A-Share Market Recap | 2026-08-18 Tuesday

0. One-Line Recap

Sentiment temperature: a divergent warm-up, but the tech side is ice cold. The SSE Composite +0.19%, the ChiNext Index −0.93%; market-wide 2,121 gainers versus 3,292 decliners, 79 limit-ups (106 yesterday), 5 limit-downs, failed-seal rate 22.5%. The strongest main line was not memory, as the pre-market brief judged, but agriculture / food security — of all 79 limit-ups, 28 (35.4%) came from the agriculture-and-food chain (crop farming 13, agri-product processing 4, animal husbandry 3, feed 2, agrochemicals 2, snack foods 2, food processing 1, beverages & dairy 1); the seed sector +14.55%, crop farming +11.83%.

Pre-market list scorecard: the direction was called wrong, the verification points were called right. The memory branch, graded S in the pre-market brief, saw the memory-chip concept up only +0.29% today while main-force money net-exited RMB 11.204 billion (yesterday: net inflow of RMB 20.184 billion); the flagship ChangXin Memory (688825) −4.22%, main-force net outflow RMB 1.560 billion. The two verification points set for this line in the pre-market brief ("downgrade if ChangXin lags or turns red" and "can main-force net inflow continue for a 2nd day") both triggered downgrade signals today — the conclusion was wrong, but the criteria were right.

The 19 names on the recommendation list averaged +0.43% today (median −0.06%); the 17 names on the Pass list averaged +3.82% (median +2.82%). This is the second consecutive session in which the Pass group beat the recommendation group. On the executable "open→close" basis, the gap narrows to +0.39% versus −0.94% (see §2).

Tomorrow's tone: the main line is rotating, but the relay quality of the new main line is no better. Of the 28 agricultural limit-ups, 7 only sealed after 14:25, and the median seal-order/turnover ratio across all 79 limit-ups is just 0.060. Memory has entered the disagreement phase; GigaDevice (603986) reporting interim results tomorrow is the only certain standalone event.

---

1. Market Overview

Turnover: market-wide total of about RMB 2.42 trillion (self-computed basis: the sum of the day's turnover across 5,899 A-shares = RMB 2.4198 trillion). Morning half-session turnover already exceeded RMB 1.6 trillion (Sina Finance¹), meaning the afternoon half-session was about RMB 0.8 trillion — volume was clearly front-loaded; the afternoon shrank rather than expanded.

Breadth and Limit-Up Structure

Price-limit basis by board: SSE/SZSE main boards ±10% / ChiNext ±20% / STAR Market ±20% / BSE ±30%. Limit-up identification in this report uses the exchanges' official limit-up pool throughout; no self-screening by a fixed threshold was performed.

Sentiment read: warming but not euphoric, and structural. Three reasons:

1. The direction of index divergence is unambiguous: SSE Composite +0.19% versus ChiNext Index −0.93%; BSE 50 +2.67% versus SZSE Component −0.56%. Money is rotating from large-cap growth into small caps and thematic names — this is not a broad-based lift in risk appetite.
2. Breadth is negative: 3,292 decliners against 2,121 gainers. Limit-ups shrank 25% from yesterday's 106 to 79.
3. Relay quality has not improved: the median seal-order/turnover of 0.060 sits in the same bracket as yesterday's semiconductor complex at 0.03–0.09; 13 limit-ups only sealed after 14:00. The criterion set in pre-market §8.4 was "ladder quality only materially improves if a semiconductor limit-up with seal-order/turnover > 0.3 appears today" — today the only semiconductor limit-up was Huaya Intelligent (003043) with a seal ratio of 0.107, so that criterion was not met.

---

2. Scorecard Against the Pre-Market List

Basis note: this table gives two change columns. "Close chg%" is versus yesterday's close; "Open→Close%" is versus today's opening price. The pre-market list is published before the open, so the earliest executable price for a reader is today's open, not yesterday's close. Any gain delivered in one shot via a gap-up is unavailable to the reader — this second column is the list's real report card.

2.1 Full Recommendation List, 19 Names (priority deep-dive / watch closely / watch only)

Recommendation list, 19 names in aggregate: average +0.43%, median −0.06%; average open→close −0.94%, median −0.72%; 9 of 19 up; 1 limit-up.

2.2 Pass List, 17 Names

Pass list, 17 names in aggregate: average +3.82%, median +2.82%; average open→close +0.39%, median 0.00%; 11 of 17 up; 3 limit-ups.

2.3 Hit Rate and Post-Mortem

Performance by group (this is the most damning table):

The list's ranking is exactly inverse to today's actual returns: the more conservative the conclusion, the better the performance.

