A-Share Market Recap | 2026-08-17 Monday

0. The Recap in One Line

Sentiment euphoric: 106 limit-ups against 1 limit-down, 4,335 advancers against 1,064 decliners, and not one of the 100 industry sectors fell; yet today's strongest main line was memory/semiconductors — and that line got not a single word in this morning's pre-market list.

All 16 names on the pre-market list rose, and their average also beat the broad market — but pull out the "watch only + Pass" groups and compute them separately, and they beat the recommended group. On a day when buying with your eyes closed still paid, the direction of stock selection was right, but the direction of the ranking was inverted.

The one big win was 中微半导 (Zhongwei Semiconductor, 688380) +16.81% — but the reason given pre-market was "earnings α," while the reason it rose today was "semiconductor sector β"; the pre-market note even wrote up "an unfavorable capital environment in the semiconductor sector" as one of its risks. Right name, wrong reason.

Tone for tomorrow: the main line sits in memory/semiconductors, sentiment runs hot but the limit-up ladder is extremely thin (91 of the 106 limit-ups were first-day seals, and the highest streak was only 4 consecutive limit-ups) — this is a liquidity-driven broad rally, not a theme relay.

---

1. Market Overview

Combined turnover across both exchanges was about RMB 2.39 trillion. Data source: East Money market data interface, closing snapshot timestamp 2026-08-17 15:30 (checked, not lagged).

Sentiment indicators:

Sentiment temperature: euphoric. But one qualifier has to be added immediately — breadth extremely high, height extremely low: of the 106 limit-ups, 91 were first-day seals, 10 were second consecutive, 2 third, 3 fourth. This is the shape of "money lifting everything indiscriminately," not the shape of a theme relay. The limit-up count doubled while the highest streak fell from 5 to 4 consecutive, which says incremental money is diffusing rather than building a ladder.

Both of last Friday's "norms" reversed today: ① Risk 3 in §8.4 of the pre-market list wrote "drifting lower after the open is the current norm (only 40.7% positive open→close)"; today positive open→close was 77.8%; ② pre-market Risk 4 wrote "the relay effect has deteriorated for several sessions running (yesterday's consecutive-limit-up index −2.23%)"; today the "yesterday's limit-up streak" concept was +4.51%, "yesterday's second-or-higher seals" +5.98%, and the 63 names that hit limit-up yesterday averaged +3.55% today, with 18 still at limit-up and 41 up against 21 down. These two reversals are the shared backdrop of every reconciliation miss today.

---

2. Pre-Market List Reconciliation

Basis note (important): the pre-market list is published before the open, so the execution price for anyone acting on it is today's opening price. This table therefore uses open→close as the executable return, with the closing change shown for reference only. The gap between the two was enormous today (e.g. 星网锐捷 closed +3.80% but open→close was only +0.59%).

2.1 Recommended (priority deep-dive / watch closely) — 9 names

2.2 Watch Only — 6 names (judged "weak"; what actually happened)

2.3 Pass List — 14 names

2.4 Hit Rate and Self-Critique

Directional hit rate: 9/16 = 56.3% (recommended names beating the market count as correct, 6/9; watch-only names lagging count as correct, 3/6; the Pass group is counted separately).

⚠️ The single most important reconciliation conclusion in this report

The Pass group (+5.30%) beat the recommended group (+2.95%) by 2.35 percentage points, and the "watch only" group (+3.14%) also beat the recommended group.

All 16 names on the list rose and the set as a whole beat the market, which says the selection universe was right — everything landed inside today's strong zone. But the internal ranking was a negative contributor: the more a name was judged "weak," the better it did today.

This is not chance. The pre-market ranking was based on "statement quality + expectation gap," while the market priced on "elasticity + sector membership" today. The Passed names — 金禄电子 (gross margin 15.13%), 芯朋微 (ex-non-recurring −53.30%), 长盈通 (PE 764) — happen to be low-priced, high-elasticity small caps, and on a day with 106 limit-ups and all 100 industries up, they were exactly what money lifted first.

