A-Share Pre-Market Brief | 2026-09-04 Friday

Caliber boundaries of this brief (judge the strength of its conclusions accordingly) ① News window = 2026-09-03 (Thursday) 15:00 close → 2026-09-04 07:00, all times Beijing time. ② A-share quotes are on the Tencent qt.gtimg.cn 9/3 closing-settlement basis, timestamps 20260903161415–20260903161457, all after the close. ③ U.S. equity and ETF quotes are measured from the CNBC quote endpoint, timestamps 2026-09-03T16:00–17:15 ET, i.e. the official U.S. 9/3 (Thursday) closing prices. ⚠️ See the note in §1: several Chinese-language "U.S. market close" articles retrieved were reporting the 9/2 session, whose conclusions run in the opposite direction to the 9/3 session inside this window — do not mix the two. ④ Treasury yields are taken from the U.S. Treasury's official daily yield-curve CSV (CNBC's Treasury change fields are not accepted). ⑤ The limit-up pool and the Dragon-Tiger List are on East Money's 9/3 basis; the Dragon-Tiger List has 63 rows, deduplicated by code and then sorted. Sectors are the Tonghuashun 90-industry-sector 9/3 closing snapshot. ⑥ Limit-ups are judged by board-specific price limits: main board ±10%, ChiNext / STAR Market ±20%, BSE ±30%. Every stock in this brief is tagged with its board. ⑦ Not obtained in this brief: confirmable public catalysts on 9/3 for the shipping sector and the insurance sector (three rounds of search returned nothing but stale articles). Handled per "no news found ≠ no news": §2 says "no confirmable catalyst obtained" rather than "no catalyst," and both are listed as must-check items for today. ⑧ No northbound flow data (daily net buying has not been disclosed since 2024-08-19); this brief gives no northbound figures.

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

I. Channels that worked

• U.S. equities / ETFs / precious-metals ETFs via the CNBC quote endpoint, 15+24 symbols returned in one shot; last_time was checked symbol by symbol and is 2026-09-03T16:00–17:15 ET throughout, confirming this is the 9/3 close and not 9/2.
• Treasuries via the Treasury's official CSV (daily_treasury_yield_curve). ⚠️ The first retrieval was wrong; see the engineering defect in section VII below. After correction: 9/3 is 2Y 4.34 / 10Y 4.77 / 30Y 5.25; 9/2 is 2Y 4.39 / 10Y 4.79 / 30Y 5.27.
• A-share single stocks via Tencent qt.gtimg.cn, 4 batches totalling 60+ names, all returned, 0 timeouts.
• Limit-up pool stock_zt_pool_em(20260903) 44 rows, consistent with the 44 rows in yesterday's recap.
• The Dragon-Tiger List stock_lhb_detail_em(20260903) successfully returned 63 rows this time — when yesterday's recap was generated at 15:30 the endpoint threw TypeError: NoneType; this brief was generated at 07:00, past the 18:00 publication point, and got the data. This confirms yesterday's judgment that it was a scheduling problem, not an endpoint failure.
• Sectors via stock_board_industry_summary_ths() 90 rows, matching yesterday's recap row by row (insurance +2.56%, precious metals +2.40%, ports & shipping +1.97%), confirming this is the 9/3 closing snapshot.

II. Channels that failed

• ⚠️ yfinance / query1.finance.yahoo.com returned HTTP 429; the GLD history series could not be retrieved. Switched to CNBC's yrhiprice/yrhidate fields plus cross-validation against an independent Chinese source (see item 3 below).
• stockanalysis.com/etf/gld/history/ regex parsing returned 0 rows (the page structure appears to be JS-rendered).
• stooq.com XAUUSD/XAGUSD CSV parsing failed (ValueError: too many values to unpack, column count not as expected).
• ak.stock_board_industry_cons_ths does not exist in akshare 1.18.79 (AttributeError); the precious-metals constituent list was replaced with a hand-built code list plus a Tencent batch pull.
• The tradingkey article returned HTTP 403; the myzaker article has been taken down; WebFetch on the 21jingji article produced no output (substituted with a JRJ article on the same topic).
• ak.stock_zt_pool_em has no 涨停价 column; the first call raised KeyError and was switched to the actual column names.

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

1. ⚠️⚠️ The "September 4 front-page digest of the four securities newspapers" returned by search was confirmed via WebFetch to have been published 2025-09-04 (the URL contains 202509043503876414, and its "London gold hits a record high above US$3,561" is a 2025 price level; in 2026 gold is around 4,400). The entire article was discarded and not a single item from it was used. This is the year-level variant of "silent lag" — lagging by a full year rather than a day.
2. ⚠️⚠️⚠️ "Gold hits a record high" was falsified — the single most important cross-check in this brief. A 9/3 article retrieved said "gold continues to set records and is performing strongly." Measured on CNBC: GLD at 410.22,52-week high 509.70 @ 01/29/26 (−19.5%); SLV at 60.55,52-week high 109.83 @ 01/29/26 (−44.9%). Cross-validation against an independent Chinese source matches exactly: spot gold hit its all-time peak of US$5,598.75 on 2026-01-29 (the date aligns precisely with the CNBC field), crashed 9.66% in a single day on 3/23 down to 4099,8/12 quoted at 4,388, "roughly 22% below the year's high." Spot is currently 4,475.70 (measured from gold-api, timestamp 2026-09-03T23:34Z), about −20% from the peak. Had "record high" been copied through, the entire precious-metals branch would have been characterised as a "breakout to new highs" instead of "a rebound inside a deep drawdown," and the risk calibration would have been exactly inverted. ⚠️ Note: the first draft wrote "spot 4406, drawdown −21%," taken from a Chinese article's intraday reading, which contradicted this brief's own GLD −19.5%; the QA pass caught it and it has been corrected to the measured value. The lesson: when two numbers in the same piece can be back-computed against each other, you must do that back-computation on the spot — 5,598.75 × (1−19.5%) ≈ 4,506, which is US$100 away from 4,406, and that contradiction sat in the first draft for an entire piece without being noticed.
3. ⚠️ The trading-day check on overnight U.S. equities overturned the search conclusion. Several Chinese articles returned by search ("semiconductor and memory names broadly up, Micron, Qualcomm and SK Hynix up more than 2%", "Broadcom's bold guidance") were checked via WebFetch for publication time (9/3 06:37, 9/3 08:35) and all report the 9/2 session. The 9/3 session measured on CNBC runs in the opposite direction: SOX +0.11%, badly lagging the Nasdaq's +1.40%; AVGO −2.74%, WDC −1.63%, STX −1.23%. Copying it through would have written a "memory sell-off day" as a "semis-rally day," and that is precisely the direction most in need of avoidance today.
4. ⚠️ The antimony-price check dismantled a reason that was about to earn Hunan Gold extra points. The plan was to award Hunan Gold points for a "gold + antimony double play"; after verification, September antimony prices are flat (1# antimony ingot RMB 190,000–200,000 per tonne), and since June there has been "heavy raw-material import pressure and oversupply." Antimony is not an additional catalyst; that score has been withdrawn.
5. Lab-grown diamonds not accepted. Henan Huanghe Whirlwind (600172) was +7.93% on 9/3 with 30.15% turnover and a Dragon-Tiger net buy of RMB 282 million, but searches show the same "diamond heat-spreader / 8-inch heat-sink wafer" narrative has been used repeatedly in February, March, May and June 2026, with no new catalyst inside the window. Handled per "AI attribution stitches old filings" and placed on the Pass list.

