A-Share Market Recap | 2026-09-17 Thursday

Data read timestamp: all figures are post-close reads taken 2026-09-17 15:30–15:40 (Beijing time); per-stock timestamps were verified one by one as 20260917. Price limits: SH/SZ main board ±10%｜ChiNext (30x)/STAR (68x) ±20%｜BSE ±30%. Every stock in this report is tagged with its board, and limit-up / failed-limit status is judged against its own limit. Limit-up / limit-down / failed-limit: official exchange-standard pools, with tc == len(pool) verified (ZT 47=47, DT 1=1, ZB 20=20). Whole-market snapshot: Eastmoney push2delay, 5,916/5,916 with no pagination gap, 5,554 valid samples. Dragon-Tiger List: the exchange discloses the Dragon-Tiger List after 18:00; this report was generated immediately at the 15:30 close, so today's seat-level data is not included; §5 substitutes net main-force capital inflow. Northbound flows: since 2024-08-19 the exchange has stopped disclosing daily net buy amounts, so this report does not provide that aggregate read.

---

0. One-Sentence Recap

Sentiment turned from euphoric to receding; the pre-market list got the leading theme wrong — and also got wrong the one bearish call it was most confident about.

Today the Shanghai Composite fell −0.41% to close at 3,875.60, with combined turnover of RMB 1.82 trillion. The index decline was small, but at the sentiment level this was an unambiguous retreat: limit-ups were halved from yesterday's 89 to 47, failed limits rose from 11 to 20, the failed-limit rate went 11.0% → 29.9%, the median seal-order/turnover ratio of limit-up stocks collapsed from 0.127 to 0.009 (one-fourteenth), and of yesterday's 89 limit-up stocks only 10.1% advanced to another limit-up.

The three semiconductor/optical-communication branches bet on pre-market all failed together today: semiconductor materials flipped from "31 up, 0 down" to 8 up, 23 down; of 186 semiconductor names 69.4% fell; components (PCB) were −2.25%, second-worst in the whole market; and the CPO concept saw main-force capital of −RMB 3.030 billion. The two genuine leading themes today were the two the pre-market list did not mention with a single word: the auto complex (10 names in the limit-up pool, 21.3% of all limit-ups) and agriculture/crop planting (+5.06%, 21 up / 0 down, strongest in the market).

The single heaviest error: pre-market, the RMB 6.670 billion share unlock at 沐曦股份 (MetaX, 688802, STAR board) was called "the only purely computable bearish item this period, one that depends on no judgment at all," with a directive to avoid the domestic-GPU direction. The actual result was +14.44%, a close near the day's high, and net main-force inflow of +RMB 2.929 billion — the largest in the entire market.

The 5 recommended names averaged +0.26% on the day (excess +0.07pp, i.e. nothing), but on an "open→close" basis — buying at today's open exactly as the list instructed — they were −0.71% (excess −1.19pp). At the execution level the list lost money. Only 1 of 5 verification points passed.

Tone for tomorrow: continuation of the retreat, not a reversal. Shrinking volume, low consecutive-limit heights and poor seal quality all point the same way. Tomorrow one should cut position size, give up on bottom-fishing high-flying tech names, and watch whether agriculture and autos — the two low-base themes that started today — can deliver second-day follow-through.

---

1. Market Overview

Indices and Turnover

Combined turnover of the two markets was about RMB 1.82 trillion.

Sentiment Indicators (vs 9/16)

One Structural Fact: Today Was "Broad Gap-Down + Intraday Repair"

• Only 18.9% of the whole market gapped up (1,051/5,553) — more than eight in ten stocks opened lower, which is the FOMC rate hike being priced in all at once at the open.
• But 55.4% of the whole market was positive open→close, with a mean open→close of +0.47% — after the lower open, the session repaired itself.

Together these two explain "the index only fell 0.4% while every sentiment indicator collapsed": at the index level the rate-hike shock was absorbed intraday. What broke was not the index, it was the relay structure of high-flying theme stocks.

Sentiment temperature call: downgraded from "euphoric" to "receding." The qualitative basis is three indicators that moved in the same direction and are mutually independent — downgraded consecutive-limit height, a nearly 3× failed-limit rate, and seal quality down to 1/14. ⚠️ But this is not panic: only 1 limit-down in the whole market, more than half of stocks positive open→close, and the index down <0.5%. This is the orderly disintegration of high-flying names, not systemic selling.