Post-mortem 1: two honest basis corrections — the headline gap alone must not be reported

The Pass group's +3.82% mean cannot be taken at face value. Three deductions are mandatory:

• Rifeng Cable (002953) was a one-price board with turnover of RMB 63 million, a turnover rate of 2.10% and a seal ratio of 1.783 — the reader could not buy it. Excluding it, the Pass group mean falls to +3.43%;
• Zhongshi Technology (300684): of its +20.00%, 19.53 percentage points came from the gap-up, leaving only +0.39% after the open. Excluding it as well, the Pass group mean falls to +2.33% and the median to +0.44%;
• On the open→close basis, the Pass group is only +0.39%, and the gap versus the recommendation group narrows from 3.39 percentage points to 1.33 percentage points.

The conclusion still stands (the Pass group did outperform), but the magnitude is only four-tenths of the headline number. Yesterday's recap made the same mistake — reporting the gap on close-to-close changes systematically overstates the cost of "missing out".

Post-mortem 2: the Pass group has outperformed for a second straight day and can no longer be treated as noise

Yesterday: Pass group open→close +5.30% versus the recommendation group's +2.95%, a gap of 2.35 percentage points. Today: Pass group +3.82% versus the recommendation group's +0.43% (close basis), a gap of 3.39 percentage points.

Two days in the same direction, and today's gap is larger. The mechanism is clear: the Pass rationales — expensive valuation, hot-money structure, negative cash flow, overheated turnover — are all "quality and crowding" metrics, whereas over these two days the market has been pricing on "is it on the main line that is currently spreading". Using quality metrics to Pass screens out, along with everything else, the names that are low quality but happen to be standing on the new main line; Shennong Seed Industry (300189) was screened out exactly this way — its Pass rationale, "turnover of 33.25%, the highest of the day", became 44.12% today, and it limited up.

Pre-market §4 had already written this lesson once and claimed two methodological adjustments had been made; today's result shows the adjustments did not work. The real problem is not the weights, it is the order: if you first slice narrative branches and then rank by financial quality within a branch, then when the branch is sliced wrong, ranking accurately inside it is meaningless.

Post-mortem 3: the costliest error was setting the wrong read moment for the "Korea leading verification point"

The pre-market brief flagged this as "the highest-information-value leading observation point today" and set the recommended read moments at 08:05 and 08:30 Beijing time (i.e. 09:05 and 09:30 Korea time, 5–30 minutes after the open). The criteria table read: catch-up gain > +4% → "the S-grade strength of branch 1 holds".

Actual readings on the day (all with the read moment annotated):

At the read moment specified pre-market (after the open), SK Hynix was up +5.53%, landing in the ">+4%" bracket, and the verification point returned "S-grade strength holds" — a confirmation signal. At the close, only +1.03% remained, open→close −4.26%, Samsung negative, KOSPI negative. The same verification point yields opposite conclusions at the open reading and the close reading, and the pre-market brief specified only the open reading.

The lesson is concrete: for a cross-market "independent pricing vote", the jump at the open is not a vote, it is the handover of the overnight US session. Only the close is a full day of pricing. And the A-share side did not follow Korea's open at all — ChangXin Memory (688825) started from a gap-down of −1.05%.

Post-mortem 4: one falsified inference — "the market will re-rank packaging & testing by profit quality"

Pre-market wording: "If JCET Group (600584) gaps up more than Tongfu Microelectronics (002156) today, it means the market is starting to re-rank the three packaging-and-testing names by 'profit quality' rather than 'money heat' — this is the inference this brief most wants to test."

Actual: JCET gapped −0.19% and closed −0.50%; Tongfu gapped +2.80% and closed +2.02%. The inference was fully falsified. JCET's three points — "non-recurring items only about 4%, a debt ratio of 43%, the lowest among peers, and growth clearly attributable to AI demand" — were all true, and not one of them paid a cent today.

At the same time, one flip-side fact must be recorded: Tongfu Microelectronics rose +2.02% while main-force money net-exited RMB 1.030 billion. Price won; money did not. This shows that "profit quality" is not useless — its payoff cycle is simply not one day. Placing a medium-term valuation judgement inside an intraday criterion such as "today's gap-up magnitude" is an error in criterion design, not an error in the judgement itself.

Post-mortem 5: the equipment branch — the same error in a new costume

The pre-market brief used a correct industry proposition ("equipment orders point to new supply in 2027–2028, the opposite direction from the current shortage narrative") to tag AMEC (688012), Piotech (688072) and Nata Opto-electronic (300346) all as "watch only".