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

• No confirmable same-day 9/3 catalyst obtained for the shipping sector. Three rounds of search (COSCO Shipping Energy / China Merchants Energy Shipping / VLCC / BDI / sanctions) returned nothing but 2025-09, 2026-02 and 2026-03 articles, not one of which can be anchored to 9/3. But the price action is hard (COSCO Shipping Energy +8.77%, China Merchants Energy Shipping +8.37%, Haitong Development limit-up), so it is written per "not found ≠ not there" and listed as a must-check for today.
• No confirmable same-day 9/3 catalyst obtained for the insurance sector (the same open item as yesterday's recap, now unresolved for a second consecutive day). Searches return only interim reports and the medium-term high-dividend logic.
• On the CME FedWatch September hike probability, searches returned four different values — 66.1%/62.2%/58.4%/44.4% — none of which can be anchored to a timestamp, so this brief gives no specific probability figure, only a direction.
• The fundamentals of the 44 limit-up stocks were not checked one by one.
• JCET Group's "advanced packaging as a share of revenue" was not obtained. The sub-agent found 59 occurrences of "advanced packaging" across the interim report and the placement plan, none of which disclose a revenue-share figure. This brief therefore gives only the capex allocation share and subsidiary data, and gives no "advanced packaging = X% of revenue" number of any kind.
• Chifeng Jilong Gold's actual 2026H1 mined gold output was not obtained: Q1 has been disclosed at 2.98 tonnes; a search yields "H1 roughly 6.75 tonnes," which is a management-communication figure; backing out from revenue gives roughly 5.95 tonnes. The three cannot all be true, so this brief uses none of them as an absolute output figure.
• The attribution for Xiaocheng Technology's Q2 swing to an ex-non-recurring loss (−RMB 4.89 million) was not obtained; it requires reading the interim report's cost / impairment / Ghana FX detail.

V. Sub-agent verification results (both fundamentals-analyst runs returned, 39 + 55 tool calls in total)

⚠️ The sub-agents overturned or corrected 5 judgments in this brief's first draft, 2 of them substantive errors:

1. ⚠️⚠️ [The most serious self-correction in this brief] The first draft ranked Zijin Mining No. 1 on the grounds of "the lowest PE in the whole table, 13.1." After the sub-agent pulled three-year valuation percentiles for all five gold miners: PE percentiles are all in the 9.4%–38.5% historical-low range, while PB percentiles are all in the 82.3%–94.5% historical-high range. This is exactly the same shape as the memory item "low PE + high PB percentile = cycle top" (last time it was the four memory names, PE percentile 3%–7% / PB percentile 70%–82%). "The lowest PE in the whole table" is, at a cyclical earnings peak, a danger signal rather than a margin of safety. The fundamentals component of the §4 scoring model has been switched from PE to a PB-percentile-first basis, the whole table recomputed, and Zijin dropped from No. 1 to No. 4 with a −10 score adjustment.
2. ⚠️⚠️ [A missed in-window filing] Xingye Silver & Tin (000426) filed on the evening of 9/3 (filing no. 2026-81), and this brief's first draft did not find it at all. The content: Yinman Mining, which contributes 46.7% of group revenue and 55.8% of ex-non-recurring profit, has had its mining area shut for more than 5 weeks since the 7/26 accident (1 fatality); on 9/3 only the concentrator resumed, the mining area has still not restarted. This is an S-grade negative inside the window, yet the first draft listed the stock as "watch only." My searches were run on topic words like "gold / silver / sector" and I did not pull company filings for each candidate stock one by one — the same failure mode as the memory item "no news found ≠ no news," and this time it was a company-level filing rather than market news, which makes missing it worse. Downgraded to Pass.
3. ⚠️ [Correcting an unfair charge] The first draft wrote that Xiaocheng Technology's "gold business is of doubtful reality — the name looks right but the main business is not gold." That is wrong. The sub-agent verified: Ghana gold-mine revenue has been 82%–94% of the total for two consecutive periods, gross margin 63%, overseas revenue 98.6% — the business is real. The real problem is a scale mismatch (RMB 295 million of half-year gold revenue vs a RMB 14.5 billion market cap) and zero institutional participation (9/3 extra-large orders −RMB 0.1 million). The conclusion is still Pass, but for an entirely different reason, and it has been rewritten.
4. ⚠️ [Found the real fake] Hunan Silver (002716) is the one whose "name looks right but the main business is not": smelting and processing is 97.06% of revenue at a 4.74% gross margin, with only 2.94% from its own mines. And the direction is inverted — H1 inventory write-downs of RMB 93.02 million (58.9% of net profit attributable to parent), with the filing itself saying costs are "clearly inverted against carried inventory cost." When silver falls, it is on the losing side. The first draft's Pass reason was "turnover overheated," which is not accurate enough; it has been changed to a business-model mismatch.
5. ⚠️ [The risk of writing the silver reference frame backwards was intercepted] The first draft leaned toward using "silver is −45% from its January peak" to argue bearishly. From Xingye Silver & Tin's interim report: the 2026H1 average COMEX silver futures price was US$78.760, +137.5% year over year, and the interval high of 121.785 falls inside the reporting period — the peak was consumed by the H1 statements, it is not outside them. "The base is a −45% low" and "the base is a +137% high" are two narratives pointing in exactly opposite directions, and the text has been rewritten per the latter.

VI. Key incremental findings from the sub-agents that this brief accepts (absent from the first draft)

• Single-quarter decomposition (Q2 = interim − Q1): the five gold miners were +79% to +177% year over year in Q1, and Q2 collapsed across the board to +12% to +48%, with every one of them down sequentially by −17% to −39%. This depends on no external information whatsoever; it is pure arithmetic.
• The gold-price base crossover point: the SGE Q4-2025 base of RMB 949.9/gram vs the current RMB 957.50 is only +0.8% apart → the Q4 year-over-year dividend goes to zero; 2027H1 turns negative at −7.7%. Stacked with nationwide H1 raw-material gold output of −14.62%, mine-segment revenue could turn outright negative year over year from Q4.
• "Smelter names have low elasticity" is a misjudgment: smelting is 91.4% of Zhongjin Gold's revenue but at a gross margin of only 4.20%, while mining is 31.6% of revenue yet contributes 95.7% of gross profit. The profit is, in substance, mining.
• "Purest ≠ most elastic": Sichuan Gold is 100% pure gold concentrate, but PB 11.14 and only 7.2% gold exposure per RMB of market cap, one third of Zhongjin's (21.8%).
• Three details on JCET's placement: actual dilution is 5.1%–6.4%, not 30% (30% is the regulatory formal ceiling); China Resources' side is subscribing 22.53% at the same book-built price with an 18-month lock-up; and H1 free cash flow of −RMB 1.7 billion is the real motivation for the placement.
• Second confirmation of the gold-price peak: SGE Au99.99 peaked at RMB 1,243.02/gram on 2026-01-29, and COMEX Dec'26's 52-week high of $5,586.20 was 01/29/26 — two independent markets confirming the same day, so the "−21%" in §0 holds (range −19% to −23%, depending on caliber). One further correction: 3/23 on the SGE was actually −11.23%, and the −11.51% on 2/2 is the largest single-day drop.