---

2. Reconciliation of the Pre-Market List

Reconciliation basis (both columns given):

• "Chg% on the day" = relative to the 9/16 close, used for comparison against the index;
• "Open→close %" = relative to today's open, this is the real return of "read the pre-market list, bought this morning as instructed."
• Whole-market benchmarks: equal-weighted +0.19% on the day, equal-weighted open→close +0.47%.

① The 5 Recommended Names

Recommended-group summary:

⚠️ On that "delivered" verdict for 光迅科技 (Accelink), one piece of self-refuting evidence must be added: it closed green +3.24% with both verification points passing, but net main-force flow on the day was −RMB 912 million, the second-largest main-force outflow in the recommended group (behind only 中际旭创 Innolight at −RMB 920 million), and turnover rate rose from 5.24% to 7.54% (+44%). The criterion set for it pre-market was "turnover rate < 8%"; the actual 7.54% passed with only 0.46pp of headroom. Meanwhile the variable that would actually lose money — main force taking RMB 912 million out while the stock rose — had no criterion set for it anywhere on that list. Conclusion: the verification point passed, but it was not measuring the variable that loses money. This "delivered" is worth less than its surface reading.

② Falsification-Type Pass (pre-market judged "primary evidence negates the bullish item itself, exclude")

⚠️ The concentration that must be stated plainly: the conclusion that the falsification-type Pass group "hugely outperformed" comes 100% from 沐曦股份 (MetaX) alone — excluding it, the group is only +0.30% and open→close is still negative. But this cannot be used to absolve the pre-market list; it sharpens the problem instead: the one name in this group that the pre-market list flagged with its highest-confidence wording — "purely computable, depends on no judgment" — happens to be the single largest reverse in the entire market.

③ Position-Type Pass (pre-market judged "the bullish item may hold, but the price level is not chaseable; cut size, keep watching")

⚠️ An observation that contradicts previous conclusions and must be recorded: the last two recaps both concluded that "position-type Pass names actually outperform on the day" (on 9/16 the excess was +1.65pp with 7/7 outperforming), and the Pass list was split into two columns on that basis. Today that reversed: position-type Pass excess was −1.23pp, and only 4 of the 14 closed green. The mechanistic explanation is self-consistent: position-type Pass holds "high-flying strong names" — on broad-rally days they have the most elasticity; on receding days they fall the most. So "position-type Pass outperforms" is not a rule but a conditional statement that only holds on broad-rally days — and the previous two periods happened to both be broad-rally days. ⚠️ With n=3 and inconsistent direction, this item is insufficient to support any new rule; recorded only, to be continued.

④ Hit Rate and Self-Critique

Four points of self-critique:

(1) The most serious one — "purely computable" was treated as "cannot be wrong." The pre-market wording on 沐曦股份 (MetaX, 688802, STAR board) was verbatim: "this is the only purely computable bearish item this period, one that depends on no judgment at all." The number — RMB 6.670 billion unlocked = 75.38% of free float — was right; the conclusion was inverted. Where it went wrong: it equated "supply volume" directly with "downward pressure," skipping the two steps of "will this supply actually be sold" and "has the market already priced it in advance." Today produced three measured counter-proofs, each the opposite of "selling pressure": ① net main-force inflow +RMB 2.929 billion, largest in the market; ② volume ratio 8.36, turnover rate 44.97% — that is enormous absorption, not enormous escape; ③ opened −3.68%, intraday low −5.54%, and closed near the day's high — the selling pressure was eaten within the first half-hour. The more important mechanism: before the unlock, 沐曦 (MetaX) had only 18.5286 million shares in free float — an extremely thin float. For such an ultra-thin float, an unlock first improves liquidity and tradability, and only secondarily adds supply. The heuristic "unlock = bearish" flips sign on ultra-thin floats. Hard fix for the next pre-market: unlock events may never be written directly as a directional call; three questions must be answered first — ① is the pre-unlock free float extremely small (<5% of total shares)? If yes, default to treating it as "liquidity improvement"; ② how long ago was the unlock announced, and has the stock risen or fallen inside that window (is it already priced); ③ the holders' paper-profit multiple and the nature of the lockup (off-market placement institutions vs major shareholders). If all three cannot be answered, it may only be written as a "watch item," never as "avoid."