Last month's lesson, "financial quality is not a short-term ranker," was validated again today — and this time the list doing the validating was one I wrote myself.

Four specific criticisms:

1. The window definition excluded the day's strongest main line entirely (root cause, see §3).
2. "It will be misread" was treated as a reason to avoid — when it was precisely the reason to be long. The pre-market text on 太辰光 read: "the market today will most likely only read 'optical communications + RMB 720 million capacity expansion,' while the interim report's ex-non-recurring figure and cash flow point the other way. This is the name this report considers most liable to be misread." — That prediction was completely accurate, and it hit +20% limit-up today. I predicted market behavior correctly and then wrote it up as a bearish conclusion. Short-term pricing turns on how the narrative gets read, not on the financial statements. Next time a phrase like "I judge the market will misread this" appears, it must be paired with the flip-side note "if the misreading happens, this stock probably goes up," rather than being flagged one-directionally as a risk.
3. The size of the gap-up is not a strength indicator — the sign is arguably inverted. The bullish gate §8.1 of the pre-market note set for 中微半导 was "a gap up of +5%~+12%"; it actually opened only +4.34% (below the bottom of the gate), yet closed +16.81% for the day. Conversely, the biggest gap-ups (星网锐捷 +3.18%, 盛科通信 +4.87%, 杭电股份 +3.62%) delivered open→close of only +0.59%, +0.89% and −0.81% respectively. On trend days, it is the flat- or lower-opening names that have room for the whole session; the gap-up names cash their gains in advance.
4. The verification point is more reliable than the conclusion — this holds for the third time today. The bearish signal §8.1 of the pre-market note set for 亨通光电 was "an open of < +2% = last Friday was a single-day pulse"; the actual open was +1.30% and main-force net outflow was RMB 919 million (3rd largest in the market) — the signal was right, while the same report's conclusion on it was "watch closely." Likewise, the gate set for 盛科通信 — "only an open of > +10% counts as buying old news as new" — was not triggered, and that call was right. Keep the order: write the verification point first, then reason back to the conclusion.

---

3. Main-Line Verification

3.1 Branch Reconciliation

3.2 The Real Main Line Today: Memory/Semiconductors

Leading stocks (all large caps, highly recognizable):

Drivers (primary-verified timestamps):

• Published by Cailianpress 2026-08-14 12:00: SK Hynix chairman Chey Tae-won said in an exclusive interview at the company's Korean headquarters that "demand right now is explosive," that customers are asking for close to twice the previous supply volume, that capacity expansion takes 4–5 years, and that 2027 may bring the most severe memory shortage in history. Cailianpress¹
• 2026-08-07: SK Hynix announced new fabs in Yongin and Cheongju, with total investment of KRW 54 trillion (about US$38.4 billion). Cailianpress²

🔴 Root cause: the pre-market time-window definition kept the day's strongest main line out of view

§1 of the pre-market list defined the scan window as "2026-08-14 (Friday) 15:00 close → 2026-08-17 (Monday) 07:00," and scanned 1,915 announcements inside that window.

But the core catalyst for the memory line was published 8/14 12:00 — three hours before the window's start.

It therefore slipped through two sieves at once: ① it was outside the time window; ② it was not an A-share announcement but an executive interview at a Korean company, while the pre-market scan targets "Shanghai/Shenzhen/Beijing A-share announcements." A window anchored at the closing bell systematically discards information that occurs during the session but only gets fully priced on the next trading day.

This and last week's miss on 卫星化学 are two forms of the same class of error: that time it was "slice the narrative first, then pick stocks"; this time it was "draw the window first, then scan." Neither was a data-retrieval error — in both cases the screening framework itself determined what could not possibly be seen.