Today: semiconductor equipment was the only semiconductor sub-sector with main-force net inflow (+RMB 868 million); Piotech +5.07% (main-force +RMB 548 million, 5th market-wide), AMEC +2.71% (+RMB 499 million), Naura Technology (002371) +1.44% (+RMB 295 million).

That industry proposition is still correct today — equipment really does speak to future supply. But it answers "does this logic hold", not "where will money go today". Using a medium-term directional judgement as a short-term ranker is exactly the "financial quality is not a short-term ranker" criticised in yesterday's recap, restaged today as "industry direction is not a short-term ranker".

Post-mortem 6: agriculture — the biggest miss, and the pre-market brief already held the evidence

The pre-market brief ranked seeds / food security 6th (last), grade C+, on the grounds that "the catalysts all fall outside the scan window (state price-support procurement on 8/5 and 8/11), and the explanatory articles were published after the close, making them ex-post attribution".

Today's facts: the agriculture-and-food chain contributed 28 of the 79 limit-ups (35.4%); seeds +14.55%, crop farming +11.83%, edible fungi +11.57%, the grain concept +7.45% (main-force +RMB 1.625 billion), agricultural planting main-force +RMB 2.632 billion.

Basis note (to avoid inflation): the 28 above are counted strictly on Eastmoney's sector classification, and exclude 4 general-retail names (Jiajiayue 603708, Sanjiang Shopping 601116, Quanxinhao 000007, Zhejiang Dongri 600113) and 2 chemical-raw-material names (Hongxing Development 600367, Jinniu Chemical 600722). Folding those 6 in as "broad consumer / agrochemical" would make the number 34 (43%) — this report does not adopt that basis, because Hongxing Development's leadership on the day is attributable to the glass-substrate concept, which has nothing to do with agriculture.

The real catalyst was not identified pre-market: the J.P. Morgan research report "Food Security Is National Security: A Compounding Storm", published on 2026-08-16 by the team of London-based global economist Nora Szentivanyi, which groups the drivers into "five Ws" (War / Weather / Warehousing / Water / Waste) and projects global food inflation rising from 2.8% in 2026H1 to 5% in 2027H1. Sina Finance², National Business Daily³, JRJ⁴

This must be spelled out, because it is not a question of "luck":

1. The publication date of 8/16 does indeed fall outside the pre-market scan window (8/17 09:00 → 8/18 07:00) — window discipline was executed correctly. What was wrong was equating "published outside the window" directly with "low strength". This catalyst was already in day 1 of pricing on 8/17 (Jinjian Cereals limited up, Shennong Seed Industry turned over 33%), and 8/18 is day 2 of pricing. The window measures "when it was published"; the market trades "when it started being priced" — a catalyst in day 2 of pricing should not be scored by its publication date.
2. The pre-market brief already held the strongest piece of tape evidence, and wrote it underneath the weakest conclusion. The exact wording for Jinjian Cereals (600127) at position 20 of §3 was: "seal-order/turnover of 0.629, among the strongest first boards of the day; Dragon-Tiger net buying at 29.86% of total turnover, the highest of the day" — among all 20 names this is the only record where "best seal quality" and "highest Dragon-Tiger concentration" hold simultaneously, and its conclusion was "watch only", with a Pass rationale of "the catalysts all predate the scan window".
3. Compare the description of the semiconductor complex in the same brief: "seal-order/turnover is broadly 0.03–0.09, extremely weak". The pre-market brief observed both the strongest and the weakest seal quality, and allocated all its resources to the latter.

This is the only fully actionable improvement item in this scorecard (written into §6.4).

---

3. Main-Line Verification

Did the pre-market brief call the strongest main line? — No

Branches 1, 2 and 3 as judged pre-market (memory, packaging & testing, chip design) contributed 0 limit-ups today, and all three saw large main-force net outflows; branch 6 (last place) contributed 35.4% of all limit-ups market-wide.

Were there surprise main lines the pre-market brief missed? — Two

1. Agriculture / food security (see §2.3, post-mortem 6), main-line grade;
2. Panels / optical & optoelectronics, secondary-line grade. BOE Technology A was today's largest main-force net inflow market-wide (+RMB 886 million), IRICO Display limited up, and the panel sector took in main-force +RMB 1.596 billion. The word "panel" does not appear anywhere in the pre-market brief. This one differs from agriculture — it has no identifiable same-day catalyst and looks more like a nearby reallocation inside the electronics space after money withdrew from semiconductors and optical modules.