(2) The leading-theme call was wrong overall, and the evidence was already in hand at the time. The three pre-market branches (semiconductor materials, optical communications, semiconductor equipment) all failed today, while the genuine leading themes were the auto complex and agriculture/crop planting, not mentioned pre-market with a single word. But the official 9/16 limit-up pool (89 names, already obtained pre-market and used to compute base rates) contained: 万向德农 (Wanxiang Doneed, 600371, SH main board, crop planting) on its first board, and 3 auto-parts names at limit-up. 万向德农 (Wanxiang Doneed) advanced to a 2nd consecutive limit-up today, and crop planting was +5.06%, 21 up 0 down, strongest in the market. This is the third repetition of the same failure mode: the limit-up pool is pulled only to count "how many names fall inside the branches I already care about," and never used to answer "which industry holds the strongest cluster in the market." Hard fix for the next pre-market: the industry distribution of the limit-up pool must be laid out as a full table and set as a mandatory item, answering at minimum — ① which industry has the most limit-ups in the market ② which industry does the highest consecutive-limit stock in the market belong to ③ do these two answers fall inside my branch ranking; if not, a reason for "why not recommended" must be written out. No reason written = missed detection, not "already excluded."

(3) Right on the single names, wrong on the portfolio — the act of "downgrading" itself needs to distinguish two cases. The reasons were all true for downgrading/passing 港迪技术 (Gangdi Technology, 301633, ChiNext) and 天孚通信 (T&S Communications, 300394, ChiNext) (港迪's H1 did swing from profit to loss; 天孚's Q2 revenue was indeed −0.9%), yet the two rose +10.33% and +3.41% respectively on the day. By contrast, the downgrades of 有研新材 (GRINM Advanced Materials, 600206), 中微公司 (AMEC, 688012) and 中瓷电子 (Zhongci Electronic, 003031) were entirely correct, with the three at −2.09%/−0.39%/−0.23% on the day. Where the difference lies: the three judged correctly were "extreme price level + weak fundamentals" (有研新材's PB is in the 99th percentile of its full history and it was sitting at a high today); the two judged wrongly were "weak fundamentals but the price has already fully corrected" (港迪 was already down 23.4% since 9/7; 天孚 is a laggard). Hard fix for the next pre-market: fundamental falsification can only negate the "logic," it cannot negate the "price level." When a stock simultaneously satisfies weak fundamentals + price already substantially retraced + news not yet priced (volume ratio <1), the conclusion should be written as "not recommended but kept under watch," and must not be placed in the same Pass column as high-flying names — today those two columns were mixed together, which directly caused the portfolio-level misjudgment.

(4) One call that was right, and the best call made today: precious metals. The pre-market explicitly wrote: "the reason for exclusion is price level (13 up 0 down consistency is already an extreme), not overseas falsification — GLD was only −0.61% overnight, and gold is actually on the firm side." Today precious metals were −5.35%, 13 names 0 up 13 down, the weakest industry in the market; 紫金矿业 (Zijin Mining, 601899, SH main board) was −3.95% with net main-force outflow of −RMB 1.183 billion. This one deserves to be recorded separately because its reasoning chain and its conclusion stand independently: it did not use one-step attribution of the "overseas fell, therefore A-shares must fall" kind (that link had already been falsified by measurement at the time); it used the purely structural criterion "consistency has reached an extreme" — and today happened to be precisely the day when overseas markets were no longer the cause and the structure ran out of room on its own.

---

3. Verification of Today's Leading Themes

⚠️ Rulings on the Three Pre-Market Criteria (item by item)

⚠️ Honest disclosure on basis: pre-market it was noted that "the constituent basis for the semiconductor-materials criterion is not fixed; the close-of-day review must use the names listed in §7 branch 1," and §7 listed only 3 names. This report gives both bases:

• Using Eastmoney BK1325 (n=31, exactly matching the sample size behind the pre-market "31 up 0 down"): 8 up / 23 down, criterion breached;
• Using the 3 names listed in §7: 有研硅 (GRINM Semiconductor Materials) +1.74%, 有研新材 (GRINM Advanced Materials) −2.09%, 立昂微 (Lion Microelectronics) +0.89% (all three negative open→close: −3.51%/−4.62%/−0.83%). Neither basis supports continuation of the branch.