The pre-market note was even holding the counter-evidence and read its direction backwards: it stated that "on 8/14 the semiconductor sector saw main-force −RMB 6.125 billion and electronics −RMB 6.470 billion, the two largest net outflows in the market — an unfavorable capital environment for the sector," and on that basis listed this as a risk for 中微半导. Today electronics was +RMB 30.946 billion and semiconductors +RMB 19.548 billion, the two largest net inflows in the market. This is one more validation of "a single-day main-force flow extreme is a contrarian indicator" — the day's outflow extreme was treated as negative evidence for the next day, when it was in fact the marker of the next day's strongest direction.

33 中微半导688380Dedicated Review: Right Name, Wrong Reason

• Actual performance: +16.81% (STAR Market ±20%, no limit-up), open→close +11.94%, turnover rate 7.39%, turnover RMB 1.317 billion.
• Reason given pre-market: "the only company in this window whose actual figures clearly exceeded its earnings pre-announcement (+4.2%), ex-non-recurring +109.74% above net profit attributable to parent, gross margin up to 38.75%" — earnings α.
• Risk given pre-market: "its downstream is home appliance/automotive MCU, not the AI chain; and on 8/14 the semiconductor sector saw main-force net outflows, an unfavorable capital environment for the sector."
• What actually happened today: the company disclosed only its 2026 interim report that day (already covered in the pre-market list, no new announcements). Meanwhile digital chip design rose +5.00% as a whole with main-force net inflow of RMB 10.193 billion, and semiconductors +4.90%, +RMB 19.548 billion. It was lifted by the entire semiconductor sector — and it was that sector's leading gainer.
• Conclusion: the earnings-α judgment picked the right name, but most of today's 16.81% came from the very sector β the pre-market note judged "unfavorable." Had the β judgment been taken at face value and the name downgraded on it, this stock would have been screened out by its own analysis. This is the archetypal "right conclusion, wrong mechanism" sample in this report, and should not be recorded as a clean success.

3.4 Two Passed Names Whose Drivers Are Now Traceable (the pre-market "driver unidentified" is resolved today)

• 中石科技 (Zhongshi Technology, 300684) (ChiNext ±20%): today +20.00% on a straight-line seal (2 consecutive), seal orders RMB 860 million / turnover RMB 294 million, seal-order-to-turnover ratio 2.928, first seal 09:25:00 — the best seal quality in the market. Driver now identified: the 8/13 announcement that the controlling shareholder, the actual controller and their concert parties intend to transfer part of their shares by agreement, with an accompanying change of interests. The pre-market note listed it as "driver unidentified, used only as a seal-quality benchmark"; that driver did in fact exist and was findable — this was a missed check. Note: a straight-line seal cannot be bought, open→close 0.00%.
• 蓝盾光电 (Landun Photoelectronics, 300862) (ChiNext ±20%): today +15.82% (no limit-up, streak broken), open→close +11.32%. Driver now identified: on 8/08 it disclosed a restructuring plan and the resumption of trading in its shares, followed by abnormal-fluctuation and serious-abnormal-fluctuation risk warnings on 8/11 and 8/13. The pre-market note judged it "the worst seal quality among the high-streak names in the market" purely on "seal order/turnover 0.01, 5 seal breaks, turnover rate 51.69%" and Passed it, without checking its trading suspension/resumption or restructuring matters. Seal quality is a trading metric and cannot substitute for verifying the nature of the catalyst — restructuring names inherently have poor seal quality (arbitrage flow keeps cashing out), so measuring them with the same ruler produces systematic misjudgment.