One flip-side fact that must be stated (to avoid hindsight bias)

The memory chain did not "crash" today; it stalled. The memory-chip concept closed +0.29% (positive), and among the three module names Demingli Technology (001309) was even up 0.44%, with Puya Semiconductor (688766) +0.17%. What was violent was the money: concept main-force flows swung from +RMB 20.184 billion to −RMB 11.204 billion, a single-day reversal of about RMB 31.4 billion. Price barely moved while money turned around — that is the shape of disagreement, not the shape of a completed ebb. Reading today as "the memory move is over" is over-interpretation — this report claims only this: the S grade (strongest market-wide) assigned to it pre-market does not hold today.

---

4. Limit-Up Ladder / Highest Boards

4.1 Highest boards: all three of yesterday's 4-board names died, and the new 4-board is unrelated to the main line

Today's highest streak = 4 boards: Shenqi Pharmaceutical (600613, SSE main board ±10%), chemical pharma. Seal-order/turnover of only 0.015, 1 failed seal, turnover rate 15.1%, turnover of RMB 561 million — the worst seal quality among all consecutive-board names. The highest board is unrelated to today's main line (agriculture).

4.2 Full consecutive-board ladder (18 names at 2 boards or above)

Advancement: yesterday's 106 limit-ups → 17 continued today, a promotion rate of 16.0%; yesterday's limit-up names averaged +1.16% today (median −0.15%), with an average gap-up of +2.33% but an average open→close of −1.12%, 50 up / 55 down, and 8 failed seals today.

The right way to read "yesterday's limit-ups" relay quality: an average gap-up of +2.33% but an average open→close of −1.12% means next-day returns on yesterday's limit-up names were delivered almost entirely by the gap; after the open they were a net loss. This is the same thing, seen from the other side, as the "yesterday's limit-up concept saw main-force net outflow of RMB 3.376 billion" observed in yesterday's pre-market §8. Today the "yesterday's consecutive boards" concept was main-force −RMB 1.062 billion, "yesterday's consecutive boards incl. one-price" −RMB 1.455 billion, and "yesterday's high turnover" −RMB 7.102 billion — relay money is still flowing out.

4.3 Flagship identification: no qualified flagship emerged today

Screened on the three tests of "large turnover + up on the day + recognisability", no agricultural stock qualifies as a flagship today — the largest turnover among agricultural limit-ups belongs to 隆平高科 (Yuan Longping High-Tech, 000998,1.022 billion yuan) and 北大荒 (Beidahuang, 600598,1.349 billion yuan), an order of magnitude below the tech names (ChangXin Memory RMB 25.460 billion, GigaDevice RMB 26.795 billion).

This is the most substantive weakness of the agriculture line: it has a ladder (28 limit-ups) and a concept (GMO +8.27%), but no flagship able to absorb large money. Yesterday's memory line was the opposite — it had a flagship (ChangXin, RMB 25.4 billion of turnover) but a thin ladder. Each line is missing the other half.

4.4 Seal quality: slightly improved but still at a low level

• Across all 79 limit-ups the median seal-order/turnover is 0.060, with only 15 above 0.3;
• Of the 28 agricultural limit-ups, 20 completed their first seal between 09:25 and 10:22 (early-session consensus), while 7 did so after 14:25 (敦煌种业 Dunhuang Seed 600354 / 新农开发 Xinnong Development 600359 / 隆平高科 Yuan Longping High-Tech 000998 / 苏垦农发 Suken Agricultural Development 601952 / 大北农 Dabeinong 002385 / 国投丰乐 SDIC Fengle Seed 000713 / 一鸣食品 Yiming Food 605179) — a late-session chase structure, the same pattern as yesterday's afternoon seals in the semiconductor complex;
• 8 sealed directly in the 09:25 auction, of which 4 had a seal ratio > 1 (学大教育 Xueda Education 000526 = 2.894, 桂发祥 Guifaxiang 002820 = 2.987, 日丰股份 Rifeng Cable 002953 = 1.783, 丰原药业 Fengyuan Pharmaceutical 000153 = 1.553) — these four cannot be bought and do not constitute a return opportunity.

---

5. Fund Flows

5.1 Northbound flows: not applicable (must be stated)

Since 2024-08-19, Stock Connect has ceased disclosing daily northbound net buying, and this report does not provide that indicator. Market data interfaces will silently return 0.0 or the quota ceiling under this basis, which cannot be used as evidence that "flows are neutral". All flow analysis in this report uses the exchanges' publicly disclosed order-classified fund flow (main-force net) basis.

