A-Share Pre-Market Brief | 2026-09-03 Thursday

Caliber boundaries of this brief (judge the strength of its conclusions accordingly) ① News window = 2026-09-02 15:00 close → 2026-09-03 07:00. Any out-of-window news that is cited is uniformly tagged "out-of-window · already digested by the 9/2 session" and is not counted as new catalyst for today. ② The publication time of a company filing is always taken from the timestamp embedded in art_code; East Money's notice_date and CNINFO's "next trading day" tag are not accepted. Every "9/2 18:28" in this brief is the real publication time decoded from art_code. ⚠️ In this round of verification a sub-agent did exactly this — it trusted notice_date and recorded Inner Mongolia First Machinery's abnormal-movement announcement as published on 9/3, when it was in fact 9/2 18:28 — corrected, see §10. ③ The 9/2 quotes are on the Tencent qt.gtimg.cn closing-settlement basis, timestamps 20260902161402–20260902161439 (BSE 50 at 20260902153545, read 39 minutes earlier, but all after the close). ④ Limit-ups / consecutive limit-ups / sealing orders are on East Money's official 9/2 limit-up pool basis, with tc=52 / len(pool)=52 consistent. Seal-to-turnover ratio = sealing-order funds ÷ that day's turnover, computed by this brief; the pool median is 0.156. ⑤ Sector data are on the Tonghuashun 90-industry-sector closing-snapshot basis. ⚠️ This conflicts with East Money's "main-force fund flow" caliber: Tonghuashun gives power-grid equipment a net inflow of −RMB 1.816 billion, East Money gives +RMB 1.482 billion. The two models differ; this brief lists both numbers, does not net them off and does not pick a side. ⑥ The RMB 1.169 billion / RMB 630 million for Far East Smarter Energy (600869) and Hangzhou Cable (603618) are outputs of East Money's "main-force fund flow" model, not Dragon-Tiger List seat data — neither stock made the Dragon-Tiger List on 9/2, so the institutional vs hot-money composition cannot be broken out, and this brief draws no conclusion about the nature of the buy side. ⑦ "Institutional net buying" on the Dragon-Tiger List is always broken down to the seat level in this brief. ⚠️ The media caliber "2 institutions net bought RMB 136 million of BoYun New Materials" is wrong: the RMB 136 million is the all-seat net buy; after netting the four institutional seats it is only RMB 18.70 million (13.7%), while two hot-money seats account for 88.3%. See §5.10. ⑧ No northbound flow data (daily net buying has not been disclosed since 2024-08-19). ⑩ ⚠️ The attribution of the oil-price gain has been corrected — this is the single most important self-correction in this brief. The first draft wrote "Brent broke through 94–95 on 9/2 (+4~5%)"; on verification the price level is right but the gain was attributed to the wrong day: the +4.90% up-candle was 9/1 (close 95.11); 9/2 closed at 95.25, only +0.15% (intraday high 97.04). That is: within this brief's window (9/2 15:00 → 9/3 07:00) Brent was essentially flat, and there is no new oil-price catalyst. What A-shares reacted to in the 9/2 morning session was the overnight jump of 9/1, and it was fully digested that same day in a "gap up then fade." A further correction: the claim that Brent "briefly hit 126 dollars" during the first round of the crisis could not be confirmed; the measured 2026 intraday high is 119.50 (3/9) and the highest close 113.86 (5/4), and COSL's own interim report likewise describes "a sharp retreat from the year's high of nearly US$120/bbl" — two independent sources cross-confirming. ⑪ The 9/2 gain in domestic INE crude futures does hold: the front contract opened at 650.0, hit an intraday high of 710.2 and closed at 693.3, vs the 9/1 close of 637.8 → +11.35% at the intraday high, +8.70% at the close. This does not contradict Brent's mere +0.15% on 9/2 — the domestic night-session/day-session split is different, and what INE's 9/2 day session was catching up to is precisely that 9/1 Brent jump. ⑨ The two strongest conclusions in this brief both rest on things the closing price cannot show: first, the opposite treatment of defense vs oil & gas, visible only in OHLC; second, the fundamentals and the companies' own filings for three defense stocks, visible only by reading primary financial reports and announcements. Looking only at the 9/2 gainers/losers list would produce conclusions exactly opposite to this brief's.

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

I. Channels that worked

• Dragon-Tiger List stock_lhb_detail_em(20260902) returned 83 rows; the seat-level stock_lhb_stock_detail_em also worked, which was the key to debunking "institutions net bought RMB 136 million."
• Limit-up pool stock_zt_pool_em(20260902) tc=52/len=52; broken-seal pool 15 rows, limit-down pool 8 rows.
• Sectors via stock_board_industry_summary_ths() (90 rows). The East Money clist series was bypassed via the established backup channel (memory: one-sided pagination bias).
• Indices and single-stock daily bars via Tencent; all 13 symbols returned, 0 timeouts (better than yesterday's 2/5 anomalies).
• Filings via the East Money np-anotice-stock + np-cnotice-stock full-text endpoints; all five key filings were obtained in the original text, with no reliance on media paraphrase.
• On the sub-agent side: Tonghuashun stock_financial_*_ths, East Money stock_zygc_em and Baidu Stock stock_zh_valuation_baidu (ten years of weekly data, 731 points) all worked; East Money stock_individual_info_em and stock_zh_a_hist failed across the board and were rerouted to Tencent.

II. Five contaminations caught this round (two of which would have written a conclusion outright wrong)

1. ⚠️ A sub-agent trusted notice_date and recorded Inner Mongolia First Machinery's abnormal-movement announcement as "published 2026-09-03." I re-checked via art_code: AN202609021828930975 = 9/2 18:28. The direction of this correction is favorable — it lands inside the window, it is new information not yet reflected in the 9/2 close, and its evidentiary weight is therefore stronger. ⚠️ The memory item on this trap was stepped on just yesterday, and a sub-agent stepped on it again today — which means this has to be written into the agent's prompt and cannot rely on the main agent re-checking alone.
2. ⚠️ web.ifzq.gtimg.cn returns stale data without erroring for delisted / absorbed symbols. China Shipbuilding Industry Corp (601989) returned 2025-08 bars (it has been absorbed by China State Shipbuilding) and Shouhang Hi-Tech (002665) returned 2024-06 bars. The price fields look like normal values; only the date betrays that it is dead data. Both removed. Same family as the "CBOE's _DJI/_COMP are dead data" trap.
3. "CSI Defense H1 net profit attributable to parent +32.92%" was falsified by a sub-agent summing all 80 constituents one by one. See §10.1 — the numerator is right, the denominator is wrong. My first draft had already declined to accept that figure because I had found conflicting evidence (2026Q1 net profit attributable to parent −12.5%); the direction of that prior was right, but the real reason (a consolidation-caliber mismatch) is more specific than my guess.
4. Searching "September 2 A-share open" returned a Zhitong article (published 9-2 09:49, updated 10:15) saying "oil & gas, coal and banks strengthened against the tide" — that is a snapshot of the first 19 minutes, whereas at the 9/2 close coal mining & processing at −3.53% was dead last among the 90 sectors, 0 up and 34 down. Copying it would have written the sector that was distributed hardest that day as "strengthening against the tide." ⚠️ But the snapshot itself is useful evidence — it confirms that oil & gas and coal did gap up, which is exactly the first half of the "gap up then fade" conclusion in §2.
5. Searching "September 2 evening positive announcements" returned a large number of stale February / July / August articles, plus CFI.net's standing "Seven O'Clock" column (headline and URL structure nearly identical month to month). Each was checked for publication date and removed. Same family as "recurring-column headlines hide the date."

⚠️ II-supplement: a silent double-counting trap in East Money's Dragon-Tiger seat endpoint (it nearly made me think my own text was wrong) While re-checking BoYun New Materials' seat data, my quick script added the flag='buy' and flag='sell' tables of stock_lhb_stock_detail_em together directly, getting an all-seat net of RMB 162.64 million and an institutional share of 17.0% — which did not match the RMB 136.37 million / 13.7% in the main text. Cause: the endpoint's buy list and sell list contain overlapping rows (in this case the "Shenzhen-HK Stock Connect dedicated" seat and one of the "institutional dedicated" seats appear in both tables), so adding them directly double-counts. After deduplicating by branch name the total is RMB 136.37 million, matching the official Dragon-Tiger net buy digit for digit; netting the four institutional dedicated seats gives RMB 18.702 million (13.71%), and the two hot-money seats give RMB 120.48 million (88.35%). The main text's numbers were correct; my verification script was the thing that was wrong. Recording the criterion: when breaking out seats you must first deduplicate by branch name, and then use "deduplicated total == Dragon-Tiger net buy" as a closed-loop check — once that check ties out, the breakdown caliber is right; if it does not tie out, it is certainly double-counting, so do not rush to change the main text.

III. Four cross-checks actively performed on method, all of which changed a conclusion

1. Far East Smarter Energy's "winning-bid positive" was overturned by back-computation. The first draft was about to list cable as the No. 1 branch (No. 1 in main-force net buying on 9/2 at RMB 1.169 billion, plus 4 limit-ups). After pulling the original text: ① it is a routine disclosure made on the 1st of every month (actual publication dates 5/6, 6/1, 7/1, 8/3, 9/1, with identical headlines month after month); ② the caliber is a full-month cumulative; ③ August's RMB 1.780 billion vs July's RMB 2.808 billion is −36.6% month over month. The market bought a routine monthly report showing a halving to the limit-up.
2. ⚠️ The sub-agent's fundamentals check overturned the No. 1 and No. 2 recommendation slots of this brief's first draft outright. The draft had Inner Mongolia First Machinery and BoYun New Materials as the top two "watch closely" names, based on seal quality (0.326 / 0.212) and "2 institutions net buying." After verification: Inner Mongolia First Machinery has net profit attributable to parent −40.72%, military-trade exports −92%, PE at the 99.9% ten-year percentile and zero institutional seats on the Dragon-Tiger List; BoYun New Materials' growth comes from tungsten carbide price increases, not military products, Q2 was −71.2% sequentially, it is not in the CSI Defense constituent list, and the netted institutional buy is only RMB 18.70 million. Both were downgraded to "watch only," and the No. 1 slot passed to China State Shipbuilding. This is the biggest conclusion reversal in this brief, and the highest-value contribution the sub-agent made.
3. Feilong Auto Components' institutional buying and its controlling shareholder's share sale are on the same day. The Dragon-Tiger List shows 1 institution net buying RMB 459 million (the largest market-wide), but the 9/2 18:28 filing shows the controlling shareholder made a block sale of 1.635 million shares that same day. ⚠️ Also note: the filing's wording "holdings reduced by a total of 4,570,724 万股" is a typo (the unit should be shares); back-computing from the company's own 0.7952% gives total share capital ≈ 574.8 million shares, which cross-checks against float market cap of RMB 31.667 billion ÷ RMB 58.07 ≈ 545 million shares. This brief uses 4.5707 million shares and does not propagate the ten-thousand-fold error.
4. China State Shipbuilding's order strength was downgraded by the 8/28 control case and then upgraded by the sub-agent's industry data. Downgrade: on 8/28 it had just announced US$2.688 billion (2.7 times as large) and the stock moved only +0.75%, then −2.01% the next day. Upgrade: the sub-agent found that this one company accounts for 24% of the CSI Defense Index's revenue, and that excluding it the other 79 constituents' H1 net profit attributable to parent is −5.45% — i.e. the entire "defense earnings improvement" narrative is being carried by this one company. After stacking both directions it still ranks No. 1, but the reason changed from "order elasticity" to "the only name with real earnings and an in-window catalyst."