Optical Communications: A Split That Must Be Taken Apart

The sector as a whole is falling while the leaders rise —

How to read it: this is not "branch continuation," it is capital withdrawing from the entire branch and making a final contraction into four large-cap names. Of 89 communications-equipment stocks, 61 fell, while capital concentrated into the leaders with turnover above RMB 10 billion. 光迅科技 (Accelink) rose 3.24% yet had net main-force outflow of RMB 912 million — the price is rising, the main force is leaving. Stage verdict: the peak has passed; entering divergence. This is directionally consistent with the pre-market judgment that "it has entered the late stage of second-tier catch-up," but the pre-market list still ranked it #2 and gave it a recommended slot; that position decision contradicts its own stage judgment.

⚠️ The Two Leading Themes the Pre-Market Missed Entirely

⚠️ One cooling note that must be added on the auto complex: the 10 auto names at limit-up today look structurally more like low-base capital rotation than industrial drivers — 众泰汽车 (Zotye Auto, 000980, SZ main board), 金杯汽车 (Jinbei Automotive, 600609, SH main board), 山子高科 (Shanzi High-Tech, 000981, SZ main board) and 安凯客车 (Ankai Bus, 000868, SZ main board) are mostly low-priced/restructuring plays; the only industrial leader, 均胜电子 (Joyson Electronics, 600699, SH main board), did not seal its limit until 13:17, making it a trailing follower rather than the leading core. Seal quality is also broadly weak (北特科技 Beite Technology: seal order RMB 1 million vs turnover RMB 565 million; 克来机电 Kelai Mechatronics: RMB 1 million vs RMB 292 million). This report does not characterize it as "a new leading theme established," only as "the largest capital cluster of the day." ⚠️ The characterization of agriculture/crop planting must also be limited: the pattern of crop planting at +5.06% with 21 up 0 down is the same kind of consistency extreme as semiconductor materials' "31 up 0 down" on 9/16 — and that one collapsed today. By the very "consistency extreme = position risk" criterion endorsed in §2 self-critique (4) of this report, agriculture is already sitting on that criterion's trigger point today. In addition, no new catalyst for today could be located (searches returned only late-August El Niño/grain-price and GMO-commercialization back news), so the specific trigger event for today's rise cannot be confirmed.

---

4. Limit-Up Ladder / Height Boards

Consecutive-Limit Ladder (official pool, 47 names)

Broken Streaks and Advances

• The 9/16 top height of 6 boards, 闽东电力 (Mindong Electric Power, 000993, SZ main board), broke its streak today, −2.76% (repaired from a −9.96% lower open to +7.99% open→close). Pre-market listed it as "position-type Pass, used as a sentiment thermometer," and that usage worked perfectly today — it is the end-marker of this sentiment cycle.
• Advance rate 10.1% (9/89), far below 9/16's 37.5%. The risk threshold set pre-market was "advance rate < 20%," and it has been triggered.
• Recognizability of a lead general: absent. There were 25 limit-up stocks with turnover above RMB 500 million today — not a small number — but not one had the form of a lead general: the top four by turnover — 招商轮船 (China Merchants Energy Shipping, 601872, SH main board, shipping & ports) RMB 3.524 billion, 先导基电 (Xiandao Jidian, 600641, SH main board, semiconductors) RMB 3.026 billion, 山子高科 (Shanzi High-Tech, 000981, SZ main board, auto parts) RMB 2.599 billion, 金健米业 (Jinjian Cereals, 600127, SH main board, agricultural product processing) RMB 2.373 billion — belong to four completely different lines, and their seal-order/turnover ratios are only 0.006 / 0.002 / 0.041 / 0.008 — heavy turnover paired with extremely weak seal orders is "high-volume gaming," not "capital locking up positions." No line formed a complete ladder of "lead general trades heavy volume + same-sector followers."

Ladder conclusion: severe discontinuity. A level is empty between 5 boards and 3 boards (no 4-board name), and the 5-board name itself had 23 failed seals. First boards are 38/47 = 80.9%. This is the classic transitional form of "the old theme is dead and the new theme is not yet established"; it does not support relay trading.