5.2 Dragon-Tiger List: not obtained for this report

The Dragon-Tiger List is published by the exchanges around 18:00 after the close, and this report was generated at 15:30, before the day's list was published, so this section is left blank. This is a known information gap in this report — yesterday's pre-market conclusions on Tongfu Microelectronics, Fastprint Circuit Tech, T&S Communications and Jinlu Electronics were changed precisely by breaking down Dragon-Tiger seats; without this piece, today's question of "whether the buying in Shennong Seed Industry (300189) and Jinlu Electronics (301282) was institutional or hot money" cannot be adjudicated, and it must be backfilled in tomorrow morning's pre-market brief.

5.3 Sector Main-Force Net Flows

Largest net inflows:

Largest net outflows:

⚠️ Two basis notes; ignore them and you will misread the orders of magnitude:

① Sector main-force nets overlap heavily and must not be added up. "Electronics −RMB 22.802 bn" already contains most of "semiconductors −RMB 12.311 bn", "components −RMB 9.746 bn" and "printed circuit boards −RMB 8.019 bn"; "telecom −RMB 13.575 bn" contains "telecom equipment −RMB 12.500 bn". Adding them together produces a number that does not exist. Likewise, "agriculture, forestry, animal husbandry & fishery +RMB 3.634 bn" already includes "crop farming +RMB 1.615 bn" and "seeds +RMB 924 mn".

② New-listing contamination check (performed). Pinzhun Laser (688826) listed on the STAR Market today and closed at RMB 1,152.00 (+516.44%, issue price RMB 186.88), with turnover of RMB 7.272 billion and a turnover rate of 81.22%. Its Eastmoney sector classification was verified as optical & optoelectronics (BK1038), not semiconductor equipment; and its main-force net for the day is recorded as roughly 0 under the Eastmoney basis (newly listed stocks do not participate in fund-flow classification on day one), so it did not contaminate any row in the tables above. Note, however: +516.44% is a new-listing pricing phenomenon and should not be read as a prosperity signal for the optical & optoelectronics sector — that sector's +RMB 1.958 billion net inflow was contributed by BOE Technology A (000725, +RMB 886 mn) and IRICO Display (600707, +RMB 627 mn); it is panel money.

5.4 Single-Stock Main-Force Flows

Top 12 net inflows: BOE Technology A (000725, SZSE main) +RMB 886 mn / IRICO Display (600707, SSE main) +RMB 627 mn / Lion Microelectronics (605358, SSE main) +RMB 612 mn / Agricultural Bank of China (601288, SSE main) +RMB 580 mn / Piotech (688072, STAR) +RMB 548 mn / Lens Technology (300433, ChiNext) +RMB 536 mn / Xinqi Microelectronic Equipment (688630, STAR) +RMB 528 mn / AMEC (688012, STAR) +RMB 499 mn / Dekeli Optoelectronics (688205, STAR) +RMB 492 mn / Haitai Optic (688677, STAR) +RMB 475 mn / Tengjing Technology (688195, STAR) +RMB 428 mn / China National Software (600536, SSE main) +RMB 422 mn

Top 12 net outflows: Innolight (300308, ChiNext) −RMB 2.761 bn / Shengyi Technology (600183, SSE main) −RMB 1.838 bn / Eoptolink (300502, ChiNext) −RMB 1.753 bn / GigaDevice (603986, SSE main) −RMB 1.615 bn / ChangXin Memory (688825, STAR) −RMB 1.560 bn / Foxconn Industrial Internet (601138, SSE main) −RMB 1.503 bn / Dongshan Precision (002384, SZSE main) −RMB 1.354 bn / Unisplendour (000938, SZSE main) −RMB 1.262 bn / SMIC (688981, STAR) −RMB 1.194 bn / Hengtong Optic-Electric (600487, SSE main) −RMB 1.161 bn / Victory Giant Technology (300476, ChiNext) −RMB 1.131 bn / Tongfu Microelectronics (002156, SZSE main) −RMB 1.030 bn

Three readings:

1. Yesterday's top main-force net inflow is today near the top of the outflow list. ChangXin Memory went from +RMB 3.678 billion yesterday (1st market-wide) → −RMB 1.560 billion today; GigaDevice from +RMB 1.518 billion yesterday (6th market-wide) → −RMB 1.615 billion today. The mean-reverting character of single-day main-force flow extremes has appeared in full once again, mapping directly onto the self-criticism in note 1 of yesterday's pre-market §2 that "branch ranking is still built mainly on single-day concept nets".
2. Dekeli Optoelectronics (688205) +7.42%, main-force +RMB 492 million. The pre-market brief listed it as grade C (news item 11, "investing remaining over-raised proceeds in new project construction"); against a backdrop of the telecom sector at −1.05% and main-force −RMB 13.575 billion, it was the day's leading gainer in the telecom equipment industry. A grade-C announcement producing a positive return inside an ebbing sector shows that company-level standalone events can break free of sector β — this does not contradict the judgement of "no recommendation slot for optical communications as a whole", but the pre-market brief graded it too low at C.
3. A clear directional switch has emerged inside semiconductors: design/memory out, equipment/materials in. Digital chip design −RMB 7.514 billion versus semiconductor equipment +RMB 868 million; at the stock level Piotech, AMEC, Xinqi Microelectronic Equipment, Konfoong Materials (300666) and Xi'an ESWIN Material-U (688783) all took in money in the same direction. This is the most certain structural change inside today's semiconductor sector.

---

6. Tomorrow's Outlook

The following are observation targets and verification points; they contain no position, buy/sell or stop-loss opinion. Each one also states its falsification direction.

6.1 Main-Line Continuity Calls

6.2 Key Focus Tomorrow (branch + stocks + verification point)

6.3 What to Avoid Tomorrow

1. Today's high-flying boards and failed-seal names. The 4-board Shenqi Pharmaceutical (600613, SSE main) had a seal ratio of only 0.015 and 1 failed seal; the 3-board Xuguang Electronics (600353, SSE main) had 10 failed seals and a seal ratio of 0.003; the 3-board Zhongshi Technology (300684, ChiNext) had 7 failed seals; the 2-board Bomin Electronics (603936, SSE main) had 16 failed seals and Lily Group (603823, SSE main) 17. When a consecutive-board name has double-digit failed seals, the seal was found by repeated trial, not built on genuine absorption.
2. The 13 names that only sealed after 14:25 today. Especially the agricultural ones: Dunhuang Seed (600354), Xinnong Development (600359), Yuan Longping High-Tech (000998), Suken Agricultural Development (601952), SDIC Fengle Seed (000713), Dabeinong (002385), Yiming Food (605179, seal ratio 0.000). Late-session seals usually produce the largest next-day gap-ups and the fastest fades after the open — today's "yesterday's limit-ups" cohort, with an average gap-up of +2.33% and an open→close of −1.12%, is exactly this pattern.
3. Non-executable limit-ups with seal-order/turnover > 1. Guifaxiang (002820) 2.987, Xueda Education (000526) 2.894, Rifeng Cable (002953) 1.783, Fengyuan Pharmaceutical (000153) 1.553, Shenzhen Huafa A (000020) 1.284. A limit-up you cannot buy generates no return and should not be counted in any expectation.
4. Three observation names in the 8/21 interim-report cluster: JCET Group (600584), Fastprint Circuit Tech (002436), Yuan Longping High-Tech (000998). Yuan Longping High-Tech deserves particular attention — it limited up at 14:40 today and is the only name in the hottest main line with an imminent interim report, giving it the greatest two-way elasticity. AMEC (688012) also reports on 8/20.
5. Single-day main-force flow extremes must not be used to rank tomorrow. Today's largest net inflow, BOE Technology A, and largest net outflow, Innolight — on this repository's observations over three consecutive days, the next-day mean-reversion probability for such extremes is significantly higher than the continuation probability. The lists in §5.4 describe where money went today; they are not a forecast of tomorrow's strength.
6. Lone survivors and branches without a ladder. Huaya Intelligent (003043, SZSE main) was today's only semiconductor limit-up, with no follow-on names in the same sector.