IV. What this brief did not obtain / did not do (listed honestly)

• Apart from Far East's routine monthly report, the cable branch yielded no in-window catalyst whatsoever. Each primary filing page was checked one by one: Hangzhou Cable stops at 8/28, Taiyang Cable at 8/26, Shenma Power at 8/28. The "UHV tendering +674%," "15th Five-Year Plan RMB 4 trillion" and "copper breaking RMB 108,000" items that turned up are all July–August or older background. ⚠️ Per "no news found ≠ no news," the wording is "no confirmable in-window catalyst obtained," not "there is no catalyst."
• Chongqing Jianshe Industry (002265) received no fundamentals check (the sub-agent's task only covered Inner Mongolia First Machinery / Great Wall Military Industry / BoYun New Materials); this brief marks its fundamentals judgment as "unverified" and its score is correspondingly conservative.
• The oil & gas chain check did not return It returned and substantially rewrote §2③, §3, §5, §7.2 and §9. ⚠️ That agent corrected one factual error in this brief (the Brent gain attributed to the wrong day), overturned one recommendation slot (SINOPEC Oilfield Equipment went from "watch only · No. 3" to Pass), and volunteered a rebuttal of this brief's own framework (the 9/2 moves cannot be explained by cost vs revenue). It also reported that East Money clist is down across the board (stock_zh_a_spot_em and stock_bid_ask_em all RemoteDisconnected), so it rerouted to Tencent/Tonghuashun on the established path — consistent with my own judgment in the main thread to bypass clist; that channel is confirmed unavailable today.
• Valuation percentiles must be discounted: Baidu Stock's history series is sparsely sampled, with 3Y samples of only n=65~275, so these are not daily percentiles. Every "percentile" statement in this brief is subject to this limitation and should not be used as a precise criterion.
• Still not obtained: Zhongman Petroleum's full-cycle cost per barrel and realized oil price (not disclosed in the interim report), the specific value of COSL's H1 new contract awards (the interim report only says "a new record" without a number), and the restart date for the 9 rigs in Iraq (the company explicitly writes "uncertainty remains").
• 9/2 intraday tick data were not obtained; "gap up then fade" is based on daily OHLC and cannot pin down the intraday distribution timing.
• The fundamentals of the 52 limit-up stocks were not checked one by one.
• bull-bear-debater was not used (the disagreement in this brief centers on "does the evidence hold," not on bullish vs bearish valuation).

---

0. Today in One Sentence

The strongest catalyst is the escalation of the U.S.–Iran conflict, but the oil-price leg is actually flat inside this brief's window — and it already ran the experiment on A-shares yesterday, with the result being rejection. ⚠️ First, a correction to a widely misreported timing attribution: the +4.90% Brent up-candle was 9/1 (close 95.11); 9/2 was only +0.15% (close 95.25). What A-shares reacted to in the 9/2 morning session was the 9/1 overnight jump. In the 09:49 snapshot of the 9/2 morning session, "oilfield services & engineering, coke, coal mining" were among the top gainers; by the close coal mining & processing at −3.53% was dead last among the 90 sectors, 0 up and 34 down. Offshore Oil Engineering open→close −5.32%, Zhongman Petroleum −6.86%, COSL −3.94% — the entire oil & gas chain was systematically distributed on the very day domestic crude futures spiked 11.35% intraday.

On the same catalyst, the defense chain was "bought from a lower open all the way to limit-up" — Great Wall Military Industry (601606) opened −0.75%, BoYun New Materials (002297) opened −1.65% and Chongqing Jianshe Industry (002265) opened −0.97%, and all three sealed at the limit; defense equipment at +2.06% ranked first among the 90 sectors with 56 up and 24 down. This divergence is visible only through OHLC.

But the most important conclusion in today's report is that it also knocks down half of the defense line itself. After primary-source financial-report checks on three defense stocks:

• Inner Mongolia First Machinery (600967) filed an abnormal-movement announcement at 9/2 18:28 stating in black and white that "the external market environment and industry policy have likewise not changed materially" and that it "has found no media reports or market rumors that could have a material impact on the company's share price" — the company itself, inside the window, personally denied the very catalyst the market is trading.
• Inner Mongolia First Machinery is the only one of the three with substantive military-trade (weapons export) business, and its 2026H1 export revenue fell −92% year over year (RMB 1.064 billion → RMB 85 million), with export gross profit evaporating by RMB 256 million — that alone is the entire reason its net profit attributable to parent is −40.72%. The one leg through which geopolitical conflict could theoretically transmit is collapsing in the latest reporting period.
• The repeatedly cited "CSI Defense H1 net profit attributable to parent +32.92%" is distorted, as verified by summing all 80 constituents one by one: same-caliber revenue growth is +7.81%, not +18.89%, and the gap comes almost entirely from the retrospective restatement for China State Shipbuilding's absorption of China Shipbuilding Industry Corp; excluding China State Shipbuilding, the other 79 constituents' H1 net profit attributable to parent is −5.45% (negative growth).

Hence the core contrast of this brief: the company that is actually carrying the entire defense sector's earnings (China State Shipbuilding, one name accounting for 24% of the index's revenue) is precisely the only one with a hard in-window catalyst — and the one nobody bought, closing 9/2 at −1.45%.

After the oil & gas check, the same contrast appeared again, and more sharply: the only "Hormuz beneficiary" bought to limit-up market-wide on 9/2, SINOPEC Oilfield Equipment (000852), listed "obstruction of shipping through the Strait of Hormuz" as a drag on results in its own 8/28 interim report — the original text reads "overseas, affected by geopolitical conflict, shipping through the Strait of Hormuz was obstructed and project progress in the Middle East slowed"; the same report shows first-half new orders of RMB 4.69 billion, −5.1% year over year, and H1 net profit attributable to parent swinging to a loss of −RMB 6.4 million. The market's reason for buying it points in the opposite direction from the operating reality the company itself wrote down.

Likely direction of flows: structural defense. 9/2 was already 1541 up / 3901 down (aggregate caliber of the 90 Tonghuashun industry sectors), with the 90 sectors showing a combined net outflow of RMB 95.964 billion and only 6 sectors net positive. U.S. stocks closed lower overnight (S&P −0.71%, Dow −0.79%, Nasdaq −1.03%), the 10-year Treasury yield hit a three-year high, and CME-implied odds of a Fed rate hike in September are about 68% — a hike, not a cut — a persistent headwind for ChiNext/STAR growth names. The conditions for a broad rally are not in place today.

---

1. News Overview

Window: 2026-09-02 15:00 → 2026-09-03 07:00. Impact levels S/A/B/C.

What did not appear in the window is information too: the defense and oil & gas chains had no company-level positive filings at all (Great Wall Military Industry's latest filing stops at 8/20, BoYun New Materials at 8/26, SINOPEC Oilfield Equipment at 8/27). The only company-level filing in the defense chain inside the window is Inner Mongolia First Machinery's abnormal-movement announcement denying the catalyst.

---

2. Strongest Positive Branches, Descending

Branch-by-Branch Notes

① Shipbuilding — the only branch with complete evidence, and the true engine of the defense sector's earnings

• China State Shipbuilding (600150) filed two announcements at 9/2 18:28, both verified against the primary original text:
  • Guangzhou Shipyard International, jointly with China Shipbuilding Trading, signed for 10 8,200-CEU LNG dual-fuel PCTCs, over US$1 billion, signed 9/1, payable in U.S. dollars, delivery 2029–2031.
  • Waigaoqiao Shipbuilding won the London ad hoc arbitration: owner ESSM1 LTD sought the return of the US$18.10 million advance payment plus interest on 1 jack-up drilling rig (H1368), and the tribunal dismissed all claims, expected to increase 2026 total profit by about RMB 98.90 million (subject to the annual audit). The case was filed in June 2017 and ran 9 years.
• ⚠️ The most important contrary evidence: on 8/28 the company had already announced 12 21,700-TEU LNG dual-fuel container ships at US$2.688 billion (2.7 times the current one), and the market reaction was only +0.75% on 8/28 and −2.01% on 8/31. Even an order 2.7 times as large could not move the share price.
• ⚠️ But one new piece of evidence props it back up: the sub-agent's one-by-one summation of the 80 CSI Defense constituents found that this one company accounts for 24% of the index's revenue, and that excluding it, the other 79 constituents' 2026H1 net profit attributable to parent is −5.45% (negative growth). In other words, the "defense upcycle" the market is buying today has its earnings source in this company's commercial shipbuilding cycle — and this company itself fell 1.45% on 9/2.
• Grading reminder: the RMB 98.90 million arbitration gain is a one-off non-operating item and should not be counted into the sustainable earnings center.

