.3%), averaging +2.90%, the lowest hit rate of any group. More precisely, the 4 names passed pre-market on the single test "limit-up + whole-market net fund outflow" (东材科技 (601208), 宏昌电子 (603002), 嘉立创 (001232), 长飞光纤 (601869)) averaged +3.98% today with 3/4 beating the median, and only 长飞光纤 (601869) was called correctly.

2.4 Hit Rate and Self-Criticism

One-line self-criticism: what was called right was the judgement that "the earnings content is negative"; what was called wrong was the judgement that "the fund flow is negative" — and in the pre-market report these two were used as the same kind of evidence.

For the type called right, the evidence is facts the companies disclosed themselves, which do not change with a single day's tape: 胜宏科技 (300476)'s Q2 single-quarter ex-non-recurring −7.7% QoQ, 德明利 (001309)'s H1 figures already pre-announced in July, 中芯国际 (688981)'s Q2 figures appearing for the third time. These 10 names went 0/10 on beating the median, with no exception.

For the type called wrong, the evidence is the single-day cross-sectional indicator "limit-up yesterday but net fund outflow". The pre-market report used it to Pass 4 names, and 3 beat the median today; this is already the second appearance of the same error — single-day fund-flow extremes are mean-reverting, they measure "who was selling yesterday", not "who will fall tomorrow". More decisively, today's strongest recommendation slot, 星网锐捷 (002396), is itself a "limit-up + net outflow of RMB 1.110 billion", and it hit limit-up. In other words, within a single day the pre-market report both passed on four names because of this pattern (all of which rose) and recommended one in spite of this pattern (which hit limit-up) — the test in fact produced no discriminating power at all, it was simply applied selectively.

Three concrete fixes for the next pre-market report:

1. "Limit-up + net fund outflow" will no longer serve as a standalone Pass reason, and is demoted to a position note ("participate but at half size"); a Pass must rest on evidence such as earnings, catalyst freshness or valuation, which do not fluctuate with a single day's flows.
2. Verification points must be laid out separately from conclusions and reconciled on their own. Today's 5 verification points clearly carried more information than the 5 conclusions — 深南电路 (002916)'s exit condition and 工业富联 (601138)'s 1% turnover line were both hit word for word. Recommend the pre-market report give verification points the same layout weight as the recommendation conclusions.
3. The value of qualifiers such as "do not chase a gap-up" has been underrated. 联瑞新材 (688300) and 赛微电子 (300456) were both called wrong on direction, yet because they carried this qualifier, actual execution did not lose but gained (their open→close figures were −4.46% and +1.53% respectively). Recommend mandatory open-execution notes on all high-positioned names, not just a select few.

---

3. Verification of Today's Main Themes

3.1 Branch Level

3.2 Did the Pre-Market Report Identify the Strongest Theme: No

Today's genuinely strongest group was agriculture + chemicals + resources, and not one of the pre-market report's 26 names was in it.

This theme's footprint in the limit-up pool is overwhelming: 6 chemical products + 5 agrochemicals + 3 chemical raw materials + 3 planting + 2 agricultural product processing = 19 names, 23.2% of today's 82 limit-ups, including 昊华科技 (Haohua Chemical Science & Technology, 600378, SSE main board, 2 boards, turnover RMB 2.976 bn), 金牛化工 (Jinniu Chemical, 600722, SSE main board, turnover RMB 2.060 bn), 赤天化 (Chitianhua, 600227, SSE main board, turnover RMB 1.601 bn), 新农股份 (Xinnong Chemical, 002942, SZSE main board, 3 boards) and 敦煌种业 (600354, SSE main board). At the same time precious metals +1.24% (net inflow RMB 1.729 bn), industrial metals +1.14% (+RMB 1.572 bn), steel +1.46% and coal +0.96% moved in the same direction — this is a complete pro-cyclical / resource repricing, not an isolated agricultural move.