---

0. Today in One Sentence

The strongest catalyst is the wholesale repricing of the precious-metals complex in overnight U.S. trading: gold-miner ETF GDX +3.95%, silver ETF SLV +2.51%, gold ETF GLD +1.85%, silver miners SIL +2.71%, with the direction being that the market pared its bets on a Fed rate hike in September (⚠️ this is a hike, not a cut: the fed funds target range has been held at 3.50%–3.75% since July for a fifth consecutive meeting, Chair Kevin Warsh leans hawkish, and CME FedWatch on 8/31 put a 66% probability on a hike at the 9/16 meeting; at the 9/3 close the 2-year Treasury at 4.34% sits roughly 71bp above the midpoint of the funds rate, independently corroborating net-hike pricing).

⚠️ But the word "catalyst" needs to be used with restraint — this brief could not anchor the trigger event to 9/3 itself. The frequently cited August ADP print (+38,000 vs 47,000 expected) was released before the U.S. open on 9/2, and the 9/2 session traded it in full. The direct source of the 9/3 jump (jobless claims, ISM services, Challenger layoffs or a Warsh speech) was not obtained in three rounds of search, and is listed as a must-check for today. What is known is the price fact; what is unknown is the cause — a distinction that matters greatly for judging whether it can persist.

But the single thing most worth remembering today is a fact that is being widely written wrong: gold has not made a new high. Spot gold is at $4,475.70 (measured from gold-api, timestamp 2026-09-03T23:34Z = Beijing 9/4 07:34), while the all-time peak was on 2026-01-29, and it is still in a deep drawdown of roughly −20%. The peak date is corroborated by three independent fields (GLD 52-week high @01/29/26, SLV @01/29/26, gold futures @GC.1 52-week high 5,586.20 @01/29/26); the precise peak level used in this brief, 5,598.75 (spot basis, a secondary source), could not be independently verified, so the drawdown is given only as the range conclusion "about −20%." This is a rebound after a deep fall, not a breakout. Sizing positions as if it were a "new-high regime" would put the risk calibration off by a full notch.

⚠️ And that up-candle is already being given back as this goes out: gold futures @GC.1 settled at 4,539.90 on 9/3 and were then quoted at 4,522.70 (−0.38%) at 19:11 ET, with spot falling in step from about 4,490 to 4,475.70. Before the A-share open, part of the overnight precious-metals gain has already been shaved off — use the latest values during the auction, not the 9/3 closing snapshot.

The direction of flows is most likely "out of compute and into precious metals," but you have to pick the right market-cap band. The Tonghuashun precious-metals sector was +2.40% on 9/3, second among 90 sectors, while the internal structure of the gold and silver names is a textbook small-cap hot-money move: Baiyin Nonferrous (601212) +10.06% limit-up, Hunan Silver (002716) +7.81% on 15.14% turnover, whereas Shandong Gold Mining (600547) gapped up +1.48% and closed down −1.03%, Zijin Mining (601899) was only +1.09% and Shanjin International Gold (000975) only +0.60%. What surged overnight in the U.S. was precisely the large miners (GDX +3.95%), so today's incremental information favours the large caps that did not move yesterday, not the small caps that have already gone wild.

⚠️ Caliber disclosure (the caliber has to be stated): this brief's "gold and silver universe" is self-built and is not the same yardstick as the Tonghuashun "precious metals" sector. On verification, Baiyin Nonferrous belongs to Tonghuashun's "industrial metals" sector (which was +1.01%, with Baiyin Nonferrous its leading gainer), and Zijin Mining is likewise not a "precious metals" constituent; the leading gainer in Tonghuashun precious metals is Hunan Silver at +7.81%. The +2.40% above and the stock list therefore do not constitute a same-caliber argument about "sector internal structure"; what they argue about is the structure of this brief's self-built universe. The conclusion (large caps have not moved, small caps have gone wild) holds within the self-built universe, but do not treat +2.40% as that universe's gain.

⚠️ But the independent fundamentals verification demands that the "duration" of this branch be shortened by a full notch — this is where this brief diverges most from the mainstream market narrative. After a single-quarter decomposition of the five gold miners (Q2 = interim − Q1, pure arithmetic, dependent on no external information):

• All five were +79% to +177% year over year in Q1 net profit attributable to parent, and all five collapsed to +12% to +48% in Q2, with Q2 profit down sequentially by −17% to −39% without exception. The "interim net profit up 46%–107%" the market keeps quoting is a Q1 legacy, not the current operating slope.
• All five display "PE three-year percentile 9%–39% (historical low) + PB three-year percentile 82%–95% (historical high)." The meaning of that combination is unambiguous: the PE is low not because the share price is cheap, but because the denominator E sits at a cyclical high. This is the textbook shape of a cyclical top, not a "valuation trough."
• The year-over-year gold-price dividend is about to go to zero: the SGE Q4-2025 base is already RMB 949.9/gram, versus a current 957.50, only +0.8% higher. Stacked with nationwide H1 raw-material gold output of −14.62% and four of the five posting negative year-over-year sales volume, mine-segment revenue could turn outright negative year over year from Q4-2026.

⚠️ This brief therefore withdraws an error of its own: the initial judgment ranked Zijin Mining No. 1 on the grounds of "the lowest PE in the whole table, 13.1." Against a backdrop of PB percentiles broadly at 82%–95%, using a low PE to argue cheapness is the low-PE trap itself. That reason has been voided; the §3 scoring model now includes "PB historical percentile" in the fundamentals component and the ranking has been rebuilt accordingly (see §3, §7, §9).

Driver type: overseas macro-data driven (not policy, not orders, not earnings). The defining feature of this kind of driver is that it can be falsified the same day. Conclusion: this is a rebound trade you can participate in, but not an industry theme you can hold on an "improving cycle" basis.

Pre-market stance: cautiously constructive, but the whole day is a "front-running window," not a "confirmation window." U.S. August non-farm payrolls are released tonight at 20:30 Beijing time, with expectations of +58,000 and a 4.1% unemployment rate (prior: July at −23,000) — this is the direct referee for the current precious-metals rebound, and it is revealed after the A-share close. Any gain in precious metals today is a bet on the data, and Friday adds the cost of carrying positions over the weekend, so it should not be traded as an established trend.

---

1. News Overview

⚠️ The first item in this section is a directional warning, so read it first: several Chinese articles returned by a search on "U.S. market close" (including "memory and semis broadly up," "Micron, Qualcomm and SK Hynix up over 2%," "Broadcom's guidance") were checked for publication time and report the 9/2 session, not the 9/3 session inside the window. The two sessions lead to opposite conclusions. Rows 1–4 below are on the CNBC-measured 9/3 close basis.