---

4. Limit-Up Ladder / High-Streak Names

Ladder structure (official limit-up pool of 106 names, tc == len(pool) verified = 106 == 106):

All names at 2 consecutive limit-ups or more (15):

Four readings:

1. Height is falling while breadth explodes. The maximum was 4 consecutive limit-ups (last Friday it was 5), and the three names at 4 consecutive belong to plastics, medical services and decoration & renovation — not one of them belongs to today's strongest memory/semiconductor line. The main line is driven by large-cap money and does not travel through the limit-up ladder; what travels the ladder are junk names unrelated to the main line. Those two systems were disconnected today.
2. Seal quality was broadly poor. Of the 15 consecutive-limit-up names, only 2 had a seal-order/turnover ratio > 0.5 (中石科技 2.928, 桂发祥 2.378); the other 13 were all below 0.3. 金螳螂 broke its seal 9 times and 共进股份 11 times yet still closed sealed, which means the seal was forced with money after repeated prying, not locked-up float.
3. 中瓷电子 sealed in the afternoon for the second straight day (last Friday 14:26:30, today 13:21:27), and still no company announcement of any kind can be found today — the most recent remains the 2026-07-29 cash management progress filing. The "catalyst unidentified" item still stands after a second limit-up day, and per last month's lesson, two consecutive limit-ups with no announcement more likely reflect industry information not yet public or pure capital behavior, and should no longer be filed simply as "no news."
4. The main line's core troops are not in the limit-up pool but among the big names: 长鑫科技 (688825, +12.00%, turnover RMB 33.518 billion), 中际旭创 (Zhongji Innolight, 300308, +6.15%, turnover RMB 35.274 billion) and 兆易创新 (603986, +6.35%, turnover RMB 25.710 billion) are the most recognizable core troops of the day.

---

5. Capital Flows

5.1 Sector Main-Force Net Inflows (East Money basis, at close)

Net outflow side (worth noting): the concepts "yesterday's limit-ups" −RMB 3.376 billion, "yesterday's limit-ups incl. straight-line seals" −RMB 3.179 billion, "yesterday's first seals" −RMB 2.956 billion. Yesterday's limit-up names rose in price overall (+3.55%) but saw net outflows — profit-taking cashing out into strength, not new money relaying.

5.2 Single-Stock Flows

Top 6 main-force net inflows: 长鑫科技 (688825) +RMB 3.678 billion, 通富微电 (002156) +RMB 2.264 billion, 中际旭创 (300308) +RMB 2.087 billion, 京东方 A (BOE Technology, 000725) +RMB 1.568 billion, 长电科技 (600584) +RMB 1.561 billion, 兆易创新 (603986) +RMB 1.518 billion. All are large-cap semiconductor/optical module names.

Top 6 main-force net outflows:

⚠️ One structural point that must be spelled out: among the market's top 6 main-force net outflows, 2 are recommended names from this morning's pre-market list (亨通光电, 生益科技), and 1 more is the core verification target from the pre-market Pass list (网宿科技). The leaders of the two main lines judged "strong" pre-market (copper-clad laminate, optical communications) were both being distributed today — only the broad rally papered it over with a green close.

53 网宿科技300017The Pass Was Right on Substance, Wrong on Price

The pre-market note made it the core verification target for "strong last Friday ≠ strong today," and set a verification point in §8.1: "a gap down of > −8% would mean the market is already pricing the negative-growth interim report; a gap up or flat open would mean the RMB 2.045 billion of money from 8/14 is still holding on."

Actual: opened −1.90% (open 17.00, previous close 17.33), closed +1.38%, open→close +3.35%. By the pre-market note's own criterion this counts as "close to a flat open" — i.e. that 8/14 money is indeed still in the name.

But today's main-force net outflow was RMB 990 million, the largest in the market. Combining the two readings: the price did not fall, but the money is leaving. About 48% of last Friday's +RMB 2.045 billion net inflow was given back today, while the share price only eased to +1.38% (on a day when ChiNext was +3.14%, a relative lag of 1.76pct).

Conclusion: the Pass judgment holds in substance (money leaving, relative underperformance), but the price expectation that "a negative-growth interim report will be punished immediately" did not materialize — on a day with 4,335 advancers, fundamental bad news was diluted by the broad rally. That is itself an important reading: in a phase of euphoric sentiment, the pricing of a refutation gets postponed, not cancelled.