⚠️ This report failed to obtain a primary catalyst for this theme. The public information retrieved was mostly stale material from March to August (methanol supply, the chemical cycle, refrigerant price increases), and not one item could be confirmed to fall within the window from the 8/27 close to the 8/28 open; per the iron rules, this report will not invent a catalyst for this theme and records it as "no reliable data available". All that can be confirmed is the price action itself: it was today's largest gainer, largest limit-up count, and held three of the top four net-inflow slots.

Why the pre-market report missed it: the pre-market scan window runs "news flow → positive branches → stocks", and what landed in concentration last night was the electronics chain's interim reports, so all scanning attention was pulled into the AI hardware chain. The "pricing progress table" in §1.1 of the pre-market report had in fact already given the clue — it showed that the limit-ups in semiconductors, components and electronic chemicals were all on "day 1", and the pre-market report read "day 1" as "just getting started"; today's outcome shows it can also mean "this one day only". "Limit-up day 1" is a signal with an undetermined direction, and the pre-market report interpreted it in one direction only.

3.3 ⚠️ One Branch Judgement That Must Be Self-Reversed: the PCB Manufacturing Downgrade

The pre-market report cut PCB manufacturing from 2nd to 6th (C+, no recommendation slot), with the core argument that "upstream materials see margin expansion while downstream PCB manufacturing gets squeezed". Today's cross-section does not support that layering.

Three conclusions:

1. The ordering was scrambled. The expected ① > ② > ③ was in fact ① > ③ > ②, with the layer downgraded to 6th beating the layer ranked 2nd by 1.87pct.
2. Layer ③'s internal spread of 16.16pct far exceeds the gap between layers (at most 4.06pct). This means "position in the value chain" had almost no explanatory power today; what did have explanatory power was each stock's own interim report. The only decliner in that layer was 胜宏科技 (300476, −8.15%), and it is precisely the one that disclosed last night "Q2 single-quarter ex-non-recurring −7.7% QoQ, attributable profit resting on unsustainable paper gains"; 生益电子 (688183, +7.81%) and 嘉立创 (001232, +8.01%), with no negative interim reports, led the gains instead.
3. The pre-market report actually foresaw this but did not change its conclusion accordingly. The original text reads: "深南电路 (002916)'s PCB segment gross margin is up +2.43pct YoY, a counterexample to this proposition… this report's squeeze judgement rests on the single-quarter data of 胜宏科技 (300476) alone, a sample size of 1". Having concluded the sample size is 1, it should not have cut the whole layer by 4 places on that basis. The correct move was to downgrade 胜宏科技 (300476) alone (that step was done right; today's −8.15% is the largest decline on the whole list) and not downgrade the layer.

The correction in this report is therefore: the layering hypothesis "upstream materials stronger than downstream manufacturing" was not supported today; what was actually validated is something more prosaic — the stocks that put out negative interim reports last night fell today, and the ones that did not, did not. The pre-market report took a correct stock-level judgement and scaled it into an industry-level conclusion, and the scaled-up part was wrong.

---

4. Limit-Up Tiers and Board Height

4.1 Consecutive Limit-Up Distribution

Tier shape: enough height (7 boards) but a thin waist. Of the 82 limit-ups, 64 were first boards (78.0%), with only 18 at 2 boards or above. This is a structure of "many new themes igniting, old themes failing to hand off" — consistent with the sector conclusion in §3.2: today is day one of a new direction, not day N of an old one.

4.2 Promotion Quality

Of yesterday's 77 limit-up stocks, 18 hit limit-up again today, a promotion rate of 23.4%. But the industry distribution of the promotions is highly directional:

The 17 AI hardware chain limit-ups from yesterday (5 communications equipment + 5 semiconductors + 4 components + 3 electronic chemicals) all landed in the +0.27% ~ +2.19% band today, none above +2.2%, and all four groups lagged yesterday's limit-up cohort overall. Meanwhile yesterday's limit-ups from traditional industries were generally above +5% today. Relay money clearly steered away from yesterday's strongest theme.