---

2. Strongest Positive Branches, Descending

Branch notes:

① Precious metals — the only "hard" branch, but you have to pick the right market-cap band. Why the positive holds: transmission from the gold price to own-mine gold producers has no intermediate link and requires no "concept mapping." Overnight, GDX +3.95% beat GLD +1.85%, meaning the market is paying an operating-leverage premium for the miners, which is a sign of branch quality. Short-term or medium-term: short-term. The driver is a single employment print, and tonight's payrolls are its referee. Already fully priced? By market-cap band, the answer flips completely. See the table below — the 9/3 gain is almost perfectly inversely correlated with market cap:

This table is the core basis for this brief's judgment: what went up yesterday were small-cap, high-turnover, high-PE (indeed loss-making) speculative vehicles; what went up overnight were the large overseas miners. These are not the same pool of money and not the same logic. If you follow today, follow the large caps playing catch-up, not the small caps that have already gone wild. One-day-wonder risk: high. Baiyin Nonferrous hit limit-up on a PE of −118.5 (loss-making) and made the Dragon-Tiger List with RMB 243 million of net buying — a textbook theme-speculation pattern.

⚠️ Fundamentals verification results (independent financial verification, akshare interim basis 2026-06-30, independently recomputed from market cap / TTM net profit attributable to parent and price / BPS, all matching item by item):

This table overturns three common judgments:

1. The "high interim growth" is a Q1 legacy. All five collapsed to +12% to +48% year over year in Q2, and profit fell sequentially at every one of them. The SGE gold price was −9.0% sequentially in Q2, and Q3 to date is another −7.0% sequentially — in a cost-rigid mine model, a −7% sequential price move gets amplified by operating leverage into a larger profit decline, so Q3 will most likely extend the sequential downtrend seen in Q2.
2. "The purest gold miner = the greatest elasticity" is wrong. Sichuan Gold is the only 100% pure gold-concentrate name, but it trades at 33.3x PE / 11.14x PB, and every RMB of market cap buys only 7.2% of gold-price exposure, one third of Zhongjin Gold's (21.8%). Business purity ≠ investment elasticity; valuation sits in between.
3. The "gold-antimony double play" was in fact "gold up, antimony down" in 2026H1. Hunan Gold's antimony gross profit was −22.1% year over year, with the gross margin falling from 50.0% to 37.3% — antimony was a net drag in the first half, not a contributor. This corroborates the "September antimony flat, oversupply" finding earlier in this brief, and the stock's bonus points have been withdrawn a second time.

A structural difference that is widely overlooked — silver's base pressure lags gold by one step: the SGE silver price is +61.5% year over year in Q3 to date, versus only +16.1% for gold. Silver-related gross profit is 21.5% of Shanjin International Gold's total, and its mined silver output at +7% is the only metal output among the five to grow(mined gold −12%). This is the one place it is differentiated in this round.

⚠️ The reference frame for silver has to be right, or you will reach the opposite conclusion. Intuitively one would use "silver is −45% from the January peak" to look bearishly at the silver names' interim results. That is wrong. Xingye Silver & Tin's interim report discloses: the 2026H1 average COMEX silver futures price was US$78.760/oz, +137.5% year over year, with an interval high of 121.785 and low of 55.539 — meaning that peak occurred inside the reporting period and the H1 statements have already consumed it. Independently recomputing with SGE Ag(T+D) matches (H1 average RMB 19,555.8/kg, +139.3% year over year). So the high base behind the interim growth is not "a −45% low" but "an +84% to +139% high average price." The real risk is in the next period: the average silver price in Q3 to date is already 22.7% below the H1 average. This is isomorphic to gold's problem — the danger is not how much it has fallen now, but that the year-over-year dividend is going to zero.

Another caliber reminder (cross-market read timestamps): the SLV +2.51% cited in this brief is on the U.S. 9/3 close basis. And the A-share 9/3 close (Beijing 15:00 = 03:00 ET) happened before this U.S. session's rally — that is, the A-share precious-metals gain on 9/3 does not contain this overnight jump. That is exactly the timing basis for this brief's judgment that "the overnight information has not yet been priced into A-shares."

② Crude tankers — the hardest price action, the weakest evidence chain. On 9/3 COSCO Shipping Energy (RMB 111.2 billion market cap) was +8.77% and China Merchants Energy Shipping (RMB 155.8 billion) +8.37%; large caps rising more than 8% on volume can hardly have had no catalyst. But three rounds of search failed to anchor any public news to 9/3. Per the iron rules, this brief writes "no confirmable catalyst obtained" rather than "no catalyst," and explicitly lists it as the first thing to check at today's open. A key distinction: this round is tankers/dry bulk, not container shipping. Container leader COSCO Shipping Holdings (601919) was up only +0.73% that day, and SCFI data show Europe lanes weakening. Calling this line "a shipping recovery" would mask the internal divergence. The path to earnings delivery is clear: tankers are a textbook freight-rate → profit high-elasticity industry, and valuations are currently extremely low (China Merchants Energy Shipping PE 14.4, COSCO Shipping Energy PE 16.6).

③ Insurance — a second consecutive day with no same-day catalyst to be found. The sector was +2.56% on 9/3, first overall. ⚠️ This brief's initial judgment called it "the branch with the best breadth, 5 up and 0 down"; that claim is retracted — the Tonghuashun insurance sector has only 5 constituents in total, so a clean sweep is a small-sample illusion, not comparable to ports & shipping's 28 up, 5 down. Yesterday's recap already listed "insurance catalyst not obtained" as a must-check, and today's searches again returned only interim results (the five majors' half-year net profit of RMB 317.4 billion, not independently verified in this brief) and high dividends — medium-term logic. ⚠️ A directional issue that needs attention: insurers typically benefit from rising long-end yields. If tonight's payrolls are strong → hike expectations rise → long-end yields rise → positive for insurance, negative for gold. In other words, precious metals and insurance are opposite bets in the face of tonight's data, and should not both be allocated as "positives."

④ Software — the most distant mapping; B is the ceiling. IGV +3.41% was one of the strongest U.S. sectors overnight. But two things must be said: first, A-share software has historically correlated weakly with U.S. software; second, the 10Y fell only 2bp and the 2Y only 5bp, a magnitude far too small to drive IGV up 3.41%, so the "falling rates favour long-duration assets" attribution can only be counted as partly holding, and it is more likely stock-specific event-driven. Do not treat it as a theme with industrial logic.

⚠️ The same yardstick has to be applied first to the branch you most want to recommend — this is a double standard this brief caught in itself, now corrected. Above, "rates moved only 2–5bp, not enough to explain IGV +3.41%" was used to downgrade software's duration attribution to "partly holding"; but §0 and item ① of this section attribute GDX +3.95%, SLV +2.51% and GLD +1.85% directly to "cooling hike expectations," on the basis of the same 2–5bp rate move. That falsification logic was not run against precious metals at the time. Now handled to the same standard: the "cooling hike expectations" attribution for precious metals is likewise downgraded to "partly holding." The overnight surge in precious metals is a confirmed price fact, but attributing it entirely to rate expectations conflicts with the criterion this brief used to reject the software duration attribution. The "logical hardness" of the precious-metals branch in the §2 table is therefore cut from "high" to "medium–high": the gold price → miner profit transmission is still hard (that leg has no intermediate link), but the "why is gold going up" leg is soft.

---

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

Component definitions are in §4. The fundamentals / industry position column follows the independent financial verification conclusions; names not covered are tagged "not independently financially verified."