6.4 Input Notes for Tomorrow Morning's Pre-Market List (7 items, ranked by importance)

1. [Most important] Add a "pricing progress" dimension alongside the "publication window". Today's lesson: the J.P. Morgan report (published 8/16) genuinely fell outside the scan window and window discipline was executed faultlessly, but at that moment it was in day 1 of pricing. Proposal: for catalysts outside the window, stop downgrading them directly and add one question — "was it already being priced on the tape yesterday?" Use the sector distribution of yesterday's limit-up pool as the criterion, not the news publication date. Yesterday's limit-up pool already contained a substantial number of agriculture-and-food-chain names; that information was in hand pre-market and went unused.
2. List "the name with the strongest seal quality" separately, regardless of which branch it belongs to. Yesterday's pre-market brief had already computed that Jinjian Cereals' (600127) seal ratio of 0.629 was the strongest first board of the day and that its Dragon-Tiger net buying share of 29.86% was the highest of the day, but that record was buried at position 20 under "watch only". Proposal: have the pre-market brief always output one fixed line: "yesterday's 3 names with the best seal quality / highest Dragon-Tiger concentration are ___, and they belong to the ___ branch" — even if that branch is rated very low.
3. Switch all cross-market verification points to closing readings, and state in the criteria table that "open readings must not be used for conclusions". Today SK Hynix opened +5.53% (triggering "confirmation") but closed +1.03%, with Samsung −2.19% and KOSPI −1.55% (which should have triggered "downgrade"). The read moment specified pre-market was 5–30 minutes after the open, and it gave the opposite signal.
4. Backfill the 8/18 Dragon-Tiger List (published after 18:00) and break down the seats. Focus: Shennong Seed Industry (300189), Jinlu Electronics (301282), Tianshan Biology (300313), Luoniushan (000735), Zhongshi Technology (300684). Determining whether the agriculture line is institutional accumulation or hot-money relay can only be done through seat-level breakdown.
5. Verification points must be falsifiable. Today's verification point for GigaDevice (603986) was "does the sector rise while it lags" — in fact the sector fell and it fell too, so the verification point cannot be adjudicated. Proposal: write all future verification points in the form "is X > threshold", and prohibit compound formulations that depend on a precondition holding.
6. Medium-term judgements must not be used as intraday rankers. Two errors of the same type today: ① "equipment is the supply side" → all three equipment names tagged watch only, while equipment was the only semiconductor sub-sector with net inflow; ② "JCET has the best profit quality" → turned into the intraday criterion "does its gap-up today overtake Tongfu's", which was falsified. Proposal: split branch ranking and single-stock ranking into two separate tables, use financial quality only for the latter, and label it explicitly as "medium term".
7. Tomorrow's calendar confirmed (source: the exchanges' scheduled interim-report disclosure calendar): 8/19 GigaDevice (603986), Puya Semiconductor (688766), Suken Agricultural Development (601952) (75 companies market-wide); 8/20 AMEC (688012) (225 companies); 8/21 JCET Group (600584), Fastprint Circuit Tech (002436), Yuan Longping High-Tech (000998) (200 companies).

---

⚠️ 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 basis; please rely on exchange/company disclosures, and do not use this directly as a basis for trading.

Data Sourcing and Run Log (not sent to clients)

1. Interface Status and Data-Sourcing Failures

• ⚠️ Dragon-Tiger List not obtained (not a fault, a scheduling problem). This report was generated at 15:30, while the exchanges publish the Dragon-Tiger List around 18:00. ak.stock_lhb_detail_em(start_date='20260818', end_date='20260818') returned TypeError: 'NoneType' object is not subscriptable — this error looks like a broken interface but is in fact the day's list not yet being published. The gap has been disclosed honestly to clients in §5.2. This is a structural problem of the recap slot: an A-share recap run at 15:30 can never obtain that day's Dragon-Tiger List.
• ⚠️ The first pull of the limit-down pool hit a silent failure. Requesting getTopicDTPool with sort=fbt:asc returned tc=5 while len(pool)=0 — a non-zero tc with an empty pool is the silent-failure pattern this repository has recorded before. Switching to sort=fund:asc gave a consistent tc=5 == len(pool)=5. All three pools had the tc == len(pool) check executed: limit-ups 79/79 ✅, limit-downs 5/5 ✅, failed seals 23/23 ✅.
  • Also noted: at the first pull at 15:32 there were 4 limit-downs; a re-pull at 15:45 gave 5 (Jianghan New Material 603281 added). Data is still settling after the close; the body adopts 5.
• push2.eastmoney.com intermittently returned 502; a push2delay.eastmoney.com fallback was added to all clist requests, successfully.
• push2his.eastmoney.com K-line interface refused connections repeatedly (RemoteDisconnected), and ak.stock_zh_index_daily_em failed likewise. The daily sequential turnover series for the two exchanges was therefore not obtained; the body instead uses "morning half-session RMB 1.6 trillion (Sina) vs full day RMB 2.42 trillion (self-computed)" as evidence of shrinking volume, and that inference is robust to the rounding direction of 1.6 (even if the morning were 1.69, the afternoon would still be only RMB 0.73 trillion).
• None of the CNBC single-stock Korean symbols could be retrieved (000660.KS-KR, 000660-KR, 000660.KS, SKHYY all returned empty); stooq 000660.kr returned 404. Closing prices for SK Hynix and Samsung were instead cross-confirmed from two independent sources (see below).
• ak.stock_report_disclosure raised ValueError: Length mismatch; switching to Eastmoney's RPT_PUBLIC_BS_APPOIN successfully obtained 500 scheduled-disclosure records for 8/19–8/21.