⚠️ One execution-related distinction that must be labelled: today's 5-board 海鸥住工 (002084) traded only RMB 39 million on a 1.0% turnover rate, the 4-board 万向德农 (600371) traded RMB 57 million on 1.7%, and 新赛股份 (600540) traded RMB 26 million on 0.9%. The pre-market report listed this "shrinking-volume one-way board" type as not worth chasing, which was wrong on price direction (all three hit limit-up) but right on executability — you simply cannot get filled on these boards. Putting them alongside "relaying the high boards" under the same avoidance reason pre-market was imprecise: 深中华 A (000017) traded RMB 1.253 billion on a 25.1% turnover rate and is buyable; 海鸥住工 (002084) is not. These two types should be written up separately.

4.3 Identifying the Core Names (large turnover, rising, high recognizability)

Seal quality is generally weak: 7 of the 8 have a seal order / turnover ratio below 0.15, and even the highest, 昊华科技 (600378), is only 0.145, on a turnover rate of just 5.2% (lowest in the group). By the empirical threshold of "seal order / turnover ≥0.15 means a real seal", almost none of today's limit-ups was hard-sealed. This does not contradict the low 16.3% failed-limit rate — a low failed-limit rate says "once sealed, it did not reopen", while a low seal ratio says "the seal was barely held". Read together: today's limit-ups were built on absorption rather than on seal-order size, and the stability of Monday's (the next trading day's) open depends on whether new money arrives, not on leftover seal orders from yesterday.

The most core-like name is 昊华科技 (600378) — the only 2-board name in the chemical theme, turnover near RMB 3 billion, a turnover rate of just 5.2% (meaning chips have not changed hands on any scale), and the highest seal ratio in the group. ⚠️ But it is simultaneously one of today's largest single-stock net fund outflows in the whole market (−RMB 2.252 bn, intraday snapshot basis, inflow RMB 349 mn / outflow RMB 2.600 bn), and this "limit-up + extreme net outflow" combination is exactly the pattern criticized in §2.4 — following that self-criticism, this report does not judge it as distribution on that basis, and only flags it as a contradictory signal.

---

5. Fund Flows

5.1 Northbound Flows: This Data No Longer Exists

Daily northbound net buying has not been disclosed since 2024-08-19, and this report does not carry that indicator. A specific caution: some data interfaces still silently return 0.0 for that field, or return the "quota ceiling" (e.g. RMB 42 billion) rather than an actual inflow figure, so any statement claiming "northbound net inflow of RMB X billion today" should be treated as a methodological error.

5.2 Industry Fund Flows (THS basis, pre-close intraday snapshot)

Top net inflows — all low-positioned pro-cyclicals:

Top net outflows — almost entirely the AI hardware chain:

The six AI hardware industries (semiconductors, communications equipment, components, consumer electronics, optics and optoelectronics, electronic chemicals) had a combined net outflow of about RMB 38.6 billion, far exceeding the sum of all other outflowing industries. Against a combined top-ten net inflow of about RMB 10.3 billion — the money leaving clearly exceeds the money entering the low-positioned sectors, and the difference went off-market or into index products, which also explains why the indices fell while individual stocks rose.

One question of magnitude worth flagging: semiconductors' −RMB 16.470 billion sits on a denominator of RMB 257.7 billion in turnover, i.e. 6.4%; while agrochemicals' +RMB 1.742 billion sits on a denominator of only RMB 21.6 billion, i.e. 8.1%. The absolute values differ by 9×, but agrochemicals is the stronger on a relative basis. Ranking sectors by absolute amount systematically favours large-cap sectors, so this section gives both bases.

5.3 Single-Stock Fund Flows (Intraday Snapshot)

Top net inflows: 生益科技 (600183) +RMB 1.008 bn (first in the whole market), 宁德时代 (CATL, 300750) +RMB 981 mn, 中际旭创 (Innolight Technology, 300308) +RMB 904 mn (share price −0.21%), 万华化学 (Wanhua Chemical, 600309) +RMB 752 mn, 生益电子 (688183) +RMB 673 mn (+7.81%), 新易盛 (Eoptolink Technology, 300502) +RMB 563 mn (share price −1.61%), 中国石化 (Sinopec, 600028) +RMB 482 mn, 长鑫科技 (688825) +RMB 420 mn.