⚠️ Ranking rule and coverage rule (caliber stated explicitly): this table is sorted strictly by total score descending, with ties broken by the higher "expectation gap." The "5 names for today" in §9① = the top 5 of this table, no exceptions and no manual picking. This brief's initial judgment had the problem of "shipping names scoring 66/65 placed ahead of Chifeng Jilong Gold and Shanjin International Gold at 68/67" while the recommendation slots skipped higher-scoring names; that has been fixed. Coverage boundary of this table: the candidates come from the union of five sources — the 9/3 gainers/losers list, the limit-up pool, the Dragon-Tiger List, overnight overseas mappings, and in-window filings — and this is not a whole-market scan; the conclusions of §7 and §9① hold only within this pool.

---

4. Stock Scoring Model (100 pts total)

⚠️ Why this brief switched the primary metric of the fundamentals component from PE to PB percentile — a methodological self-correction. The initial judgment scored on "the lower the PE the better," putting Zijin Mining (PE 13.1) at No. 1. After primary financial verification: all five gold miners' PE three-year percentiles are in the 9%–39% historical-low range, while their PB three-year percentiles are all in the 82%–95% historical-high range. At a cyclical earnings peak, a low PE does not mean cheap, it only means the denominator E is at the top of the cycle — as E falls with the gold price, a low PE automatically repairs itself into a high PE, while PB does not fall for that reason. Continuing to sort by PE would systematically put the names at the very top of the cycle at the very front. The whole table has been recomputed on the new caliber.

Example (Shanjin International Gold 000975 = 75): source 14 + directness 17 + earnings elasticity 12 + fundamentals 14 (PE percentile 9.4% and PB percentile 82.3% both the lowest, OCF/attributable 1.27, net cash) + expectation gap 9 (only +0.60% on 9/3) + persistence 6 + trading characteristics 8 + risk −5 (Q2 −26.6% sequentially) = 75.

Example (Hunan Gold 002155 = 46): source 14 + directness 9 (88.9% of revenue is pass-through at a 0.094% gross margin) + earnings elasticity 8 + fundamentals 5 (OCF −RMB 106 million, inventory +121%, PB percentile 89.1%) + expectation gap 3 + persistence 5 + trading characteristics 7 + risk −5 (negative OCF) = 46.

Example (Baiyin Nonferrous 601212 = 35): source 12 + directness 13 + earnings elasticity 8 + fundamentals 3 (loss-making) + expectation gap 0 (already limit-up) + persistence 5 + trading characteristics 4 + risk −10 (PE<0 deducts 5; turnover 7.95% with hot-money Dragon-Tiger net buying deducts 5) = 35.

---

5. Top 10 Stocks in Detail

⚠️ The order of this section has been rebuilt on the independent financial verification results. The initial ranking rested on "the lower the PE the better"; verification showed that caliber systematically mis-sorts at this point in the cycle (see the method note in §4), and everything has been recomputed.

1山金国际 Shanjin International Gold000975SZ main board ±10% | Score 75 | Priority deep-dive

• Related news: measured 9/4 04:00, U.S. 9/3 close GDX +3.95%, SLV +2.51%, GLD +1.85% (CNBC quote endpoint, timestamp 2026-09-03T16:10 ET). The source is the 9/2 evening ADP print of +38,000 August jobs vs 47,000 expected.
• Positive logic: it hits revenue directly. The mining & dressing segment (RMB 5.131 billion of revenue) contributes 100.6% of gross profit, so a rising gold price lifts the gross margin with no intermediate link.
• Stage of the branch: launch phase. On 9/3 the sector was +2.40% while it was only +0.60% on 1.84% turnover — it barely participated in yesterday's move at all.
• Fundamentals verification (primary financial verification):
  • Its valuation combination is the only one of the five that is "not expensive on either end": PE three-year percentile 9.4% (lowest), PB three-year percentile 82.3% (lowest). Against an industry backdrop of PB percentiles at 82%–95%, that "lowest" position has real meaning.
  • Financial quality: OCF/attributable 1.27, debt ratio 27.3%, cash of RMB 5.504 billion vs short-term borrowings of RMB 20 million (net cash).
  • The only company not stockpiling: inventory fell from RMB 1.160 billion to RMB 966 million. Among peers, Zhongjin Gold's inventory is +RMB 5.42 billion (financed by +RMB 3.32 billion of short-term debt) and Hunan Gold's +121% — stockpiling means adding a directional gold-price position on the balance sheet, stacked in the same direction as the core exposure rather than hedging it. It does not have that problem.
• Differentiation (the point this brief considers most overlooked): silver-related gross profit is 21.5% of the total, while the SGE silver price is +61.5% year over year in Q3 to date versus only +16.1% for gold — silver's year-over-year base pressure lags gold by one step. And its mined silver output at +7% is the only positive metal-output growth among the five (mined gold −12%).
• Risks (which must be stated alongside): Q2 attributable profit was up only +13.4% year over year and −26.6% sequentially, flaming out in step with the whole industry — it is not an exception; 46.8% of revenue is loss-making trading at a −0.61% gross margin (a noise item, not fatal); ⚠️ beware the keyword trap: screening on "silver" will hit its RMB 2.171 billion "silver trading" segment, which loses money; what actually makes money is "silver contained in lead-zinc concentrate" (gross profit RMB 745 million, +229% year over year).
• Final judgment: given the premise of "a gold-price rebound," it is the best combination of fundamental quality and valuation position among names that did not rise yesterday. But the premise itself faces its referee tonight at 20:30.

2山东黄金 Shandong Gold Mining600547SH main board ±10% | Score 68 | Watch closely

• Related news: as above.
• Positive logic: the purest large-cap play on gold-price elasticity (versus Zijin, no copper business dilutes it).
• Stage of the branch: likewise launch phase. ⚠️ This brief's initial judgment wrote that it was "the only decliner in the whole sector," and that is wrong — Zhaojin Gold (000506) was also −0.43% that day, and the Tonghuashun precious-metals sector itself was 12 up, 2 down (that number is written in this brief's own §2 sector table; the first draft had two pieces of its own evidence fighting each other). The correct statement is: it is the deepest decliner among the large- and mid-cap gold miners.
• Fundamentals verification: ⚠️ not within the coverage of this brief's five independently verified names. Market cap RMB 172.9 billion, PE (TTM) 31.6 — on the high side within the sector. Its PB three-year percentile, single-quarter decomposition and cash-flow quality were all left unverified in this brief, and given that all five verified peers without exception show "low PE + high PB + a sequential Q2 decline," it should be assumed to be in the same shape until proven otherwise.
• Technical sentiment: ⚠️ On 9/3 it gapped up +1.48% (38.45), hit a high of 38.61, closed at 37.50 for −1.03%, with a low of 37.34 — a complete gap-up-then-fade. That means even with gold up 1% that day, the stock still faced selling pressure.
• Final judgment: it has the largest expectation gap among large- and mid-cap gold miners (the deepest decline), but the gap-up-then-fade is a genuine negative signal, and the two must be stated side by side. If today's overnight positive can carry it above the 9/3 high of 38.61, the 9/3 selling pressure has been absorbed; another gap-up-then-fade should be read as the sector leader refusing to cooperate. This is the single most worthwhile verification point to watch today.