Limit-Downs

Only 1: 海南发展 (Hainan Development, 002163, SZ main board) −9.97%. Whole-market limit-downs went 4 → 1, showing the selling pressure has not spread to ordinary stocks; the retreat is concentrated inside high-flying themes.

---

5. Capital Flows

⚠️ Dragon-Tiger List left blank: the exchange discloses the Dragon-Tiger List after 18:00, while this report was generated immediately at the 15:30 close, so today's seat-level detail is not included. Consequently, the second verification point set pre-market for 有研硅 (GRINM Semiconductor Materials, 688432) ("the Shanghai-Connect seat does not appear in the day's top five sellers") cannot be judged this period. This section substitutes net main-force capital inflow. ⚠️ Northbound flows: since 2024-08-19 the exchange has stopped disclosing daily net buy amounts, so this report does not provide that read.

Industry Net Main-Force Inflow (top 8 each side)

⚠️ Two notes on basis: ① Autos (+RMB 3.708 billion) and auto parts (+RMB 3.040 billion), and electronics (−RMB 12.668 billion) versus components/semiconductors/printed circuit boards, stand in parent-child inclusion relationships and must not be added together; ② general equipment shows an apparent net main-force inflow of +RMB 1.315 billion, but +RMB 865 million of that (65.8%) comes from today's new listing N 沈鼓 (N Shengu, 601091, SH main board, +373.80%); excluding it leaves only +RMB 450 million — a new listing contaminates its sector's capital flow on debut day, so this table does not adopt the general-equipment read.

⚠️ Reversal of Yesterday's Capital Extremes

All five branches flipped sign on capital flow. The primary basis for the pre-market branch ranking was exactly the single-day breadth and single-day capital of 9/16 — and all five reversed today without exception.

Single-Stock Main-Force Capital

Top 8 net inflow:

Top 8 net outflow:

⚠️ The most important contrast in this section: 4 of the 5 names recommended pre-market saw net main-force outflow on the day (光迅 Accelink −RMB 912 million, 旭创 Innolight −RMB 920 million, 有研硅 GRINM Semiconductor Materials −RMB 647 million, 澜起 Montage −RMB 581 million; only 北方华创 Naura was positive at +RMB 213 million); while 沐曦股份 (MetaX, 688802), which the pre-market list explicitly told readers to "avoid," had the largest net main-force inflow in the whole market (+RMB 2.929 billion). Capital gave the answer opposite to this list in the most direct way possible.

Domestic GPU / AI Chips: Judged Pre-Market as a Direction to Avoid, Actually the Strongest Capital Flow Today

• 沐曦股份-U (MetaX, 688802, STAR board) +14.44%, main force +RMB 2.929 billion
• 摩尔线程-U (Moore Threads, 688795, STAR board) +8.68%, main force +RMB 586 million
• AI-chip concept: the index was only −0.05%, yet net main-force inflow was +RMB 2.702 billion — price flat while capital pours in

Control group: fellow semiconductor name 长鑫科技 (ChangXin Memory, 688825, STAR board) −2.96% with main force −RMB 2.110 billion; IC packaging & testing −1.57%, −RMB 1.728 billion. This shows today was not "semiconductors as a whole receiving capital," but capital rotating inside semiconductors out of memory/packaging/materials and into domestic GPUs.

---

6. Outlook for Tomorrow (2026-09-18 Friday)

① Continuity Calls on Tomorrow's Leading Themes

② Tomorrow's Focus (branch + stock + verification point)

⚠️ This section follows the new rule from today's self-critique (1): every verification point must target "the variable that loses money," and must not use criteria with an excessively high base rate such as the day's gain.