Top net outflows: 昊华科技 (600378) −RMB 2.252 bn (yet limit-up), 兆易创新 (603986) −RMB 1.798 bn, 澜起科技 (688008) −RMB 1.285 bn, 亨通光电 (Hengtong Optic-Electric, 600487) −RMB 1.225 bn, 胜宏科技 (300476) −RMB 1.195 bn, 星网锐捷 (002396) −RMB 1.110 bn (yet limit-up), 长飞光纤 (601869) −RMB 1.045 bn, 中天科技 (Zhongtian Technology, 600522) −RMB 1.021 bn.

Two readings:

1. The CCL complex is the only part of today's AI chain with genuine money absorbing supply: 生益科技 (600183) at +RMB 1.008 billion led the whole market, and 生益电子 (688183) at +RMB 673 million. Against a backdrop where semiconductors, communications equipment and components saw a combined outflow of nearly RMB 30 billion, these two took RMB 1.68 billion of net inflow against the tide — the only evidence that still stands behind the pre-market "CCL as the 1st branch" judgement, and it is not their percentage gains (+3.32% / +7.81% are both modest) but the direction of the money.
2. Optical modules show a "price down but money in" divergence: 中际旭创 (300308) −0.21% with +RMB 904 mn, 新易盛 (300502) −1.61% with +RMB 563 mn, while their optical-communications peers 亨通光电 (600487) −RMB 1.225 bn, 长飞光纤 (601869) −RMB 1.045 bn and 中天科技 (600522) −RMB 1.021 bn saw heavy outflows. Optical modules (旭创 / 易盛) and fibre-and-cable (亨通 / 长飞 / 中天) were treated distinctly differently by the money today. ⚠️ Given this is an intraday snapshot and both groups moved into the close, this report only flags that the divergence exists and does not conclude that "money is bottom-fishing optical modules" — that needs re-checking on Monday's full-day basis.

5.4 Dragon-Tiger List: Left Blank in This Report

The exchanges publish the dragon-tiger list after 18:00 while this report goes out at 15:30, so today's seat data was not obtained. This is a fixed timing gap, and 8/28's dragon-tiger list will be added in Monday's pre-market report. The pre-market report added 8/27's dragon-tiger list in exactly this way; the process works.

---

6. Outlook for the Next Trading Day

⚠️ Tomorrow (2026-08-29) is a Saturday, market closed. This entire section refers to the next trading day, 2026-08-31 (Monday). The delivery window for every "watch / avoid" item is 8/31, and positions held into today's close must span two calendar days and cannot be adjusted within the week.

6.1 ① Theme Continuation Assessment

6.2 ② Key Names to Watch on 8/31 (with Verification Points)

This section is executed per the self-criticism in §2.4: verification points are laid out with equal weight to conclusions, and "single-day net fund outflow" is not used as an exclusion reason.

⚠️ This section sets no "priority deep-dive"; all entries are watch slots. Reason: this report did not obtain a primary catalyst for today's strongest direction (agriculture / chemicals / resources) (see §3.2), and giving a top-tier recommendation slot without knowing what the driver is amounts to reverse-engineering logic from price action, violating this machine's iron rule against hindsight fitting. The first task of the 8/31 pre-market report is to fill in that catalyst.