3中金黄金 Zhongjin Gold600489SH main board ±10% | Score 66 | Watch closely

• Related news: as above; it was already named by JRJ on 9/3 as one of the top gainers.
• Positive logic: a state-owned gold miner. ⚠️ One common misjudgment needs correcting: the market often says "Zhongjin Gold is a smelter, so its gold-price elasticity is small." The segment breakdown shows that claim is right on revenue and wrong on profit — the smelting segment is 91.4% of revenue but at a gross margin of only 4.20%, while the mining segment is 31.6% of revenue yet contributes 95.7% of gross profit (gross margin 62.67%). Its profit is, in substance, a mine, not a smelter.
• Fundamentals verification (primary financial verification):
  • ✅ Gold-price exposure carried per RMB of market cap is 21.8%, the highest of the five — this is its strongest point, the best elasticity/valuation combination; PB percentile 83.6%, second lowest of the five.
  • ⚠️ But its financial quality is the worst of the five: OCF/attributable 0.38, debt ratio 46.6% (both the worst of the five). The cause has been located: inventory rose from RMB 12.346 billion to RMB 17.766 billion (+RMB 5.42 billion), while over the same period short-term borrowings rose from RMB 9.795 billion to RMB 13.111 billion (+RMB 3.32 billion) — stockpiling gold financed by short-term debt.
  • ⚠️ Q2 revenue was up only +0.9% year over year, essentially stalled; ex-non-recurring +46.0% is materially below attributable +61.7%, so about a quarter of the profit growth comes from non-recurring items.
• Technical sentiment: +4.01% on 9/3, opening +2.02%, turnover only 2.07% — it rose without overheating, with healthy volume.
• Final judgment: "best elasticity" and "worst financials" hold simultaneously in the same stock, and you cannot tell only one side. The RMB 5.4 billion of inventory built with short-term debt is an unhedged directional gold-price position — if gold rises it benefits twice, if gold falls it loses twice. Suited to trading a rebound, not to holding.

4紫金矿业 Zijin Mining601899SH main board ±10% | Score 66 | Watch closely

• Related news: as above.
• ⚠️ This brief withdraws an error of its own, and the error has two layers. The initial judgment ranked it No. 1 on the grounds that "PE 13.09 is the lowest in the whole table."
  • Layer one (cycle position): primary financial verification shows that for the five names in this sector, PE three-year percentiles are all in the 9%–39% historical-low range while PB three-year percentiles are all in the 82%–95% historical-high range — the classic shape of cyclical earnings topping out, where the low PE comes from the denominator E sitting at the top of the cycle and does not mean the share price is cheap.
  • Layer two (the word "whole table" was itself wrong): §3 is a cross-branch overall ranking, and within that same table China Life at PE 4.48, New China Life at 4.36, China Pacific at 5.81 and Ping An at 6.59 are all below 13.09. The first draft left the fundamentals cells for the three insurers blank as "—", which happened to make the contradiction invisible on the page. The only correct statement is "the lowest within the precious-metals branch."
  • That reason has been voided and the score cut by 10 points. ⚠️ Note also that the PE caliber itself is unstable: on East Money's forward PE, Zhongjin Gold is 14.29, Shanjin International Gold 15.28 and Xingye Silver & Tin 15.41, whereas on the Tencent TTM basis they are 18.88/19.48/21.99 — a very large difference. This brief uses the Tencent TTM basis for PE throughout.
• What still holds: only +1.09% on 9/3 with turnover of 0.82%, the coldest in the sector, so the expectation gap objectively exists; RMB 885.7 billion market cap, first in domestic reserves and output, leadership beyond dispute.
• ⚠️ What was not obtained: this stock is not within the coverage of this brief's independent financial verification, so its PB three-year percentile, single-quarter decomposition and cash-flow quality were all left unverified here.
• Additional variable: the dual gold-copper business dilutes the gold share, so it is not a pure gold-price proxy; also, the company acquired 25.85% of Chifeng Jilong Gold for RMB 18.258 billion in 2026-03 and consolidated it financially, creating a related-party link with No. 7 in this ranking.
• Final judgment: the expectation gap is real; "cheap" is unverified. Kept in the watch slot, but the reason changes from "cheapest" to "least moved + leadership position."

5招商轮船 China Merchants Energy Shipping601872SH main board ±10% | Score 66 | Watch closely

• Related news: ⚠️ None. +8.37% on 9/3 (17.81→19.30) on RMB 2.79 billion of turnover, but three rounds of search obtained no confirmable public catalyst for that day.
• Positive logic: crude tankers are a high-elasticity freight-rate → profit industry. If this is VLCC-rate driven, the path to earnings delivery is clear.
• Stage of the branch: fermentation phase, but on an unclear basis.
• Fundamentals verification: PE (TTM) 14.4, market cap RMB 155.8 billion. A low-valuation large cap.
• Technical sentiment: up 8.37% but turnover only 1.82% — meaning it was driven by existing holders with very light selling pressure, not hot-money wash trading. That combination (big gain + low turnover) is usually healthier than "big gain + high turnover."
• Final judgment: the price action and the valuation are both good; the only problem is that I could not prove why it rose. Per the iron rules, no catalyst is fabricated. The first order of business today is to establish what drove 9/3; until then, its score carries a 7-point deduction for "catalyst unconfirmed."

6中远海能 COSCO Shipping Energy600026SH main board ±10% | Score 65 | Watch closely

• Same source as China Merchants Energy Shipping. +8.77% on 9/3 (18.69→20.33), turnover value RMB 2.33 billion, turnover rate 3.02%, PE 16.6, market cap RMB 111.2 billion.
• Likewise no confirmable catalyst obtained. The judgment matches the previous name.
• One useful distinction: COSCO Shipping Energy is the domestic crude-tanker leader, purer in tankers than China Merchants Energy Shipping (which also has dry bulk, ro-ro and so on). If tanker rates are confirmed as the driver, COSCO Shipping Energy has more elasticity; if it is dry bulk, look at Haitong Development instead. That fork can be tested with the relative strength of the three intraday today.

7赤峰黄金 Chifeng Jilong Gold600988SH main board ±10% | Score 64 | Watch closely

• Same overnight precious-metals positive. +1.81% on 9/3, opening +1.88% (it spiked and faded but still closed up), turnover 3.44%, market cap RMB 88.6 billion.
• Fundamentals verification (primary financial verification):
  • ✅ The cleanest business model of the five: own-mined gold is 88.0% of revenue and 94.3% of gross profit, while purchased gold is only 1.2% of revenue and 0.04% of gross profit — almost no low-margin pass-through business to dilute it. OCF/attributable 1.09, net cash, debt ratio 29.6%.
  • ⚠️ But it is also the most expensive of the five: PB 6.18, three-year percentile 94.5%.
  • ⚠️ The rare-earth segment's gross margin of −50.2%, also negative a year earlier (−16.3%), is a continuous drain; overseas revenue is 75.7% (Sepon in Laos + Wassa in Ghana), and its price benchmark is the LBMA dollar gold price rather than Shanghai gold, so FX is an extra variable.
  • Output is −9% to −11% year over year (Q1 already disclosed mined gold of 2.98 tonnes, −10.71%).
• A related-party relationship that must be disclosed: Zijin Mining acquired 25.85% of the company for RMB 18.258 billion in 2026-03 and consolidated it financially; the current share price of RMB 46.60 is 12.7% above that transaction price of RMB 41.36. Post-consolidation, the capex pace and dividend policy carry uncertainty.
• Final judgment: the cleanest business, the priciest valuation. On business quality alone it should rank higher; it is the 94.5% PB percentile that pushed it down to No. 7.