③ What to Avoid Tomorrow

1. The entire PCB / components line — the most emphatic item. Components −2.25%, 54 decliners, main force −RMB 8.096 billion; the 5-board 澳弘电子 (Aohong Electronic, 605058, SH main board) had 23 failed seals and a seal-order/turnover ratio of only 0.002; a top height board of this quality usually breaks its streak the next day. 生益科技 (Shengyi Technology, 600183, SH main board), 华正新材 (Huazheng New Material, 603186, SH main board) and 沪电股份 (WUS Printed Circuit, 002463, SZ main board) were all among today's largest net outflows.
2. Semiconductor materials / wafers — the criterion has already been breached (23 decliners vs a threshold of ≤2). 有研新材 (GRINM Advanced Materials, 600206, SH main board) and 有研硅 (GRINM Semiconductor Materials, 688432, STAR board) were −4.62% / −3.51% open→close today, a gap-up distribution pattern.
3. Precious metals and non-ferrous — precious metals −5.35% (13 names 0 up 13 down), gold concept main force −RMB 4.915 billion, industrial metals −RMB 3.426 billion, copper −2.37%. 紫金矿业 (Zijin Mining, 601899, SH main board) had net main-force outflow of RMB 1.183 billion.
4. The low-priced/restructuring names inside today's auto complex — 众泰汽车 (Zotye Auto, 000980, SZ main board), 金杯汽车 (Jinbei Automotive, 600609, SH main board), 山子高科 (Shanzi High-Tech, 000981, SZ main board), 安凯客车 (Ankai Bus, 000868, SZ main board). Seal quality is broadly weak (several with seal-order/turnover <0.02), and there is no verifiable industrial catalyst.
5. High-level leftovers from yesterday's limit-ups — advance rate only 10.1%, failed-limit rate 29.9%. With the ladder this broken, playing the day after a limit-up has a historical base rate (44.8% positive open→close) that is already worse than a coin flip, and today's measured figure was lower still at 36.0%.
6. Lone-runner patterns — today's 47 limit-ups were spread across 33 industries, and apart from the auto complex (10 names) no industry had ≥ 3 limit-ups. Every limit-up outside autos and agriculture should be treated tomorrow as a lone runner.

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

1. [Mandatory item · new] The industry distribution of the limit-up pool must be laid out as a full table, explicitly answering three questions: ① which industry has the most limit-ups in the market ② which industry does the market's highest consecutive-limit stock belong to ③ do these two answers fall inside my branch ranking; if not, a reason for "why not recommended" must be written out. Today and 9/16 were two consecutive occasions where skipping this step caused the strongest theme in the market to be missed.
2. [Mandatory item · new] An unlock event may not be written directly as a directional call. Three questions must be answered first: ① is the pre-unlock free float extremely small (<5% of total shares) → if yes, default to treating it as "liquidity improvement" ② how long ago was it announced, and has the stock risen or fallen inside that window ③ the nature of the holders and their paper-profit multiple. If all three cannot be answered, it may only be written as a "watch item." ⚠️ There is a direct test of the same pattern tomorrow: 金帝股份 (Jindi, 603270, SH main board ±10%) unlocks RMB 1.204 billion on 9/18, equal to 46.67% of its free-float market cap of RMB 2.580 billion, and it already rose 3.72% today on a 17.09% turnover rate — the same "rises ahead of the unlock" pattern as 沐曦 (MetaX). Tomorrow morning's pre-market should use it as a live-market test of this new rule rather than tagging it bearish outright. Total whole-market unlocks on 9/18 come to only RMB 1.855 billion, of which this stock is 64.9%.
3. [Mandatory item · new] The Pass list needs three columns, not two. Today proved that "position-type Pass" mixed two completely different things: high-flying strong names (有研新材 GRINM Advanced Materials, 紫金矿业 Zijin Mining, 生益科技 Shengyi Technology — all judged correctly today) and names that have already fully corrected to a low base (港迪技术 Gangdi Technology, 天孚通信 T&S Communications — both judged wrongly today). The proposal is to add a third column, "low-base awaiting launch (not recommended but kept under watch)," with qualifying conditions: weak fundamentals + retracement from the recent high >20% + volume ratio <1.
4. Branch ranking must not use single-day breadth and single-day capital alone. All five branches flipped sign on capital flow today (CPO went from +RMB 15.720 billion to −RMB 3.030 billion). That single-day capital extremes are a contrarian indicator has now been confirmed by measurement for the second time in this repository. The suggestion is to change the ranking to require "at least 2 consecutive days in the same direction" before counting toward strength.
5. Sentiment baseline for tomorrow (for direct citation pre-market): limit-ups 47, failed limits 20, failed-limit rate 29.9%, highest 5 boards, advance rate 10.1%, median seal-order/turnover of limit-up stocks 0.009, advancer share 46.4%, whole-market positive open→close 55.4%, gap-up share 18.9%.
6. Trigger review of today's pre-market risk signals (for threshold calibration):
   • [Sentiment dimension] ① whole-market positive open→close <40%? measured 55.4%, not triggered ② limit-ups <40 and largest cluster <4? measured 47 names and a largest cluster of 10, not triggered ③ advance rate of yesterday's limit-ups <20%? measured 10.1%, ✅ triggered
   • [Structural dimension] ④ two of the three lead generals turning red? 光迅 Accelink +3.24%, 北方华创 Naura +0.39%, 有研硅 GRINM Semiconductor Materials +1.74%, all closed green, not triggered ⑤ USDCNH breaking above 6.75? measured 6.7070 (−0.08%), not triggered ⑥ 沐曦 (MetaX) falling on heavy volume? measured a heavy-volume surge of +14.44%, the reverse
   • Conclusion: under the pre-market rule of "at least one from each of sentiment and structure," the risk signal was not triggered — and the index did indeed only fall 0.41%. The rule itself is effective.
   • ⚠️ But it missed what actually happened today: the index was fine, the theme ladder collapsed. None of the six risk signals measures "seal quality" or "consecutive-limit height downgrade," and today those two were exactly what broke worst (median seal-order/turnover 0.127 → 0.009, top board 6 → 5). The suggestion is to add these two as the seventh and eighth risk signals in tomorrow's pre-market.