6.3 ③ What to Avoid on 8/31

1. Bounces in the AI hardware chain: the outflow scale across semiconductors (−RMB 16.470 bn), communications equipment (−RMB 9.313 bn) and components (−RMB 3.927 bn), together with the damage to breadth (semiconductors 151 down 33 up), is not repairable in a single day.
2. The substrates layer as a whole: the only one of the three layers down overall today (−0.59%), and 深南电路 (002916) has already triggered its exit under the pre-market rule.
3. 深南电路 (002916), SZSE main board ±10%: the exit condition written by the pre-market report itself has triggered (turnover 1.36% < 1.5% and the price did not move), and this report enforces that rule and removes it from the list.
4. 澜起科技 (688008), STAR Market ±20%: −3.72% today with a net outflow of RMB 1.285 billion, and its interim report discloses on 8/29 (Saturday), producing a direct gap reaction on 8/31. No reason to hold before the event; revisit after it.
5. Today's shrinking-volume one-way high boards: 海鸥住工 (002084, 5 boards, turnover RMB 39 mn), 万向德农 (600371, 4 boards, turnover RMB 57 mn), 新赛股份 (600540, 2 boards, turnover RMB 26 mn). The reason to avoid is non-executability (you cannot get filled), not a directional call — all three hit limit-up today, so avoiding them pre-market on "direction" was wrong, and avoiding them on "liquidity" is right.
6. 深中华 A (000017), SZSE main board ±10%, 7 boards: seal order / turnover of only 0.008 and a 25.1% turnover rate, with high-level relay risk rising exponentially with the board count. But this must be recorded honestly: the pre-market report avoided it for the same reason (at 6 boards then) and it hit limit-up again today. The historical hit rate of this avoidance item is negative; weight it down accordingly.
7. The pharmaceutical chain: healthcare services −2.63% (5 up 51 down), biologics −2.41% (5 up 48 down), chemical pharmaceuticals −0.76% (net outflow RMB 4.017 bn), the worst breadth of any group today.

6.4 ④ Input Notes for the 8/31 Pre-Market List

1. First task: establish the primary catalyst behind the agriculture / chemicals / resources theme. This report did not obtain it (§3.2), and it is this report's largest information gap. Suggested lines of verification: agricultural and chemical product price data within the 8/27–8/28 window, relevant ministry policy, and the company announcements of 昊华科技 (600378), 金牛化工 (600722), 赤天化 (600227) and 敦煌种业 (600354). Until the catalyst is established, no recommendation slot may be given to this theme.
2. Stop using "limit-up + single-day net fund outflow" as a standalone Pass reason. Of the 4 names passed on that test today, 3 beat the median; and today's strongest name, 星网锐捷 (002396), is itself that pattern. Convert it into a position note.
3. Roughly 700 interim reports will land over the 8/29–8/30 weekend (verified against the CNINFO calendar: 952 undisclosed, of which 706 are scheduled for 8/29 and 151 for the 8/31 statutory deadline). The 8/31 pre-market scan window will cover a pile of announcements far denser than usual, and must first be triaged by "how many times has this been said" (anything already public via a pre-announcement, flash report or HK-listing announcement is not incremental), and only then by whether the numbers are good — a method already validated this week on 德明利 (001309), 金安国纪 (002636), 中天科技 (600522) and 中芯国际 (688981).
4. Layering hypotheses must be tested on the cross-section before being used to rank. Today the PCB manufacturing layer's internal spread was 16.16pct while the largest gap between layers was only 4.06pct (§3.3); when within-layer variance far exceeds between-layer variance, the layering has no explanatory power and one should return to stock-level evidence. Recommend the pre-market report compute a within-layer / between-layer variance ratio once before publishing branch rankings.
5. "Verification points" were the highest-information module produced this week. Four of today's 5 verification points triggered precisely, and among them 深南电路 (002916)'s exit condition and 工业富联 (601138)'s 1% turnover line were hit word for word. Recommend the 8/31 pre-market report give verification points their own chapter and spell out "what to do once triggered" (as this report did in enforcing 深南电路' exit rule).
6. The avoidance list must distinguish "directional judgement" from "executability judgement". Today's three shrinking-volume one-way boards all hit limit-up (direction called wrong) but each traded below RMB 60 million (not executable), and mixing these two types under one reason distorts the hit-rate statistics of the avoidance list.
7. Fund-flow data currently rests on the single THS line and is an intraday snapshot (East Money's clist and push2his are both down). When the 8/31 pre-market report cites any fund-flow figure it must label the snapshot nature and avoid subtracting it directly against historical data on the full-day basis.

---

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