5.4 Data That Could Not Be Obtained (stated plainly)

• Northbound flows: daily net buying has not been disclosed since 2024-08-19, so this report provides no such figure. The "420.0" returned by the interface is a quota ceiling, not an inflow figure, and must not be cited. Today Shanghai Connect names were 1,205 up / 375 down and Shenzhen Connect names 1,476 up / 362 down (advancer/decliner counts of eligible names only, not a capital-flow measure).
• Dragon-tiger list: the exchanges publish it after 18:00 each day; this report was generated at 15:30, so today's dragon-tiger data is not yet published and cannot be included here. Last Friday's (8/14) dragon-tiger list was used in this morning's pre-market list. Tomorrow's pre-market should backfill today's dragon-tiger list, focusing on the seat structure of five names: 太辰光, 中瓷电子, 金禄电子, 芯朋微, 长盈通.
• Southbound net buying today was RMB 3.265 billion (Shanghai RMB 1.276 billion + Shenzhen RMB 1.989 billion). The read timestamp is 15:30, when the Hong Kong market had not yet closed, so this figure is an intraday reading, not a full-day final value.

---

6. Tomorrow's Outlook

6.1 Main-Line Continuation Assessment

6.2 Key Watch List for Tomorrow (with verification points)

Per last week's lesson, write the verification point first and let it determine the conclusion. All of the below are observation targets and do not constitute recommendations.

6.3 What to Avoid Tomorrow

1. High-streak junk limit-up names as a whole. The 4-consecutive names 天洋新材 (603330), 澳洋健康 (002172), 金螳螂 (002081) — none of the three has anything to do with today's main line — and 金螳螂 broke its seal 9 times today; 共进股份 (603118, 3 consecutive) broke its seal 11 times with a seal order/turnover of 0.011, already the second consecutive trading day with the worst seal quality in the market.
2. Optical communication leaders. 亨通光电 (600487) main-force −RMB 919 million, 中天科技 (600522) −RMB 358 million, 剑桥科技 (603083) −RMB 403 million. The sector index masked distribution in the leaders.
3. Relaying yesterday's limit-up names as a group. The "yesterday's limit-ups" concept saw main-force −RMB 3.376 billion today — prices up, money out.
4. Names locked at a straight-line seal and therefore not executable. 中石科技 (300684) was at a 20% straight-line seal today with open→close 0.00%, and a seal-order/turnover ratio of 2.928 means it will most likely still be unbuyable tomorrow; 桂发祥 (002820) likewise.
5. Pure-concept memory followers with no actual business. All 100 industry sectors rose today; the moment sentiment cools tomorrow, the first to be dumped will be the "touches memory but has no real business" cohort.
6. Macro-sensitive consumption/property chains. See below.

6.4 July Economic Data Released at 15:00 Today (after the close, priced only tomorrow)

The National Bureau of Statistics released the July national economic performance at 2026-08-17 15:00 (the closing bell):

• July value added of industry above designated size +4.5% year on year (June was +5.3%, a sequential deceleration)
• January–July total retail sales of consumer goods +1.2% year on year (first half was +1.3%, still trending down)
• January–July fixed asset investment and property market conditions were released at the same time

Basis note: the above are the figures carried in the headings of the National Bureau of Statistics' 2026-08-17 release page. The single-month July retail sales year-on-year figure and the specific line items for fixed asset investment and property could not be verified item by item in this report and are therefore not cited here, to avoid confusing them with cumulative measures. National Bureau of Statistics data releases³

Implication for tomorrow: the data are broadly soft and pointing down, but because they were released at the closing bell, they were not priced at all today. Note that today's main line (memory/semiconductors) is driven by the global industry cycle and has low sensitivity to domestic macro data; what genuinely comes under pressure is consumption, property and the pro-cyclical chain. If tomorrow brings a structural split of "tech strong, consumption and property weak," it should be read as the lagged pricing of macro data, not as the main line itself weakening.