② Defense — the best money structure, the emptiest fundamentals, and the company itself denies the catalyst

• Why it is a positive: on 9/2 U.S. forces completed roughly 6.5 hours of strikes on Iran, Iran retaliated against four countries, and ceasefire talks are deadlocked (Iran has told mediators it is unwilling to meet the U.S. side in Islamabad in the near term).
• ⚠️ Four layers of falsification of the transmission chain (the section this brief invested the most verification in):
  1. The beneficiary directions the market named do not include these three. Public attribution on 9/2 pointed to "drones, counter-drone, air and missile defense, electronic warfare"; the three companies' main businesses are aircraft carbon brake discs + cemented carbide cutting-tool materials, ammunition and initiating explosives + prestressed anchorage, and main battle tanks + railway freight cars. Not one of them is in the sub-segments being named.
  2. Chinese defense orders come from domestic five-year-plan procurement, not from Middle East hostilities — and this can be verified directly in the financial statements. Great Wall Military Industry explicitly attributes its 50.65% drop in military-product revenue to "contract-signing progress" and "delayed delivery due to changes in customer demand," pointing to the domestic ordering cadence; Inner Mongolia First Machinery's 9/2 18:28 filing states "the external market environment and industry policy have likewise not changed materially"; BoYun New Materials' profit driver is the tungsten carbide price, with no transmission path to the Middle East.
  3. The one channel that could theoretically transmit — military-trade exports — has collapsed this period. Inner Mongolia First Machinery's 2026H1 export sales were −92% (RMB 1.064 billion → RMB 85 million), with export gross profit evaporating from RMB 265.5 million to RMB 9.3 million. Great Wall Military Industry's interim report characterizes military trade as "highly constrained by international geopolitics…" — in its own disclosure, geopolitics is a risk factor, not an opportunity. Moreover, Chinese military trade runs on long-cycle government-to-government contracts (Inner Mongolia First Machinery goes through the NORINCO platform, VT4/VT5/VN20), and signing to delivery is measured in years; Iran is not an in-service buyer of Chinese main battle equipment, and the United States certainly is not.
  4. The sector earnings figure underpinning the narrative is itself distorted. See §10.1.
• Is it already fully priced: it has been fermenting for 3 days. Inner Mongolia First Machinery's cumulative price deviation reached 20% over the three consecutive sessions of 8/31, 9/1 and 9/2, which has already triggered an abnormal-movement filing.
• But the contrary evidence must be recorded fairly: seal quality is objectively the best tier market-wide (Inner Mongolia First Machinery's seal-to-turnover ratio 0.326, 0 broken seals, turnover of only 3.82%), and the three names being pushed to the limit from lower opens is also a fact. "Money behavior is good" and "fundamentals are empty" are simultaneously true — this brief's treatment is: acknowledge the former, do not let the former add points to the latter, and uniformly downgrade the label to "watch only."
• One separate and valid medium-term thread (which must be severed from the Middle East): the domestic order recovery from the start of the "15th Five-Year Plan" plus the approach of the 2027 centenary of the armed forces; Inner Mongolia First Machinery's interim report says it "successfully won bids for key military-product projects in both army equipment and navy equipment… and expects continued ordering going forward," with contract liabilities +86% year over year. But this thread has nothing to do with the U.S.–Iran conflict and has not yet landed in any company's income statement. Treating the Middle East event as its trigger conflates two different things.

③ Oil & gas — oil did not move inside the window, and last round's interim reports already falsified this chain once

• ⚠️ First set the ruler straight: Brent closed 95.11 on 9/1 (+4.90%) and 95.25 on 9/2 (+0.15%). Inside this window oil was essentially flat. Domestic INE crude spiked +11.35% intraday in the 9/2 day session, catching up to that 9/1 jump — not new information.
• What happened on the equity side — this is the most important table in the brief:

Note ①: 8 of the 9 "gapped up then faded" on the first trading day after Brent's +4.90% on 9/1, the same day domestic INE spiked +11.35% intraday. Looking only at the closing change (−1% to −3%) you would think they simply drifted down with the index; only OHLC shows that they gapped up 2–5% and were sold all day long.

Note ②: this is not an isolated case. 8/31 (the first trading day after the U.S. strike on Larak Island on 8/30) had the same shape — COSL opened 13.00 and closed 12.89 on 8/31, Offshore Oil Engineering opened 6.20 and closed 6.16. Two geopolitical escalations, two gap-up-then-fade days. If there is a third gap up today, the burden of proof sits with the bulls.

⚠️ The control group from the previous round of the crisis — the most invested and most valuable piece of verification this brief did on the oil & gas chain.

The current Hormuz crisis has been running for more than 6 months since February 28, 2026 (Brent's 2026 intraday high is 119.50 on 3/9; "126 dollars" could not be confirmed and has been corrected). The previous spike left a complete, testable sample:

Three readings, each bearing directly on today's decisions:

1. The share-price peak leads the commodity peak by about 8 weeks. All six topped out on 3/3–3/5, whereas Brent's intraday high was 3/9 and its highest close 5/4 (113.86). By 4/29, when Brent closed at the year's second-highest 112.47, the four oilfield service/equipment names had already retraced 20.6%~26.8% from their March peaks. "New highs in oil" and "new highs in oil-service stocks" fully decoupled last round, and the decoupling point was early March.
2. Interim-report delivery: all four oilfield service/equipment names were falsified. The previous spike covered exactly H1 (Q2 average Brent US$96.02, +44.7% year over year, a historically high level), and the earnings delivered were: SINOPEC Oilfield Equipment swung to a loss, Zhongman Petroleum −67.03%, COSL ex-non-recurring −5.48%, Offshore Oil Engineering ex-non-recurring −8.28%.
3. The ones that actually delivered are the upstream equity-production holders: CNOOC Ltd H1 net profit attributable to parent +23.4% (Q2 alone +41.6%), PetroChina +22.0%. In the last oil spike, the money landed in the pockets of those who own the oilfields, not in the pockets of the service and equipment providers. The most elastic name, Zhongman, fell the most (−62.1%), and Offshore Oil Engineering, leaning on its backlog, fell the least (−6.1%).

The real transmission cadence of this chain (why oilfield services and equipment cannot catch it):

Sustained high oil prices → next year's capex budget revised up (usually published the following January)
                          → tendering/awards (+1~2 quarters) → contract signing, revenue recognized on percentage-of-completion (+2~4 quarters)
Total lag 4~8 quarters, and the first gate is the human decision called "the budget," not the price.

The gate is verifiable: CNOOC Ltd's 2026 capex budget is RMB 112.0–122.0 billion, and in July 2026 it reiterated that "the operating guidance is unchanged"; H1 actual spending on purchases and construction of fixed assets was RMB 53.38 billion, only +2.2% year over year. In a half-year when average Brent was +22.9% and its own net profit attributable to parent was +23.4%, it spent only 2.2% more on capex. COSL's H1 operating data confirm the same thing: drilling-rig operating days 9,741, −1.7% year over year; calendar-day utilization 88.7%, −2.5pp year over year — in a half-year when oil was +22.9%, offshore drilling workload posted negative growth.

• Reverse check (for whom is high oil a cost): the direction is right, but it must be cut by quarter; H1 cumulative figures will mislead you. Sinopec's H1 net profit attributable to parent was RMB 25.627 billion, +19.3%; on that number alone, "high oil prices hurt refiners" is falsified on the spot. Cut by quarter and it immediately reverses:

The heavier the refining weight, the worse that high-oil-price quarter. Sinopec's Q2 net profit attributable to parent was RMB 8.62 billion, half of Q1's RMB 17.01 billion; the H1 +19.3% is carried entirely by Q1. There is primary evidence for the mechanism: the Petroleum Price Control Measures set US$130/bbl as the control ceiling, and on 2026-03-23 the NDRC launched the first temporary price control on refined products in history — the mechanism called for a gasoline increase of RMB 2,205/tonne but the actual adjustment was RMB 1,160/tonne, about RMB 0.85 per litre less, a pass-through rate of only 52.6%, with the shortfall absorbed by refiners. This gate has already come down once for real during the first round of the crisis.

• ⚠️ But one rebuttal of myself must be given (and it matters): the 9/2 tape does not support explaining that day's moves via "cost vs revenue." CNOOC Ltd (pure upstream, the biggest beneficiary, H1 +23.4%) fell −2.14% that day, more than Sinopec (refiner, −1.63%). If the market had been pricing oil economics that day, those two numbers should be the other way around. So 9/2 is more likely profit-taking / positioning-driven, not a fundamental repricing. Reading it as "the market recognized refining's cost exposure" would be a misreading — refining being hurt is a fact that holds at the financial-statement level, but it is not the cause of that 9/2 candle.
• Filling in a cell I missed in the first draft: the gas sector was +0.63% on 9/2 with 16 up and 10 down, the only sub-sector of the whole energy complex to close green (leader Guizhou Gas (600903) +9.95% with a RMB 79.68 million sealing order and zero broken seals).

④ Data-center thermal management — a meaningful amount but a questionable entity

Beijing Creative Distribution Automation (002350) won a bid worth RMB 91.5344 million, equal to 4.25% of 2025 audited revenue. But three deductions must be stated at the same time: ① the winning entity, Shenzhen Cyber Engine Technology, is a subsidiary in which the company only acquired 51% for RMB 5.10 million on July 13, 2026; ② it was founded less than a year ago and had not previously carried on any business, so this is the first major project since its formation; ③ only a notice of award has been received; no formal contract has been signed.

⑤ Cable — the heaviest money, the falsest evidence

Four limit-ups appeared on 9/2: Far East Smarter Energy (600869), Taiyang Cable (002300), Hangzhou Cable (603618) and Shenma Power (603530), and on the East Money caliber Far East's main-force net buying of RMB 1.169 billion ranked first market-wide, with Hangzhou Cable's RMB 630 million second. But the catalyst does not survive back-computation:

This is a routine disclosure made on the 1st of every month (actual publication dates 5/6, 6/1, 7/1, 8/3, 9/1, with headlines identical month after month), on a full-month cumulative caliber, and halved sequentially. The filing itself states that "the company will proactively disclose information on relevant winning bids and signed contracts" — it is a standing column, not a sudden positive. Separately, Far East filed on 8/31 18:28 that a fire accident occurred at a controlled second-tier subsidiary. Hangzhou Cable, Taiyang Cable and Shenma Power all had no new filings around 9/2 and were simply following the move.

---

3. Overall Stock-Level Positive-Strength Ranking (descending by stock)

Board limits: SH/SZ main boards ±10% / ChiNext 30xxxx ±20% / STAR Market 68xxxx ±20% / BSE 8xxxxx·9xxxxx ±30%.