8中国太保 China Pacific Insurance601601SH main board ±10% | Score 62 | Watch only

• +3.69% on 9/3, with the sector +2.56%, first among 90 sectors and 5 up, 0 down (perfect breadth).
• But the same-day catalyst has now gone unobtained for a second consecutive day. Searches return only medium-term logic such as "the five majors' interim net profit of RMB 317.4 billion and RMB 39.0 billion of dividends."
• ⚠️ The most important point: insurance and gold are opposite bets in the face of tonight's payrolls. Strong payrolls → hike expectations rise → long-end yields rise → positive for insurers, negative for gold; weak payrolls, the reverse. Holding both heavily is self-hedging; pick one.
• The label is "watch only," because it is already up 3.69% on 9/3 and lacks a same-day catalyst.

9长电科技 JCET Group600584SH main board ±10% | Score 62 | Watch closely

• Related news (the original filings have been retrieved and verified, art_code prefix AN20260903..., 12 documents in total): on 9/3 it held the fifth extraordinary meeting of its ninth board and proposed a private placement of A-shares raising no more than RMB 6.5 billion. Use of proceeds: capacity expansion of a high-performance-computing high-end advanced packaging platform RMB 1.5 billion, capacity upgrade for high-end power-module advanced packaging and testing RMB 1.0 billion, upgrade and expansion of the wafer-level packaging advanced-process platform RMB 1.1 billion, system-level packaging and testing for high-density, high-capacity memory chips RMB 1.0 billion, and working capital and repayment of bank loans RMB 1.9 billion.
• Three details that change the nature of it (missed in the first draft):
  1. Controlling shareholder Panshi Runqi (ultimately controlled by China Resources) is subscribing for 22.53%, about RMB 1.46 billion on a RMB 6.5 billion raise, locked for 18 months (other participants lock for 6 months). And the issue is book-built with the pricing base date being the first day of the issue period; Panshi Runqi does not participate in the bidding but accepts the bid result — no haggling, no floor, the same price as outside institutions. That is a high-signal-cost action.
  2. ⚠️ Actual dilution is roughly 5.1%–6.4%, not 30%. The plan's "no more than 536.8 million shares = 30% of pre-issue share capital" is a regulatory formal ceiling; the real constraint is RMB 6.5 billion ÷ the issue price, which at the 9/3 close of RMB 71.27 implies about 91 million new shares and 5.10% dilution, or 6.37% at the 80% floor price of RMB 57. If you see a claim of "up to 30% dilution," that is mistaking the regulatory ceiling for the issuance plan.
  3. Working capital and loan repayment of RMB 1.9 billion / RMB 6.5 billion = 29.2%, just inside the 30% regulatory red line.
• Capacity utilisation (primary, quotable): the interim report says "capacity utilisation running at a high level"; the 8/31 results briefing said "traditional packaging is close to full load and is expected to stay there in the second half" and "back-end high-end packaging and testing will become the bottleneck link in the industry's volume ramp"; the placement plan says "wafer-level packaging lines are close to full capacity" and "advanced packaging capacity is approaching saturation." The capacity rationale for the expansion holds.
• ⚠️ This brief gives no figure for "advanced packaging as a share of revenue." Searching the interim report and the plan turned up 59 occurrences of "advanced packaging," none with a disclosed revenue-share number. All that can be quoted is the capex allocation (in H1, high-end advanced packaging was more than half and mainstream packaging less than a tenth) and subsidiary data: JCET Advanced generated RMB 1.311 billion of revenue (6.7% of the group) and contributed RMB 394.5 million of net profit (46.7% of group attributable profit).
• The real financial motive for the placement: H1 operating cash flow of RMB 2.964 billion − capex of RMB 4.7 billion ≈ free cash flow of −RMB 1.7 billion (the 2026 full-year capex budget is RMB 10.0 billion, of which RMB 4.7 billion was spent in H1, up nearly +80% year over year); the debt ratio rose from 43.64% to 46.22%. It is not short of projects; its operating cash flow can no longer cover the pace of expansion.
• Counter-factors: ① valuation at PE (TTM) 65.8 and PB 4.41, both at the 93.9% percentile of the past 5 years; ② the use of proceeds includes memory SiP, while U.S. memory was being sold off overnight; ③ it still requires state-asset approval → shareholder meeting → SSE → CSRC, four gates, with the plan itself stating "considerable uncertainty exists," and zero cash impact on fundamentals before approval; ④ the plan landed after 4 consecutive days of net main-force outflows and a share-price retracement from 76.54 to 71.27 (−6.9%).
• Final judgment: China Resources subscribing at the same book-built price with an 18-month lock-up is the hardest thing in this filing, and the capacity rationale is solid; but the valuation percentile, the approval timeline and the external headwind in memory all exist at the same time. Rated "watch closely," no higher label.

10四川黄金 Sichuan Gold001337SZ main board ±10% | Score 60 | Watch only

• +3.17% on 9/3, opening +0.34%, turnover 5.36%, market cap RMB 22.9 billion.
• Fundamentals verification (primary financial verification) — it is the "best-looking data yet ranked tenth" name in this list, and the reason is worth spelling out:
  • ✅ Best operating quality of the five: gross margin 71.0%, half-year ROE 21.86%, OCF/attributable 1.31, with no goodwill, no trading business and no purchased material, and gold concentrate at 100% of both revenue and gross profit — the only company without a single line of trading.
  • ✅ The only company with real output growth (implied +30%), and also the only one still +48% year over year in Q2 — a genuine exception against a backdrop of all five flaming out in Q2.
  • ⚠️ But the valuation has front-run all of it: PB 11.14, three-year percentile 91.2%, PE 33.3. Every RMB of market cap buys only 7.2% of gold-price exposure, the lowest of the five and just one third of Zhongjin Gold's (21.8%).
• This stock is the best example of "business purity ≠ investment elasticity": it is the one that looks most like a "pure gold miner," and also the one that benefits least on a market-cap-weighted basis when gold rebounds. Both statements hold at once, and the second is said far less often.
• ⚠️ Pending verification: the debt ratio rose from 26.2% to 38.8% within a year and total assets grew about 58%, while long-term borrowings did not change materially — where the increment came from is unverified. This determines whether its implied +30% output growth is a sustainable capacity ramp or a one-off consolidation, and that is the only fundamental support for its 33.3x PE / 11.14x PB.

---

6. Pass List

A note on classifying Pass reasons (carrying over yesterday's recap's lesson): yesterday's recap found that the "the company's own statements point the opposite way to the buy case" type of directional falsification went 3 for 3, while the "I cannot find a catalyst" type of insufficient evidence went only 1 for 6. This brief therefore deliberately separates the two: in the table above, 601212 (loss-making), 002909/002536 (the companies issued their own risk warnings) and 603986/300308 (overnight overseas direction is opposite) are directional-falsification type and carry high confidence; whereas 600172 (stale narrative) is insufficient-evidence type, and it could perfectly well keep surging today — "the evidence does not hold" is not the same as "it will fall tomorrow." Please do not treat this table as a short list.