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

1. ⚠️ Change the scan window definition immediately. The root cause of today's biggest failure was anchoring the window's start at "the previous trading day's 15:00 close." Change to: start from 09:00 of the previous trading day (covering the entire trading session), and separately flag items "published intraday, not fully priced that day." The memory line's catalyst was published 8/14 12:00, only 3 hours before the original window start.
2. ⚠️ The scan universe must include overseas industry sources. Today's main line was driven by an executive interview at SK Hynix (Korea) and appears in no A-share announcement. The pre-market scan should permanently include: earnings guidance, capacity-expansion announcements and public executive statements from overseas leaders (SK Hynix, Samsung, TSMC, Micron, NVIDIA). Per the existing lesson, a foreign supply shock must be checked against the source country's original IR materials; Chinese-language sources corroborating each other are an echo, not verification.
3. ⚠️ Do not use "the day's main-force flow extremes" to rank branches for the next day. Refuted again today: 8/14 semiconductors −RMB 6.125 billion (largest outflow in the market) → 8/17 +RMB 19.548 billion (second largest inflow in the market). A single-day flow extreme is a contrarian indicator.
4. Backfill today's dragon-tiger list (published after 18:00), focusing on: 太辰光 (300570), 中瓷电子 (003031), 金禄电子 (301282), 芯朋微 (688508), 长盈通 (688143). Institutional net buying must be broken down by seat; the total net buy figure is not a substitute.
5. Tracking down 中瓷电子's (003031) catalyst is upgraded to a must-do. Two straight limit-up days + two straight days with no announcement + two straight afternoon seals — this can no longer be closed out as "no news found."
6. Any phrasing along the lines of "I judge the market will misread X" must be accompanied by the flip-side note. Today's 太辰光 (+20%) proves it: predicting a misreading accurately and then flagging it one-directionally as a risk directly forfeits the largest gain.
7. Tomorrow, run the market-wide capital-inflow and gainers rankings first, then slice the narrative branches — this was written into the process last week but still was not executed properly today; today's real main line (memory) sat at the very top of every full-market ranking.

---

⚠️ 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 measurement basis; please defer to exchange/company disclosures. It must not be used directly as a basis for trading.

Data Retrieval and Run Log (not sent to clients)

1. Interface Status

• push2.eastmoney.com intermittently returned 502 again (same cause as on 8/14 and in the 8/17 pre-market run), and both akshare interfaces stock_board_industry_name_em / stock_board_concept_name_em failed. Switched to the clist interface on push2delay.eastmoney.com for industry/concept sectors, successfully.
• The limit-up pool used the official push2ex pool with tc == len(pool) verified: limit-ups 106 == 106, limit-downs 1 == 1, passed.
• Market-wide quote pagination: the total field read 5,896 and 5,896 were actually pulled, no gap. Valid quotes numbered 5,539 (the rest were suspended/without quotes). Pagination was deliberately pulled in full here to avoid the known bias whereby sorted pagination returns only part of the set with the gap skewed toward the decliner tail.
• The quote snapshot timestamp was verified via f86 as 2026-08-17 15:30, not lagged data.
• The dragon-tiger interface stock_lhb_detail_em raised TypeError: 'NoneType' object is not subscriptable — this is an empty return caused by the data not having been published yet, not an interface failure. The exchanges publish after 18:00 and this slot runs at 15:30, so it is structurally unobtainable.
• The northbound flow interface returned net buy amount 0.0, consistent with the known fact (daily net buying disclosure ceased on 2024-08-19). The 420.0 in the fields is a quota ceiling, not an inflow figure, and has been confirmed as not cited.