---

⚠️ Data-fetch failures and open items this period (not sent to clients)

1. The first pull of the official limit-up pool failed silently: the sort parameter of push2ex was written as the already-URL-encoded fbt%3Aasc; after requests encoded it a second time, the endpoint returned non-zero tc (47/1/20) but an empty pool array, with HTTP 200 and no error at all. If one reads only tc without verifying len(pool), this report would have lost every single-stock detail while "47 limit-ups" remained correct. It was re-pulled with the unencoded sort=fbt:asc, and tc == len(pool) was verified one by one for the ZT/DT/ZB pools (47=47, 1=1, 20=20). The limit-down pool additionally requires sort=fund:asc; using fbt:asc likewise returns empty.
2. The Dragon-Tiger List is unavailable today (a scheduling problem, not an endpoint failure): stock_lhb_detail_em(20260917) returned TypeError: NoneType — the exchange discloses it only after 18:00, and this report is generated at 15:30. The second verification point set pre-market for 有研硅 (GRINM Semiconductor Materials) (whether Shanghai-Connect appears in the top five sellers) therefore cannot be adjudicated this period, and the body leaves it blank as such. ⚠️ Ops to-do: the A-share recap slot is fixed at 15:30, which means the Dragon-Tiger List dimension is structurally unobtainable forever. This is the second time this repository has recorded the same problem. The suggestion is either to move the A-share recap slot to after 18:30, or to split it in two (15:30 for quotes + reconciliation, 18:30 for an incremental Dragon-Tiger List supplement).
3. The onshore renminbi 133.USDCNYC returned data: None; only the offshore USDCNH at 6.7070 (−0.08%) was obtained. The body uses only the offshore read and makes no claim about the onshore basis.
4. New-listing contamination was intercepted: N 沈鼓 (N Shengu, 601091) listed today at +373.80% with turnover of RMB 2.181 billion, a single name contributing 65.8% of the general-equipment sector's net main-force inflow (+RMB 865 million / +RMB 1.315 billion). Reading the sector table directly would have put "general equipment received RMB 1.315 billion of capital inflow" into the body. It has been explicitly excluded and annotated in §5. This is the second time this repository has hit the same trap; the suggestion is to write "sector capital flow must exclude same-day new listings" into the skill's data-fetch chapter.
5. Date misalignment in search sources (two items, both discarded): ① searching "reasons for auto-parts limit-ups on September 17" returned an intraday broadcast from 09:54 (citing the sector at +0.96%), inconsistent with the close of +1.33% and 10 limit-ups; ② searching "MetaX unlock" returned a Sina piece saying "up more than 5% intraday," a midday read far from the closing +14.44%. Neither was used; the body relies entirely on locally measured closing data. "Searching for the day's closing review returns the opening broadcast" is now the third time this has been recorded in this repository.
6. Today's catalyst for agriculture/crop planting could not be located: searches returned only 8/25–8/31 back news on El Niño/grain prices/GMO commercialization, with no new event found for 9/17. The body honestly states "the specific trigger event for today's rise cannot be confirmed," and no August back news was stitched into a bullish item for today.
7. Two self-interceptions this period:
   • (a) Nearly wrote "the falsification-type Pass group hugely outperformed (+3.13%)" into §2 as a conclusion. Taking it apart revealed only +0.30% excluding 沐曦 (MetaX), and −0.89% open→close; 100% of the group's excess comes from one name. The concentration is now written out explicitly in the body, along with why it does not absolve the pre-market list.
   • (b) Nearly wrote "position-type Pass underperformed today" as "this overturns the rule from the last two periods." In fact n=3, today was a receding day and the previous two were broad-rally days, so the sample is insufficient to support a new rule. It has been downgraded to "recorded, to be continued," with a conditional-statement mechanistic explanation.
8. Conflicting bases are presented side by side rather than chosen between: on the constituent basis for the semiconductor-materials criterion, pre-market wrote "use the names listed in §7 branch 1" (only 3 names) while the threshold assumed n≈31. This report gives both bases: BK1325 (n=31, exactly matching the pre-market sample size) 8 up/23 down; the 3 names in §7 all negative open→close. The two bases reach the same conclusion, so no adjudication ambiguity arose — but the next pre-market must fix the constituent pool code (writing the BK code directly is recommended).
9. The marginal fail on the 北方华创 (Naura) verification point needs human review: +0.39% vs the industry equal weight of +0.4076%, a gap of 0.0176pp. Eastmoney's f3 field carries two decimals, so this gap is already smaller than the field's precision; strictly speaking it is "indistinguishable." The body writes it as "❌ failed (short by 0.02pp)" while also tagging it "⚖️ effectively a tie," and it is not counted in the 1/5 pass tally, nor used as a failure case in the self-critique. If this criterion is reused next period, the suggestion is to compute the gain to four decimals from f2/f18.
10. Explicitly blank items: (a) today's Dragon-Tiger List and seats; (b) northbound daily totals (disclosure discontinued); (c) the day's catalyst for agriculture/crop planting; (d) the day's catalyst for the auto complex (searches returned only background such as CAAM export data from 9/8, with no same-day event); (e) roughly 35 of today's 47 limit-ups have no single-stock attribution; (f) Hong Kong/overseas reads for today were not taken (Hong Kong had not yet closed when this A-share report was generated at 15:30, and per this repository's existing rule an intraday snapshot is not passed off as a close).