2. Execution of the Full-Universe Basis (guarding against one-sided pagination bias)

• Single-stock quotes: declared total=5,899, actually retrieved 5,899, complete. The 2,121/3,292 advance-decline count is a full-universe statistic, not an extrapolation from the top N of a sorted list. Cross-check: Sina's 8/18 closing commentary said "more than 3,200 stocks declined", consistent with 3,292.
• Single-stock fund flows: declared total=5,560, actually retrieved 5,560, complete.
• Industry sectors 496/496 and concept sectors 504/504, both complete.

3. Back-Calculation and Cross-Check Log

4. Two Data Traps Detected

1. ⚠️ A search summary labelled the 8/14 Korean session as 8/18. An English search returned "On August 18, 2026, the KOSPI climbed 2.42% to close at 6,978. SK Hynix jumped 3.26%". But CNBC's .KS11 previous_day_closing is precisely 6,977.94 — i.e. 6,978 is the close of the previous trading day (8/14, since Korea was closed on 8/17), not 8/18. Taking that summary at face value would have produced the exact opposite conclusion, "Korea's big rally confirms memory", and would have inverted today's entire scorecard. What caught it was the previous_day_closing field returned by CNBC, not news verification.
2. ⚠️ Estun Automation (002747) closed at the limit price but is not in the official limit-up pool. Previous close 36.40 → close 40.04, exactly the limit price, with f3 = 10.0; but it sits in the failed-seal pool (first seal 14:56, 1 failed seal), not the limit-up pool. The official pool's tc=79 check passed, so the body uses the official-pool basis of 79 throughout and did not add it back on a "closing price = limit price" test. This discrepancy is recorded for later reconciliation — self-screening by threshold would have produced 80 today, inconsistent with the official basis.

5. Not Done / Not Obtained in This Report

• No sub-agent was invoked (this run explicitly required not using the Agent tool). This report therefore has no valuation back-calculation from fundamentals-analyst and no red-flag list from risk-auditor — per this repository's records, the most serious error ever made in the pre-market slot (the PE basis) was precisely one the main slot could not find itself and that was exposed only by a sub-agent's back-calculation. All self-checking here was done by the main slot and should be discounted accordingly.
• The Dragon-Tiger List is missing, so the buy-side structure of Shennong Seed Industry (300189), Jinlu Electronics (301282) and Tianshan Biology (300313) cannot be adjudicated by seat breakdown.
• No same-day catalyst was found for the panels / optical & optoelectronics surprise main line. This is explicitly flagged in §6.2 item 6 as "no verifiable catalyst found, no strength grade given"; no attribution was fabricated.
• The daily sequential turnover series was not obtained (push2his refused connections); only half-day versus full-day was used to argue shrinking volume.
• Whether Pinzhun Laser is included in the optical & optoelectronics sector's change calculation, and at what weight, was not verified. It has been confirmed that its main-force net is recorded as roughly 0 and does not contaminate the nets; the contamination magnitude on the sector change is estimated at < 0.03pct (free-float market cap of RMB 8.776 bn / sector scale), so the conclusion is insensitive to this item.

6. Methodology Notes (worth recording)

1. A new failure mode appeared today: setting the wrong "read moment" for a verification point is more insidious than getting the conclusion wrong. The pre-market Korea verification point had a well-designed criteria table (three brackets, with a falsification direction), but the read moment was fixed at 5–30 minutes after the open, and that instrument opened +5.53% and closed +1.03%, an open→close of −4.26%. The same criterion gives opposite conclusions at the two moments, and the pre-market brief specified only one moment. This is the same mechanism as this repository's "pre-market volume and price accumulate; a single snapshot is not a fact", recurring in a cross-market setting.
2. The "window discipline" rule itself produced a systematic miss. It measures the publication date; the market prices on trading days. Proposal: make "pricing progress" (was the catalyst already visible in yesterday's limit-up pool) a second dimension independent of the publication window, otherwise any theme that ferments across days will be structurally underrated — and the overwhelming majority of A-share themes ferment across days.
3. The two-column scorecard basis (close chg / open→close) produced a substantive difference today: a headline gap of 3.39 percentage points versus an executable gap of 1.33 percentage points. Reporting only the former would exaggerate the severity of "the Pass group outperformed" to 2.5 times the actual level, leading to over-correction of the pre-market methodology. Both columns must be permanently retained.