---

4. Stock Scoring Model (out of 100)

Components: source authority 0–15 / directness of the positive 0–20 / earnings elasticity 0–15 / industry position and fundamentals 0–15 / expectation gap 0–10 / theme persistence 0–10 / A-share trading characteristics 0–10 / risk deduction 0 to −15.

Scoring notes (three things that must be made clear):

1. Inner Mongolia First Machinery still gets a full 10 on "trading characteristics" while getting only 2 on "source authority" — these do not contradict each other. Its seal quality is objectively first market-wide (seal-to-turnover ratio 0.326, 0 broken seals, turnover 3.82%), and that deserves the score; but its "news" is an abnormal-movement filing in which the company itself denies the catalyst, maximally authoritative but negative in direction, hence the lowest score. This is precisely what this brief wants to convey: trading structure and news evidence can diverge completely.
2. The low scores for Far East Smarter Energy / Hangzhou Cable are not because their fundamentals are bad, but because "source authority" and "directness of the positive" are near zero. Their 9/2 money volumes were the largest market-wide — money inflow ≠ evidence holding. The memory item "single-day extremes in main-force fund flow are a contrarian indicator" applies here.
3. SINOPEC Oilfield Equipment's 0 on "directness of the positive" is one of the two zeros in this table (the other is Zhongman Petroleum (603619)). It is not that "the positive is weak" but that the direction is reversed — the reason it was bought (Hormuz obstruction) is listed as a drag in its own interim report. Zhongman Petroleum gets a 0 for the same reason. When a stock's "positive," once verified against primary sources, points the other way, directness is not a low score but a zero.

---

5. Detailed Stock Analysis (12 names)

The order broadly follows the descending overall ranking in §3, but stocks in the same branch are placed adjacently for comparison (e.g. the oil & gas chain at 5.3–5.5 and 5.7), so a few positions deviate slightly from strict score order. Each name's score is as given in §3/§4.

51 China State Shipbuilding600150SH main board ±10% — Watch closely — 61 pts

• Related news (in-window, all verified against the primary original text):
  • 9/2 18:28 Announcement 2026-030: controlled subsidiary Guangzhou Shipyard International, jointly with China Shipbuilding Industry Trading, signed for 10 8,200-CEU LNG dual-fuel car carriers, total value over US$1 billion, signed 9/1, payable in U.S. dollars, delivery 2029–2031.
  • 9/2 18:28 Announcement 2026-029: the London ad hoc arbitration between Waigaoqiao Shipbuilding and owner ESSM1 LTD concluded, with the award dismissing all of the claimant's arbitration claims. There is no need to refund the US$18.10 million advance payment plus interest on 1 jack-up drilling rig (H1368), and it is expected to increase 2026 total profit by about RMB 98.90 million (subject to the annual audit). The case was filed in June 2017.
• The positive logic splits into two layers of different strength. The RMB 98.90 million from the arbitration directly adds to 2026 current-period total profit; the order explicitly does not affect the current period — the company itself says it "does not have a material impact on the company's current-period profit," and what it affects is 2029–2031 revenue. It is the only name in the window with "a confirmed amount + a primary filing," but "goes straight into the statements" is true only for the arbitration half.
• Branch stage: not started. 9/2 closed at 33.93, −1.45%, opening 34.20 → open→close −0.79%. It has been range-bound at 33.9–34.8 for the past 7 sessions, with no limit-up ladder, and it is not part of any theme currently being played.
• Fundamentals check (an unexpected find of this brief's verification): while summing the 80 CSI Defense constituents one by one, the sub-agent found that this one company accounts for 24% of that index's revenue (the second-largest constituent, Chujiang New Materials (002171), is 11%), and that excluding it, the other 79 constituents' 2026H1 net profit attributable to parent is −5.45%. The "defense upcycle" the market is chasing today has its earnings source in this company's commercial shipbuilding cycle. Also: its absorption of China Shipbuilding Industry Corp (formerly 601989) completed in September 2025 as a business combination under common control, so prior-period statements were restated retrospectively (2025H1 revenue was restated from the originally disclosed RMB 40.325 billion to RMB 72.640 billion).
• Industry position: the absolute global shipbuilding leader.
• ⚠️ The company itself played down the current-period significance of this order. From the "Important Notice" section of the filing: "Impact on the listed company's results for the year: the company will recognize revenue under the revenue standard, and it does not have a material impact on the company's current-period profit (subject to the final audit result)." It also specifically notes that "the contract performance period is long, and performance may be affected by changes in the shipping and shipbuilding markets and by U.S. dollar / RMB exchange-rate risk." This US$1 billion order is evidence of medium-to-long-term competitiveness, not an increment to 2026 earnings.
• Two further details affecting evidentiary strength: ① the counterparty is exempt from disclosure (under the Measures for the Administration of Deferral and Exemption of Corporate Information Disclosure, it is described only as "a well-known shipowner"), so the owner's identity cannot be verified; ② the contract is an ordinary-course operating contract requiring no board or shareholder-meeting approval, i.e. the company itself did not treat it as a material matter needing governance-level endorsement.
• ⚠️ The most important deduction: the 12 21,700-TEU ships announced on 8/28 at US$2.688 billion (2.7 times the current one) bought only +0.75%, followed by −2.01% the next day. The market has already stated clearly that it will not pay for shipbuilding orders.
• Final judgment: watch closely. Evidence quality is first market-wide, but the expectation gap may be negative — what it needs is patience, not same-day elasticity. The RMB 98.90 million arbitration gain is a one-off and is not counted into the valuation center. It is ranked No. 1 not because it will rise today, but because it is the only name on this list that, if it rises, will have risen for a reason.

52 Inner Mongolia First Machinery600967SH main board ±10% — Watch only — 34 pts (No. 1 in the first draft, downgraded after verification)

• ⚠️ The related news is a "reverse" filing. At 9/2 18:28 the company issued an Announcement on Abnormal Share Trading Fluctuations (Interim 2026-026), which reads:

  "The company's shares recorded a cumulative closing-price deviation of 20% over 3 consecutive trading days on August 31, September 1 and September 2, 2026, constituting an abnormal share-trading fluctuation." "The company's current operations are normal, its principal business has not changed materially, and the external market environment and industry policy have likewise not changed materially." "Upon self-examination, the company has found no media reports or market rumors that could have a material impact on the company's share price." "The controlling shareholder and its persons acting in concert, and the company's directors and senior management, did not trade the company's shares during this abnormal-fluctuation period."

  The company itself, inside the window and in a written filing, denied the very catalyst the market is trading. This filing was published after the 9/2 close and is not yet reflected in the 9/2 price.

• Fundamentals check (uniformly negative results):

Item	2026H1	YoY
Revenue	RMB 5.232 billion	−8.64%
Net profit attributable to parent	RMB 172 million	−40.72%
Ex-non-recurring	RMB 168 million	−41.61%
Gross margin	7.97%	−1.8pp
Operating cash flow	−RMB 760 million	(−RMB 160 million in the prior-year period)

Table scrolls horizontally

Net profit attributable to parent has stepped down for three straight years: RMB 851 million (2023) → RMB 500 million (2024) → RMB 335 million (2025) → down another 40.7% in 2026H1; gross margin slid from 16.12% to 7.97%. 2026Q2 net profit attributable to parent alone was RMB 34 million, −75.4% sequentially from Q1's RMB 138 million.

• ⚠️ The regional breakdown reveals the key point — the one leg through which geopolitics could transmit is collapsing:

	2026H1 revenue	Share	Gross profit	2025H1 revenue	Gross profit
Domestic sales	RMB 5.147 billion	98.38%	RMB 407.7 million	RMB 4.663 billion	RMB 293.0 million
Overseas sales (military trade)	RMB 85 million	1.62%	RMB 9.3 million	RMB 1.064 billion	RMB 265.5 million

Table scrolls horizontally

Export revenue fell −92.0% year over year and export gross profit evaporated by RMB 256.2 million — 100% of the decline in total gross profit (RMB 141.5 million) is caused by the shrinkage in military trade, while the domestic business is actually offsetting it (gross profit +39%). It is the only one of the three with substantive military-trade business, and that business is close to zero in the latest reporting period. (To be fair: military-trade deliveries are extremely lumpy — exports were only RMB 225 million in 2025H2 and only RMB 279 million in 2024H1, so a single low period does not equal a trend; but "the catalyst has already landed in the statements" definitively does not hold.)

• Industry position: real and exclusive. From the interim report: "the company is China's only high-tech weapons and equipment R&D and manufacturing group combining main battle tanks, wheeled combat vehicles and medium-calibre artillery." This is an institutional barrier and cannot be replicated. But the same report admits competition is worsening — "the once relatively closed defense system is gradually opening up… market participants are increasingly diverse," consistent with gross margin sliding from 16.12% to 7.97%: the monopoly position is intact, the pricing power is eroding.
• Valuation: PE (TTM) 105.30, the 99.9th ten-year percentile — the most expensive in ten years (historical median 28.05). It is not that the share price has gone mad; it is that TTM net profit attributable to parent is down to RMB 217 million. PB 1.90 sits at the 50.3rd ten-year percentile. PB says it is not expensive, PE says there are no earnings; these do not contradict each other: the assets are there, the earnings are not.
• Technical sentiment (the only positive): a seal-to-turnover ratio of 0.326, the highest of the 11 limit-up stocks this brief tracks and 2.1 times the pool median of 0.156; 0 broken seals; first sealed at 09:34:45 and never unsealed; turnover of only 3.82%. Low turnover + a heavy sealing order = very light selling pressure; this is the best-quality seal market-wide. ⚠️ But the RMB 212 million of Dragon-Tiger net buying came entirely from ordinary seats, with zero institutional seats.
• Final judgment: watch only (downgraded from "watch closely · No. 1" in the first draft). The seal quality is real, the fundamentals are empty, and the company itself denies the catalyst. When all three are simultaneously true, the correct treatment is not to average them but to acknowledge that it may keep rising while not giving it a recommendation slot — because what is driving it up cannot be backed by evidence, and once sentiment reverses there are no fundamentals to catch it. A PE at the 99.9th ten-year percentile means the margin for error is close to zero.