---

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

Data sources: exchange-standard limit-up/limit-down/failed-limit pools and seal orders, consecutive-board counts, first seal times (Eastmoney push2ex, 2026-09-17, with tc == len(pool) verified: ZT 47, DT 1, ZB 20)｜whole-market snapshot and industry breadth (Eastmoney push2delay clist, 5,916/5,916 with no pagination gap, 5,554 valid samples, industries grouped locally by the Shenwan field f100)｜industry/concept sector moves and net main-force inflow (Eastmoney push2delay, 496 industry entries, 504/504 concepts with no gap; the general-equipment entry was excluded and unused because the new listing N 沈鼓 N Shengu accounts for 65.8%)｜sub-industry constituents and equal-weighted means (Eastmoney BK1325 semiconductor materials n=31, BK1326 semiconductor equipment n=25, BK1331 digital chip design n=54, BK1340 printed circuit boards n=48, computed locally after pulling constituents BK by BK, no secid collisions)｜single-stock price/volume/open/turnover rate/volume ratio (Tencent Finance qt.gtimg.cn, timestamps verified one by one as 20260917; key names such as 沐曦股份 MetaX and N 沈鼓 N Shengu were cross-validated against Eastmoney push2delay as a second source)｜indices (Tencent Finance qt.gtimg.cn)｜the 9/16 limit-up pool and reconciliation benchmark (Eastmoney push2ex, 89 names, tc == len(pool) verified)｜unlock calendar (Eastmoney RPT_LIFT_STAGE)｜offshore renminbi (Eastmoney 133.USDCNH)｜pre-market list: this repository's reports/a-share/2026-09-17.md