2. This Report's Core Finding Was Forced Out by the Numbers During Reconciliation, Not Presupposed

The first-draft framing was "pre-market recommendations rose broadly → decent hit rate." Only when the average open→close was computed separately for the Pass group and the recommended group did it emerge that the Pass group (+5.30%) beat the recommended group (+2.95%). Had only the closing changes been reported (recommended +4.66% vs Pass +5.32%, a gap of just 0.66pct), this conclusion would have been heavily diluted — it was the "versus the pre-market/opening price" column that exposed the problem, and that column must be kept permanently.

The root cause of the missed memory line was also worked out rather than assumed: first came the observation "zero pre-market coverage of memory," and only when chasing the catalyst timestamp did it emerge that the Cailianpress piece was published 8/14 12:00, 3 hours before the window start the pre-market run had defined for itself. This is the highest-value item in this report, and it came from primary verification of the timestamp on the URL, not from a search-summary paraphrase.

3. Known Incomplete / To Be Verified

• Today's dragon-tiger list was not obtained (published only after 18:00). The seat structures for all five of 太辰光, 中瓷电子, 金禄电子, 芯朋微 and 长盈通 are missing, and tomorrow morning's pre-market run must backfill them. Every capital judgment in this report about those five rests solely on the "main-force net" measure, unverified against seat data.
• The July economic data were obtained only at heading level from the statistics bureau's release page (value added of industry +4.5%, January–July retail sales +1.2%). An attempt to scrape the detail page directly returned 404 (the URL was constructed by guesswork), so the body text does not cite the single-month July retail sales year-on-year figure or the fixed asset investment and property line items. Tomorrow morning's pre-market run should backfill these from the bureau's official pages.
• The split between "earnings α" and "sector β" within 中微半导's +16.81% is qualitative and no quantitative attribution was done. Strictly, the median gain of comparable companies in the same sector should be used as the benchmark. This report used "digital chip design +5.00%, semiconductors +4.90%" as reference points and concluded that the β contribution was significant without giving a precise proportion.
• No catalyst has been found for 中瓷电子's two consecutive limit-ups. The announcement interface has been checked (most recent filing 2026-07-29); the interactive Q&A platform and investor-visit minutes were not searched.
• The supply/demand direction of the memory main line was not independently verified. SK Hynix's US$38.4 billion capacity expansion (8/07) is fundamentally an increment on the supply side, pointing the opposite way from the demand-side "memory shortage" narrative; the body of this report treats it as an "industry-level catalyst" and did not assess how the expansion, once delivered, suppresses supply-side tightness in 2027–2028. Per the existing lesson, revenue on an equipment/expansion basis speaks to future supply rather than current demand, and must not be conflated with cycle strength. This item is the basis for "semiconductor equipment/packaging & testing needs discrimination" in tomorrow's outlook, but was not developed.

4. Methodology: Two Framework-Level Defects Exposed Today

1. Anchoring the window's start at the closing bell → intraday information is discarded. Already written as a hard rule in §3.2 and §6.5 of the body. This is the same-origin error as last week's "slice the narrative first, then pick stocks": neither was a data-retrieval error; in both cases the screening framework determined what could not possibly be seen.
2. Using financial quality to rank short-term trades → negative contribution. This report is the second empirical case for that conclusion (the previous one was 盛美上海, ACM Research Shanghai). The pre-market ranking basis ("statement quality + expectation gap") and the market's pricing basis that day ("elasticity + sector membership") are not the same function. The suggestion is for the pre-market list to score the two in separate columns rather than combining them into a single total — once combined, "high financial quality but low elasticity" and "low financial quality but high elasticity" end up with similar totals, while their short-term performance is worlds apart.

An honest self-assessment: this report's reconciliation conclusions are fairly damning for the pre-market run (negative contribution from the ranking, zero coverage of the strongest main line), but every one of those points was computed by this slot itself, with no reliance on sub-agents or an auditor. Compared with the 8/17 pre-market report, where "two of three self-reversals came from sub-agents," this one shows improvement in self-refutation. One caveat, though: a recap is inherently easier to self-refute than a pre-market note, because the answers are already public — this does not prove the pre-market slot's self-check ability has improved too.