53 CNOOC Ltd600938SH main board ±10% — Watch only — 58 pts

• Related news: ⚠️ oil barely moved inside the window (Brent +0.15% on 9/2). The real fulcrum is its own already-delivered earnings.
• Positive logic: the purest oil-price play in the whole chain — upstream equity production benefits directly, unlike refining where oil is a cost and unlike oilfield services which must wait for capex. Logically it is the right answer on this chain, and the verification proves upstream is the only link that has ever delivered.
• Fundamentals check (the only unambiguously positive one of the four oil & gas names): H1 net profit attributable to parent +23.4%, Q2 alone +41.6% and +19.2% sequentially. Against Brent's Q2 +44.7% year over year, the observed elasticity is about 0.74, the closest to 1 of any immediate elasticity in the chain. In the last oil spike the money landed precisely in the pockets of equity-production holders like this.
• ⚠️ But the market has refused it twice in a row: on 9/2 it opened 35.41 (a gap of +2.61%) and closed 33.77 (−2.14%), open→close −4.63%; 8/31 was also a gap up.
• ⚠️ A rebuttal of myself: it fell −2.14% on 9/2, more than the refiner Sinopec (−1.63%). If the market had been pricing oil economics that day, those two numbers should be reversed. So 9/2 looks more like profit-taking / positioning-driven than a fundamental repricing — I do not read that day's decline as "the market turning bearish on upstream."
• Final judgment: watch only. The logic is the soundest, the earnings the most real, and the price action the worst — all three at once. If it gaps up again today, that is the third run of the same experiment; only if it opens flat or lower and then gets bought up does the character of this chain change.

54 Offshore Oil Engineering600583SH main board ±10% — Watch only — 50 pts

• Positive logic: indirect, and the driver is the backlog rather than the oil price.
• Fundamentals check (the only one of the four with a genuine forward improvement): H1 new contract awards of RMB 21.185 billion (RMB 12.068 billion in H1-2025, +75.6%); period-end backlog of about RMB 73.3 billion (about RMB 40.7 billion a year earlier, +80.1%), covering roughly 2.7 times full-year 2025 revenue of RMB 27.163 billion. Its financials are the most stable of the four: PB 0.93, PE 12.88, net cash +RMB 7.698 billion.
• ⚠️ But three offsets must be stated at the same time: ① the backlog is a cumulative figure first disclosed with the interim report on 8/18 and has been sitting in the tape for 11 trading days, so it is not new information (the memory item "cumulative-period order figures are old news" applies); ② gross margin fell from 16.25% to 14.87% — volume bought with price (new awards include low-margin overseas work such as Saudi Aramco CRPO 161); ③ accounts receivable of RMB 6.021 billion, +49.8%, while revenue was only +9.4% and operating cash flow was −30.1% — order quality and collection quality are diverging. H1 ex-non-recurring was −8.28%.
• Technical sentiment: on 9/2 it opened 6.39 (a gap of +1.75%) and closed 6.05, open→close −5.32%.
• Final judgment: watch only. It is the only name in oilfield services whose case can be argued from the backlog rather than an oil-price story, and PB 0.93 provides a margin of safety; but the backlog has been public for 11 days, ex-non-recurring is still falling and receivables have surged. Suitable for the tracking pool, not for chasing today.

55 SINOPEC Oilfield Equipment000852PassSZ main board ±10% — 22 pts (in the first draft "watch only · No. 3," downgraded after verification)

After BoYun New Materials, this is the second stock in this brief whose buy case is directly overturned by its own primary interim report.

• The market's reason for buying it: Hormuz obstruction + rising oil → oilfield equipment benefits. On 9/2 it opened 5.82 and closed 6.06 at the limit, open→close +4.12%, the only positive among the 9 oil & gas chain stocks.

• ⚠️ But its own 8/28 interim report writes Hormuz up as a drag. Excerpts from the original text:

  Oil & gas steel pipe segment: "overseas, affected by geopolitical conflict, shipping through the Strait of Hormuz was obstructed and project progress in the Middle East slowed." "Expansion in the overseas Middle East market was obstructed by geopolitical conflict; affected by project suspensions and restricted personnel travel, some project orders could not be closed." "New orders in the first half were RMB 4.69 billion, down 5.1% year over year." "Sinopec and PetroChina cut their fracturing-equipment investment plans, and the company's fixed-pressure equipment and high-pressure manifold revenue fell by RMB 155 million year over year."

  Its only forward indicator (new orders) is falling; its two big customers are cutting capex; and the catalyst it was bought on is a deduction item in its own report.

• Fundamentals (the worst of the four): H1 revenue RMB 3.123 billion (−10.15%), net profit attributable to parent −RMB 6.4 million (a swing to loss), ex-non-recurring −RMB 13.2 million; gearing 69.21% (highest of the four); operating cash flow −RMB 699 million (−RMB 172 million a year earlier); cash of RMB 434 million against short-term borrowings of RMB 2.125 billion; inventory of RMB 3.979 billion + receivables of RMB 3.079 billion = 65% of total assets, and the interim report's own third listed risk is "occupation by the two funds and cash-flow risk." By region, domestic revenue was RMB 2.323 billion, −17.9%.

• The transmission chain itself does not support it either: it sells oil & gas drilling and production equipment, so its revenue is a function of capex, lags oil prices by 4–8 quarters, and the first gate is the human decision called "the budget." The official limit-up pool labels its industry as specialty equipment, not oil & gas.

• Technical sentiment: ⚠️ a seal-to-turnover ratio of only 0.056 (a third of the pool median), seal broken once, final seal at 11:03:15, Tonghuashun net fund flow −RMB 174 million (50% of turnover) — the weakest sealing order among the 6 energy-related limit-ups that day (compare Guizhou Gas (600903), with a RMB 79.68 million sealing order and zero broken seals). (Note: its absence from the 9/2 Dragon-Tiger List is normal — a single-day limit-up requires being in the top 5 by price deviation — and carries no extra information.)

• Final judgment: Pass (downgraded from "watch only"). "The only Hormuz beneficiary bought to limit-up market-wide on 9/2 lists Hormuz as a drag in its own interim report" — this is the sharpest counter-evidence this brief found. The limit-up is a fact, but the reason for buying points in the opposite direction from the operating reality.

56 Beijing Creative Distribution Automation002350SZ main board ±10% — Watch only — 45 pts

• Related news: a 9/2 filing states that controlled subsidiary Shenzhen Cyber Engine Technology won the cooling-source module general-integration tender of Guangdong Qichuang Network Technology for a Thai data center, with a winning amount of about RMB 91.5344 million, equal to 4.25% of the company's 2025 audited operating revenue (Jiemian⁵).
• ⚠️ Three deductions that must be stated together:
  1. The winning entity, Cyber Engine, is a subsidiary in which Beijing Creative Distribution Automation only acquired 51% with its own funds for RMB 5.10 million on July 13, 2026;
  2. that company was founded less than a year ago and had not previously carried on any business, so this is the first major project since its formation;
  3. only a notice of award has been received; no formal project contract has been signed.
• Final judgment: watch only. A subsidiary acquired two months ago for RMB 5.10 million, founded less than a year ago, with zero prior business, has won a RMB 91.5344 million order and has not yet signed — the amount is meaningful but the ability to perform is entirely unverified. Reassess when a formal contract-signing announcement appears.

57 COSL601808SH main board ±10% — Watch only — 40 pts

• The oil-price positive applies, but its revenue comes mainly from CNOOC's capex, not from the spot oil price.
• Fundamentals check (falsifying "oil rises → oilfield services benefit"): H1 revenue RMB 23.787 billion (+2.00%), net profit attributable to parent RMB 2.009 billion (+2.28%), ex-non-recurring RMB 1.833 billion (−5.48%); Q2 alone net profit attributable to parent +7.1%, ex-non-recurring only +1.8% — in a quarter when Brent was +44.7% year over year. The operating data are even more direct: drilling-rig operating days 9,741, −165 days (−1.7%) year over year, jack-ups −5.2%, calendar-day utilization 88.7%, −2.5pp year over year. In a half-year when oil was +22.9%, offshore drilling workload posted negative growth. The only bright spot is semi-submersibles (operating days +11.4%, available-day utilization +9.3pp).
• The gate sits on CNOOC's budget sheet: CNOOC Ltd's 2026 capex budget is RMB 112.0–122.0 billion, and in July 2026 it reiterated that "the operating guidance is unchanged"; H1 actual spending on purchases and construction of fixed assets was RMB 53.38 billion, only +2.2% year over year. If the full-year budget is unchanged, the remaining H2 allowance is already locked, so however high oil goes it will not change 2026 revenue; the real variable is the 2027 budget (usually published the following January).
• Valuation is the only positive: PE 15.57 (5th percentile on 3Y), PB 1.28 (7th percentile on 3Y) — genuinely cheap. ⚠️ But the percentiles must be discounted: the history series is sparsely sampled, with 3Y samples of only n=65~275, so these are not daily percentiles.
• On 9/2 it opened 13.20 (a gap of +2.25%) and closed 12.68 (−1.78%), open→close −3.94%; 8/31 was the same gap-up-then-fade. Twice.
• Watch only. Its leadership position and low valuation are both real, but the buy case of "oil is rising" has been falsified by its own operating data — until that changes, any statement of "oil rises → COSL benefits" lacks transmission evidence.

58 Chongqing Jianshe Industry002265SZ main board ±10% — Watch only — 43 pts

• No company-level in-window filing. On 9/2 it opened 19.30 (a lower open of −0.97%) → closed 21.44 at the limit, open→close +11.09%, and was listed by several media outlets as a defense leader (Sina⁸).
• ⚠️ Fundamentals unverified (not covered by this round's sub-agent task). Note that East Money classifies it under "auto parts" rather than defense — suggesting the civilian revenue share may not be small, and the "defense" label needs revenue-mix support.
• Technicals: seal-to-turnover ratio 0.216 (above the median), seal broken once, first sealed at 09:32:42.
• Watch only. The lower open pushed to the limit and a healthy seal-to-turnover ratio are facts, but against the backdrop of two other names in the same branch having had their fundamentals falsified, an unverified stock in the same branch should not receive greater trust.

59 Shengnan Technology920006BSE ±30% — Watch only — 36 pts

• On 9/2 it was +29.96% (a BSE 30cm limit-up), the leading gainer in the defense equipment sector. Its main businesses are aviation equipment manufacturing and military power supplies.
• ⚠️ The BSE limit is ±30%; do not read it with ±10% intuition. +29.96% is a limit-up; the reverse, −30%, is equally possible in a single day. There is also a RMB 500,000 asset eligibility threshold, so a fair share of readers cannot buy it — that is information, not a reason for exclusion.
• Fundamentals unverified. Watch only. The next-day risk/reward after a first-board 30cm move is very poor.

510 BoYun New Materials002297PassSZ main board ±10% — 33 pts (No. 2 in the first draft, downgraded after verification)

This is the name most thoroughly changed by verification in this brief. It is not a defense stock rising; it is a tungsten-materials stock rolling over.

• The headline results look excellent: 2026H1 revenue RMB 745 million (+88.33%), net profit attributable to parent RMB 170.2 million (+1911.60%), ex-non-recurring RMB 162 million, gross margin 44.40% (+23.3pp).
• ⚠️ But the growth has nothing to do with "military." Segment composition:

Segment	2026H1 revenue	Share	Gross margin	2025H1 gross margin
High-performance cemented carbide and related materials	RMB 585.5 million	78.6%	40.32%	2.62%
Aerospace and civil carbon/carbon composites	RMB 144.6 million	19.4%	56.56%	53.43%

Table scrolls horizontally

Cemented carbide revenue was +132% and its gross margin jumped from 2.62% to 40.32%, contributing almost the entire increment; carbon/carbon composites grew only 10.2%. The company writes it very plainly in the 2026 interim report:

"Cemented carbide business: in the first quarter of 2026, affected by the continued rise in the price of the main raw material tungsten carbide, the company raised its product prices accordingly, driving significant year-over-year growth in this segment's revenue and gross margin; in the second quarter of 2026, as the market price of tungsten carbide retreated from its high for the year, product profitability narrowed relative to the first quarter."

This is a price pass-through during a tungsten upcycle, not a ramp in military products. And the company has already said Q2 turned.

• ⚠️ The turn is already provisioned in the accounts: 2026Q2 net profit attributable to parent was RMB 38.0 million, −71.2% sequentially from Q1's RMB 132.0 million; on the same day it announced asset impairment provisions including "certain contracts to be performed turning into onerous contracts," with a combined impact on total profit of −RMB 89.1106 million. At the same time inventory rose from RMB 552 million to RMB 802 million (+45%), i.e. stockpiling during a period of falling raw-material prices.
• ⚠️ Cash flow: net operating cash flow of −RMB 43.5389 million, against net profit attributable to parent of +RMB 170.2 million → cash content of earnings of −0.26x; cash-to-revenue ratio 0.53; contract liabilities of only RMB 19.7556 million — domestic defense suppliers typically carry sizeable advance payments for military products, and BoYun's near-zero contract liabilities are themselves evidence that military prepayments are not the main body of its business model.
• ⚠️ Military share: the company does not disclose it and it cannot be broken out. The segment name itself is a combined caliber ("aerospace and civil"), so 19.4% is the absolute ceiling of the military share and the true figure must be lower. A full-text search of the interim report for the character "军" (military) returns only 2 hits, and both are the names of funds among the shareholders. The core competencies the company describes all point to civil aviation and aerospace programs (strategic cooperation with COMAC, C919 carbon brake discs, the first domestic PMA issued by the CAAC).
• ⚠️ It is not even in the CSI Defense (399967) constituent list — the 80-name list was checked one by one and 002297 is not there. The market pins a defense label on it; the index compiler does not.
• ⚠️ The valuation is a textbook shape: PE (TTM) 51.26, the 0.9th ten-year percentile; PB 4.99, the 97.4th ten-year percentile. "Low PE + high PB percentile" is the classic signal of a cyclical top — the denominator E has been pushed to a historical high by one tungsten-price cycle, making PE look extremely cheap, while on a book-value basis the share price sits at its most expensive level in ten years. If E returns to 2025 levels (full-year net profit attributable to parent of RMB 61.9 million), the current market cap corresponds to a PE of about 185x.
• ⚠️ The Dragon-Tiger List must be restated: the media caliber "2 institutions net bought RMB 136 million" is wrong. Seat-level breakdown:

Seat	Net amount
Kaiyuan Securities Xi'an Xidajie (hot money)	+RMB 73.888 million
Guotai Haitong Shanghai Changning Jiangsu Rd (hot money)	+RMB 46.591 million
Shenzhen-HK Stock Connect dedicated	+RMB 17.375 million
Founder Securities Changsha Furong Rd	−RMB 20.189 million
4 institutional dedicated seats, netted	+RMB 18.702 million
Total (= official Dragon-Tiger net buy)	+RMB 136.37 million ✓ closed loop

Table scrolls horizontally

The RMB 136 million is the all-seat net buy; the buy list has 2 institutional seats, but the sell list has another 2 institutional seats selling, and after netting all four the institutional net buy is only RMB 18.70 million, or 13.7%, while the two large hot-money seats account for 88.3%. This was a hot-money-led Dragon-Tiger appearance.

• Final judgment: Pass (downgraded from "watch closely · No. 2" in the first draft). The seal-to-turnover ratio of 0.212 and 0 broken seals are still facts, but the reason it is being bought (defense) and the reason it makes money (tungsten prices) are simply not the same thing — and the latter has already turned.

511 Great Wall Military Industry601606PassSH main board ±10% — 18 pts

• ⚠️ Military-product revenue halved; it is the worst of the three.

Segment	2026H1	Share of principal business	Gross margin	2025H1	Share of principal business	Gross margin
Equipment manufacturing (military)	RMB 238.5 million	50.84%	14.98%	RMB 483.3 million	69.14%	22.93%
Civil products (prestressed anchorage, plastics)	RMB 216.4 million	46.14%	12.76%	RMB 199.9 million	28.60%	12.55%

Table scrolls horizontally

From the interim report: "affected by contract-signing progress for some final-assembly products and by delayed delivery due to changes in customer demand for others, special-product operating revenue fell 50.65% year over year." The cause of the decline is the domestic military ordering cadence, not external hostilities. Civil products (prestressed anchorage, used in construction/bridges, unrelated to defense) now account for nearly half of revenue.

• Financials: 2026H1 revenue RMB 469 million (−32.89%), net profit attributable to parent −RMB 96.45 million (loss widened 252%), ex-non-recurring −RMB 105 million. Three consecutive years of ex-non-recurring losses (2024FY −RMB 378 million, 2025FY −RMB 12.9 million, 2026H1 −RMB 105 million). Gearing 52.33%, total liabilities of RMB 2.360 billion > net assets attributable to parent of RMB 2.140 billion. Accounts receivable of RMB 1.307 billion = 2.79 times H1 revenue, with 516 turnover days. Operating cash flow was +RMB 27.1377 million, but the company explains it came from "stepping up collection of accounts receivable" — that is collecting existing balances, not new receipts, and this road cannot be walked twice. Annualized ROE is −9.01%.

• ⚠️ The company has publicly poured cold water on itself twice. From the Announcement on Share Trading Risk (2026-07-28):

  "The company's share price has risen a cumulative 39.87% since July 20, 2026… but the company's fundamentals have not changed materially, and there is a situation of overheated market sentiment" "The company's price-to-book ratio is 11.70x… the price-to-book ratio of the SWS defense industry in which the company sits is 5.47… there is a risk of overvaluation"

  The current PB has reached 12.39 (the 90.1st ten-year percentile), higher than when the company issued that warning. The interim report also lists geopolitics as a risk item: "military-trade business… is highly constrained by international geopolitics and by changes in domestic and foreign policies and regulations."

• Final judgment: Pass. The loosest chip structure in the defense branch (seal-to-turnover ratio 0.112, turnover 8.20%, turnover value RMB 2.140 billion) + the worst fundamentals + the company's own overvaluation warning. The direction may be right, but this name should get no position at all.

512 Zhongman Petroleum603619PassSH main board ±10% — 10 pts

The mapping "owns its own oilfields = direct leverage to the oil price" is cut off by three gates in this name.

• ⚠️ A new in-window negative (disclosed 9/2, not yet reflected in the close): the Announcement of a Shareholder Share-Reduction Plan (No. 2026-048) — pre-IPO shareholder and former actual controller Zhu Fengxue plans to sell, between 2026-09-23 and 12-22, all 5.5634 million shares he holds = 1.20% of total share capital (about RMB 129 million at RMB 23.22), for reasons of "his own funding needs." (It is standard practice on the Shanghai market to publish reduction plans after the close, so it should not be used to explain the intraday gap-up-then-fade of 9/2; but it is a new variable for 9/3 and had not been traded before.)
• Equity production is disclosed and need not be inferred: H1 crude output was 473,000 tonnes (+7.77%). Wensu (Xinjiang, 100% working interest) contributed 324,200 tonnes of equivalent; Jianguo (Kazakhstan, 51% working interest at period end, raised to 100% by the disclosure date) 220,100 tonnes; Anbian trial production 10,000 tonnes.
• ⚠️ Three gates:
  1. What it produces, it cannot sell. H1 sales volume was only 353,000 tonnes, just 74.6% of output, leaving roughly 120,000 tonnes sitting in inventory; inventory was RMB 1.134 billion, +39.5%. The oil price only works on barrels that have been sold.
  2. The Middle East conflict shut down its core business directly. From the interim report: "the company has deployed a total of 14 drilling crews in the Iraqi market; after the outbreak of the Middle East conflict… as of June 30, 10 of them were on standby/idle. As of the disclosure date of this report (8/28), 9 crews remain idle, and the timing of restart remains uncertain." → drilling and completion segment revenue was RMB 554 million (−27.30%), with gross margin collapsing from 15.59% to 1.63%; overseas revenue gross margin fell from 29.46% to 10.94%.
  3. High oil prices are themselves a tax and an expense. Taxes and surcharges +45.8% (the higher the oil price, the higher resource tax and the special gain levy), selling expenses +252.1% (transfer and transport costs for unsold crude), and an effective tax rate of 54.1% (27.8% a year earlier).
• Result: H1 net profit attributable to parent of RMB 99 million, −67.03%, ex-non-recurring −67.14%; Q2 alone net profit attributable to parent −4.4% — in a quarter when Brent was +44.7% year over year.
• ⚠️ The same crisis raised the price at which it sells oil while shutting down 9 of its 14 rig crews. Its oilfields are in Central Asia/Xinjiang; its cash cow (drilling services) is in Iraq. Its immediate elasticity to Brent is negative this round.
• Technicals: on 9/2 open→close −6.86%, the worst in the chain; turnover 12.70% (highest of the four), net fund flow −RMB 125 million. Valuation PE 34.82 (94th percentile on 3Y), PB 2.46 — high PE + mid PB, the shape of an earnings squat.
• Final judgment: Pass.

---

6. Pass List

---

7. Intra-Branch Ranking

7.1 Defense branch (9/2 sector +2.06%, ranked 1st, 56 up and 24 down)

• Highest trading recognizability: Inner Mongolia First Machinery (600967) — objectively first in seal quality, and the anchor of the branch.
• ⚠️ But not one name in the whole branch deserves a recommendation slot. Of the five, two have had their fundamentals clearly falsified (BoYun New Materials, Great Wall Military Industry), one has had the catalyst denied by the company itself (Inner Mongolia First Machinery), and two are unverified (Chongqing Jianshe Industry, Shengnan Technology). This is not a question of "which one to pick"; it is a question of "the evidence base of this chain does not support a recommendation."
• Shared risk: the catalyst switch is in Tehran, not Beijing. A single ceasefire headline can falsify all 5 at once.

7.2 Oil & gas branch (9/2 sector −1.30%, 3 up and 16 down)

• The core judgment on this branch has changed. The first draft said "the problem is not stock selection, it is timing"; after verification it is: the problem is both timing and stock selection. The dispersion inside this chain is enormous — upstream (CNOOC Ltd) truly delivered with Q2 net profit attributable to parent +41.6%, while all four oilfield service/equipment names were falsified in the same quarter (SINOPEC Oilfield Equipment swung to a loss, Zhongman −67%, COSL ex-non-recurring −5.48%, Offshore Oil Engineering ex-non-recurring −8.28%).
• ⚠️ The most ironic part: the best performer on 9/2 was SINOPEC Oilfield Equipment (the only limit-up), with the worst fundamentals; CNOOC Ltd, which fell hard, is the only one that actually made the money. That day's market ranking is almost exactly the inverse of the fundamental ranking.

---

8. Opening Verification Signals for Today (9/3)

8.1 Call-auction signals

• Inner Mongolia First Machinery (600967) — today's single most important observation point:
  • The abnormal-movement filing was published at 9/2 18:28 and is not reflected in the 9/2 close. Today's auction is the market's first pricing of "the company denies the catalyst."
  • Opens 0–2% lower, then quickly turns green and seals at the limit: money is ignoring the filing, sentiment still dominates, and the branch continues.
  • Opens more than 3% lower, or gaps up and then dives: the filing is starting to bite, and profit-takers after the 3-day 20% deviation are heading for the exit — a branch-top signal.
  • ⚠️ Yesterday's pre-market lesson has been fixed: none of yesterday's three scenarios covered "a lower open," and a lower open is exactly what happened. This brief states explicitly: a lower open is not necessarily a bad signal (on 9/2 Great Wall Military Industry, BoYun New Materials and Chongqing Jianshe Industry were all pushed to the limit from lower opens); the criterion is "can it turn green before 09:45 after the lower open," not "did it open lower."
• Oil & gas (CNOOC Ltd 600938 / COSL 601808):
  • ⚠️ A gap up = the third run of the same experiment; avoid. Both gap-ups on 8/31 and 9/2 were distributed.
  • Opening flat or lower and then being bought up = the character has changed and it can be reassessed. That is the only shape that can overturn the "distribution" conclusion.
• China State Shipbuilding (600150): after the twin filings, a gap up of less than 2% would be isomorphic to the 8/28 episode (+0.75%), showing the market still will not pay for shipbuilding orders; a gap up of more than 4% that does not fade would be a revision of the 8/28 conclusion and worth reassessing. Also watch whether turnover expands — on 9/2 it did not start at all.
• Feilong Auto Components (002536): the share-sale filing was published at 9/2 18:28 and is not reflected in the close. Whether it plays catch-down today is a natural observation point for "information vs price dislocation."
• Far East Smarter Energy (600869) / Hangzhou Cable (603618): the catalyst has been falsified, so any gap up is a chance to reduce, not an entry point. Hangzhou Cable's seal-to-turnover ratio of 0.038 makes a broken seal likely.

8.2 Sector signals

• Defense confirmed continuing: 3 or more rapid limit-ups in defense equipment before 09:45, with Inner Mongolia First Machinery holding its seal on heavier volume; defense electronics (first market-wide in net inflow on 9/2 at +RMB 190 million) strengthening in step.
• Defense confirmed fading: 2 or more of the 4 defense limit-ups from 9/2 break their seals today, or Inner Mongolia First Machinery cannot re-seal after unsealing.
• Leadership-rotation signal: if the oil & gas chain opens lower today and gets bought while defense gaps up and fades, that is a signal of money shifting from "sentiment" to "physical assets" and requires an intraday reassessment.
• ⚠️ Overall backdrop: 9/2 was already 1541 up / 3901 down, with the 90 sectors showing a combined net outflow of RMB 95.964 billion and only 6 sectors net positive. Against that backdrop any sector's "strength" is relative; do not read a structural opportunity as a broad rally.

8.3 Stock-level signals

• Inner Mongolia First Machinery (600967): absorption volume in the first 5 minutes, whether it can seal quickly, whether the sealing order can hold above RMB 200 million (it was RMB 276 million on 9/2), and whether Shengnan Technology (920006) squeezes it out of the leadership slot.
• China State Shipbuilding (600150): whether turnover expands — this says more about whether the market is starting to accept the two filings than the percentage change does.
• Chongqing Jianshe Industry (002265): whether it can post a second consecutive limit-up without breaking the seal — it is the only main-board name in the defense branch whose fundamentals have not been falsified.

8.4 Risk signals

• Gap up then dive: any defense name gapping up more than 5% and turning red within 30 minutes.
• A lone limit-up: only 1 defense limit-up left today, with all the others fading.
• The anchor does not follow: Inner Mongolia First Machinery breaks its seal while the back row (Shengnan Technology and others) is still pushing higher — the most dangerous shape, usually the end stage of a branch.
• Yesterday's strong theme fading: ⚠️ the cable branch (Far East / Hangzhou / Taiyang / Shenma) will most likely fade today, as the catalyst has been falsified. If they all break their seals and drag the index, that will also suppress risk appetite for defense.
• External: three straight down sessions in U.S. stocks overnight + the 10Y Treasury yield at a three-year high + 68% odds of a Fed rate hike in September — a persistent headwind for ChiNext/STAR growth names (on 9/2 the ChiNext Index was −2.39% and STAR 50 −1.82%).
• ⚠️ Ceasefire risk: on 9/1 the U.S. side proposed a 48-hour temporary ceasefire to Iran through friendly countries, and Iran declined. But this thread can reverse at any time, and it is the sole fulcrum of the defense branch.

---

9. Final Recommendations

① The 5 stocks most worth watching today

⚠️ First, be clear about the character of today's list: after primary-source checks of financial reports and filings, not one of the stocks that hit the limit on 9/2 makes the top two of this list. Only 1 name on the list is tagged "watch closely"; the other 4 are all "watch only." This is not conservatism; it is that the evidence does not support a stronger label.

⚠️ This list is ordered by "observation value today," which is not fully identical to the total-score ranking in §3, and the two inconsistencies are deliberate: Inner Mongolia First Machinery (600967) scores only 34 and Chongqing Jianshe Industry (002265) 43, both below Beijing Creative Distribution Automation (002350)'s 45, yet the first two are included and the latter is not. The reason: Inner Mongolia First Machinery faces the test of "first pricing after the abnormal-movement filing" today, making it the key observation point for whether the whole defense branch stays or goes; Chongqing Jianshe Industry is the only main-board name in the defense branch whose fundamentals have not been falsified. Beijing Creative Distribution Automation did not hit the limit on 9/2 and its order has no formal contract yet, so there is no corresponding verification event today. The total score measures "evidentiary strength," this list measures "is it worth watching the tape today," and the two need not be equal — but it must be stated, not switched silently.

A special note on slots 2 and 3: the verification points for CNOOC Ltd and Offshore Oil Engineering are partly reversed — they are included not because of a bullish view on today's performance, but because they are the best observation points for judging whether the character of the whole oil & gas chain has changed. Execute them as "watch only"; entering on a gap up is not advised.

⚠️ The first draft's No. 3 was SINOPEC Oilfield Equipment; it has been removed and downgraded to Pass. Reason: its 8/28 interim report lists "obstruction of shipping through the Strait of Hormuz" as a drag on results, first-half new orders were −5.1% year over year, and H1 net profit attributable to parent swung to a loss. It is the only oil & gas chain stock bought to limit-up market-wide on 9/2, and also the one with the worst fundamentals on that chain.

② The 3 strongest branches today

③ Directions not worth chasing today, and why

1. Cable / power grid (Far East Smarter Energy 600869, Hangzhou Cable 603618, Taiyang Cable 002300, Shenma Power 603530) — the catalyst does not survive back-computation. Far East's "RMB 1.780 billion of winning bids" is a routine disclosure made on the 1st of every month on a full-month cumulative basis, and it is down 36.6% from July's RMB 2.808 billion; the other three had no new filings around 9/2. Hangzhou Cable's seal-to-turnover ratio of 0.038 is the flimsiest seal in this brief. The combination of the heaviest money and the weakest evidence market-wide.
2. Great Wall Military Industry (601606) — military-product revenue −50.65%, three consecutive years of ex-non-recurring losses, receivable days 516. The company itself issued a "market sentiment overheated" / "overvaluation" warning on 7/28 (PB was 11.70 then); PB is now 12.39, higher still.
3. BoYun New Materials (002297) — the defense label is mismatched. 78.6% of revenue is cemented carbide, the growth comes from tungsten carbide price increases and Q2 was already −71.2% sequentially, with RMB 89.11 million of onerous contracts provisioned; it is not even in the CSI Defense constituent list. "Low PE (0.9 percentile) + high PB (97.4 percentile)" is the classic shape of a cyclical top.
4. Feilong Auto Components (002536) — on the limit-up day itself, an 18:28 filing disclosed a block sale of 1.635 million shares by the controlling shareholder, and that information is not yet reflected in the 9/2 close.
5. SINOPEC Oilfield Equipment (000852), Zhongman Petroleum (603619) — the reason for buying points in the opposite direction from the operating reality the companies describe. SINOPEC Oilfield Equipment's interim report lists "obstruction of Hormuz shipping" as a drag, new orders −5.1%, H1 swung to a loss; Zhongman Petroleum had 9 of its 14 Iraqi rig crews shut down by the same conflict, H1 net profit attributable to parent −67.03%, and it has just announced on 9/2 that the former actual controller plans to sell his entire 1.20%.
6. Refining (Sinopec 600028 and peers) — the direction was got backwards at the financial-statement level. For refiners crude is a cost: in Q2 (Brent +44.7%) Sinopec's net profit attributable to parent was only +4.9% year over year and −49.3% sequentially, while CNOOC Ltd was +41.6% over the same period. ⚠️ But note: this is a financial-statement conclusion, not the cause of that 9/2 candle — that day CNOOC Ltd (−2.14%) actually fell more than Sinopec (−1.63%), indicating 9/2 was profit-taking-driven. Do not mix the two judgments.
7. Coal — at the 9/2 open "coal mining" was still on the top-gainers list, but it closed −3.53%, dead last among the 90 sectors, 0 up and 34 down. "Oil rises → coal substitution" was already falsified in full yesterday.
8. Shenzhen China Bicycle A (000017) — a cumulative price deviation of 100% over 10 consecutive days, and it has announced it may apply for a trading halt for verification.
9. High-priced tech growth (ChiNext/STAR) — the 10Y Treasury yield is at a three-year high and the odds of a Fed rate hike in September are 68%; on 9/2 the ChiNext Index was −2.39%, STAR 50 −1.82%, and Innolight had the largest main-force net selling market-wide at RMB 3.066 billion. This is a macro headwind, not a sentiment wobble.

④ Final judgment in one sentence

Today's real discovery is not "which line is strongest" but "the two strongest lines both fail under scrutiny" — and they fail in exactly the same way: the market's reason for buying points in the opposite direction from the operating reality the companies themselves wrote down.

The defense line: Inner Mongolia First Machinery, the company concerned, denied the catalyst in writing inside the window ("the external market environment and industry policy have likewise not changed materially"), and its only geopolitically transmissible business, military trade, is −92% year over year; BoYun New Materials' profits come from tungsten carbide price increases rather than military products, Q2 was already −71.2% sequentially, and it is not in the CSI Defense constituent list; Great Wall Military Industry's military revenue halved by 50.65%, it has three consecutive years of ex-non-recurring losses, and the company itself warned in July of "overheated market sentiment" and "overvaluation." And the "CSI Defense H1 net profit +32.92%" that props up the whole narrative is distorted once all 80 constituents are summed one by one — same-caliber revenue is +7.81%, not +18.89%, and excluding China State Shipbuilding the other 79 constituents' net profit attributable to parent is −5.45%, i.e. negative growth.

The oil & gas line: SINOPEC Oilfield Equipment, the only "Hormuz beneficiary" bought to limit-up market-wide on 9/2, lists "obstruction of shipping through the Strait of Hormuz" as a drag on results in its own 8/28 interim report, with new orders −5.1% and H1 swinging to a loss; while Zhongman Petroleum, the one that "owns its own oilfields," had 9 of its 14 Iraqi rig crews shut down by the same conflict, with H1 net profit attributable to parent −67.03%. More importantly: the +4.90% Brent up-candle was 9/1, not 9/2, and inside this brief's window oil was only +0.15% — even the premise of "a new catalyst" does not hold.

Which produces the biggest contrast in this brief: the companies on those two lines that actually made money are precisely the two nobody bought yesterday — China State Shipbuilding (600150), one company accounting for 24% of CSI Defense Index revenue and holding twin in-window filings, fell 1.45% yesterday; CNOOC Ltd (600938), with Q2 net profit attributable to parent +41.6% and, together with PetroChina (601857), one of the two upstream names that truly delivered, fell 2.14% yesterday — more than the refiner did.

On execution: today is a structural-defense day, not an offensive one. With 1541 up / 3901 down, a market-wide net outflow of RMB 95.964 billion and 68% odds of a Fed hike overnight, cutting position size matters more than picking the right direction. Of the 5 slots on this list only 1 is tagged "watch closely" and 4 are "watch only," and 2 of those have reversed verification points — this is not conservatism; it is that after verification the evidence can only support conclusions this strong. The trading structure in defense is genuinely the best market-wide (Inner Mongolia First Machinery (600967) with a seal-to-turnover ratio of 0.326), and this brief neither denies nor evades that — but "money is buying" and "worth buying" are two different judgments, and this brief is responsible only for the latter.

---

10. Verification Log: Key Facts Overturned or Corrected

10.1 "CSI Defense H1 net profit attributable to parent +32.92%" — the numerator is right, the denominator is wrong, and the growth rate is materially overstated

Summing the current 80 CSI Defense (399967) constituents one by one (sample 80/80, none missing):

The two absolute 2026H1 figures match the media caliber (RMB 377.494 billion / RMB 24.772 billion) digit for digit, which shows the numerator and the constituent list are both right. The discrepancy is entirely in the base period: the 2025H1 base implied by the media's +18.89% is RMB 317.52 billion, RMB 32.63 billion lower than the same-caliber base.

That RMB 32.63 billion has been traced to a single company: China State Shipbuilding (600150). Its originally disclosed 2025H1 revenue was RMB 40.325 billion; its absorption of China Shipbuilding Industry Corp (formerly 601989) completed in September 2025 as a business combination under common control, so prior-period statements were restated retrospectively, and on the current caliber 2025H1 is RMB 72.640 billion, a difference of RMB 32.315 billion. Recomputing with the base swapped back to the originally disclosed value gives RMB 317.83 billion, only 0.10% away from the media's implied base.

Conclusion: about 11 percentage points of the +18.89% were manufactured by a consolidation-caliber mismatch, not by organic industry growth. More importantly, the structure:

• Excluding China State Shipbuilding, the other 79 constituents' revenue was +3.05% and net profit attributable to parent −5.45% (negative growth)
• The median stock-level revenue growth was only +0.58%, with 42/80 growing and 13/80 loss-making in 2026H1
• China State Shipbuilding alone accounts for 24% of index revenue, with the second-largest constituent Chujiang New Materials (002171, copper processing) at 11%

The narrative of "across-the-board improvement in defense earnings" is carried mainly by one commercial shipbuilder's ship cycle plus one merger restatement.

10.2 "2 institutions net bought RMB 136 million of BoYun New Materials" — the number is right, the attribution is wrong

The RMB 136 million is the all-seat net buy. After netting the four institutional seats, institutional net buying is only RMB 18.70 million (13.7%), while two hot-money seats total RMB 120.48 million (88.3%). Seat details are in §5.10. Also: Inner Mongolia First Machinery appeared on the list the same day with RMB 212 million of net buying, all from ordinary seats and zero institutional seats, and the reason for its appearance was "cumulative price deviation of 20% over three consecutive trading days" (i.e. it had already hit the limit on 9/1); Great Wall Military Industry did not make the 9/2 Dragon-Tiger List.

10.3 The publication time of Inner Mongolia First Machinery's abnormal-movement filing — the sub-agent recorded 9/3; it was actually 9/2 18:28

The sub-agent trusted CNINFO's / East Money's "next trading day" tag. Re-checked via art_code as AN202609021828930975 = 2026-09-02 18:28. The direction of this correction is favorable to the conclusion — the filing falls inside this brief's window and is new information not yet reflected in the 9/2 close, making the evidence stronger than a "published 9/3" version. This is also why caliber boundary ② exists: the same trap was stepped on just yesterday in the recap.

10.4 "Brent +4~5% on 9/2" — the price level is right, the gain was attributed to the wrong day (this brief's most important self-correction)

Following the media, the first draft wrote "Brent broke through 94–95 on 9/2, +4~5%." Measured via futures_foreign_hist('OIL'): the +4.90% close at 95.11 was 9/1; 9/2 closed at 95.25, only +0.15% (intraday high 97.04). That removes outright the premise of a "new catalyst" for the oil & gas branch inside this window — what A-shares reacted to in the 9/2 morning session was the 9/1 overnight jump, and it was digested that same day in a "gap up then fade." Without the correction, this brief would have treated an already-digested old catalyst as a new positive for today. Corrected at the same time: the claim that Brent "briefly hit 126 dollars" during the first round of the crisis could not be confirmed; the measured 2026 intraday high is 119.50 (3/9), and COSL's own interim report describes "a sharp retreat from the year's high of nearly US$120/bbl" — two independent sources cross-confirming.

10.5 "Oil rises → oilfield services and equipment benefit" — last round's interim reports already falsified it once

The previous oil spike covered exactly H1 (Q2 average Brent US$96.02, +44.7% year over year), and the four oilfield service/equipment names delivered: SINOPEC Oilfield Equipment swung to a loss, Zhongman Petroleum −67.03%, COSL ex-non-recurring −5.48%, Offshore Oil Engineering ex-non-recurring −8.28%. What actually delivered were the upstream equity-production holders: CNOOC Ltd H1 +23.4% (Q2 +41.6%), PetroChina +22.0%. On the price side, all six topped out on 3/3–3/5, whereas Brent's intraday high was 3/9 and its highest close 5/4 — the share-price peak leads the commodity peak by about 8 weeks. "New highs in oil" and "new highs in oil-service stocks" fully decoupled last round.

10.6 A rebuttal of this brief itself: the 9/2 moves cannot be explained by "cost vs revenue"

§2③ of this brief argues that refining is the cost-side loser from high oil prices (supported by quarterly data: Sinopec Q2 net profit attributable to parent −49.3% sequentially, versus CNOOC Ltd +19.2% over the same period). But the 9/2 tape does not support using that framework to explain the day's moves — CNOOC Ltd (pure upstream, the biggest beneficiary) fell −2.14% that day, more than Sinopec (refiner, −1.63%). If the market had been pricing oil economics that day, those two numbers should be reversed. So 9/2 is more likely profit-taking / positioning-driven. The financial-statement conclusion and the attribution of that day's candle are two different things, and this brief keeps them separate in the main text rather than mixing them.

---

⚠️ Risk warning: this list is a pre-market information review and observation note only and does not constitute investment advice. A-share volatility risk is extremely high, and automatically generated content may contain timeliness gaps or industry-chain mapping errors; it must not be used directly as a basis for trading.