---

7. Within-Branch Ranking

Branch 1: Precious metals (gold / silver)

The fundamental-support column has been switched to a "PB three-year percentile + cash flow" caliber, no longer PE (reasons in §4).

Hardest single name: 山金国际 Shanjin International Gold (000975) — the only one with both the lowest PE percentile and the lowest PB percentile, the cleanest cash flow, and virtually no move yesterday. Highest trading recognisability: 山东黄金 Shandong Gold Mining (600547) — the market's accepted "gold-price proxy," but it has to clear the 9/3 gap-up-then-fade. Greatest elasticity but worst quality: 中金黄金 Zhongjin Gold (600489) — exposure of 21.8% per RMB of market cap, first in the field, while OCF 0.38 and a 46.6% debt ratio are the worst in the field; both sentences have to be said together. Who is following: Western Region Gold and Shengda Resources are in catch-up slots. ⚠️ A classification error that must be called out: the market treats Hunan Gold, Hunan Silver and Xingye Silver & Tin all as "silver/gold elasticity plays," but the reasons the three are Passed are entirely different — Hunan Gold is a profit-structure mismatch (88.9% of revenue is zero-margin pass-through, and antimony is still a drag), Hunan Silver is a reversed business model (a smelter that is hurt when silver falls), and Xingye Silver & Tin is an event shock (the flagship mine's mining area is shut). Only the third is reversible; the first two are structural. Who is Passed and most dangerous: Baiyin Nonferrous — a loss-making company (PE −118.5) hitting limit-up with RMB 243 million of hot-money net buying on the Dragon-Tiger List, with earnings and share price completely decoupled.

Branch 2: Crude tankers / dry bulk shipping

Hardest single name: 中远海能 COSCO Shipping Energy (600026) — the purest in tankers, and the most elastic if freight rates are indeed the driver. Note: the whole branch is marked down a notch, because the catalyst is unconfirmed.

Branch 3: Insurance

The whole branch is uniformly labelled "watch only", because: ① the same-day catalyst has gone unobtained for two consecutive days; ② it already rose as a group on 9/3; ③ it points the opposite way to precious metals in the face of tonight's payrolls, so it should not be held alongside the first branch.

---

8. Today's Open Verification Signals

Auction signals

• Whether the large-cap precious-metals names gap up: focus on the three that did not rise on 9/3 — Zijin Mining (601899), Shandong Gold Mining (600547) and Shanjin International Gold (000975). If they gap up 1–2% and do not fade → the overnight GDX +3.95% has transmitted effectively and the branch holds.
• ⚠️ Whether Shandong Gold Mining can clear 38.61 (the 9/3 high) — this is the single most critical verification point set by this brief. Clearing it = the 9/3 selling pressure absorbed; another gap-up-then-fade = the leader refuses to cooperate and the whole branch is downgraded.
• ⚠️ Counter-signal: if the small caps (Baiyin Nonferrous / Hunan Silver) gap up sharply while the large caps open flat → it is still one-sided hot-money behaviour with no institutions entering, and it should be downgraded outright to a one-day wonder.

Sector signals

• Whether the precious-metals sector produces three or more fast limit-ups, at least one of them above a RMB 50 billion market cap. If the limit-ups are all concentrated in names below RMB 30 billion, the branch quality is unsatisfactory.
• Core names on volume: Zijin Mining's turnover was only 0.82% on 9/3; if it expands above 1.5% today and closes up, that is a sign of institutional entry.
• Whether the three tanker names (China Merchants Energy Shipping / COSCO Shipping Energy / COSCO Shipping Specialized Carriers) strengthen in step, and the relative strength of Haitong Development (dry bulk) against them — that can be used to back out whether the 9/3 driver was tankers or dry bulk.

Stock signals

• Follow-through in the top 5: among Zijin, Shandong Gold Mining, Zhongjin Gold, Chifeng Jilong Gold and Shanjin International Gold, how many close with gains above the sector average.
• Large-order net inflows: the precious-metals sector had net inflows of RMB 1.465 billion on 9/3; if today it can hold above RMB 1.5 billion and stay concentrated in the large caps, the quality is better than yesterday's.
• JCET Group's (600584) reaction to the placement — placement dilution vs the controlling shareholder's 22.53% subscription; today's open is a direct vote on how the market weighs the two.

Risk signals

• ⚠️ Gap up then dive: payrolls come tonight, so the incentive to take profits intraday on a Friday is strong. If precious metals gap up more than 2% and fall back to flat within half an hour, treat it as a failed front-run.
• ⚠️ A lone-runner tape: if only Baiyin Nonferrous / Hunan Silver hit limit-up while the large caps do not follow → an ebbing-tide signature.
• ⚠️ Yesterday's hot themes ebbing: Guangzhou Jointas Chemical (002909, 4 consecutive limit-ups) and Henan Feilong Auto Components (002536) both issued risk-warning filings on the evening of 9/3 (⚠️ the initial judgment here mistakenly wrote "two of Jointas, Jinfu Technology and Jindi Co."; the ones that issued warnings were in fact Jointas and Feilong, and there is no evidence for Jindi Co., which has been deleted). Stacked with AVGO's −2.74% on its first day post-earnings overnight, the probability of the high-flyer cohort ebbing is not low, which could drain market liquidity and weigh on the index.
• ⚠️ A weak index: 9/3 was already a fade-into-the-close day (the Shanghai Composite was +0.43% at midday and closed +0.02%), with a 42.9% failed-limit-up rate and 16 limit-downs. The money-making effect was already contracting, so do not assume a broad rally today.

---

9. Final Recommendations

① The 5 stocks most worth watching today

A note on board distribution (the iron rules require no fencing, so this is disclosed proactively): four of the five are SH main board and one is SZ main board, with no ChiNext / STAR Market / BSE names. This is not fencing, it is the natural outcome of the screen — this round's catalysts are the gold price and freight rates, and the ChiNext name on those chains (Xiaocheng Technology 300139) was verified to be a scale mismatch (PB 12.3, half-year ROE 4.6%), while the STAR Market and BSE have no names matching the core business. If a ChiNext gold miner strengthens on volume today, that should be treated as an omission of this brief, not as "that board should not be bought."

② The 3 strongest branches today

⚠️ Self-questioning on the ratings of branches 2 and 3 (this brief's own check): both are tagged "catalyst unconfirmed," yet both rank ahead of the software branch, which has a measured overnight catalyst (IGV +3.41%) — which shows that the "catalyst unconfirmed" deduction did not genuinely constrain the conclusion in this brief's initial judgment. Both have now been cut from B+ to B. Readers can take this as the caliber of the table: the ranking mainly reflects the strength of price action and valuation, and is not equivalent to the degree of evidence confirmation. If tanker freight-rate data still cannot be obtained before today's open, branch 2 should be handled straightforwardly as "pending verification" and should not be traded on.

③ Directions not worth chasing today

1. ⚠️ Memory / the semiconductor memory chain (GigaDevice 603986, Demingli 001309, Longsys 301308) — ⚠️ this brief's initial judgment called this "the best-evidenced avoidance of all"; verification overturned that and it has been downgraded. The reality: the overnight decliners WDC −1.63% and STX −1.23% are both hard-disk names, while MU +0.22% and SNDK (SanDisk, NAND) +0.10%, which are directly comparable to these three, were up. Using HDD to falsify DRAM/NAND is a sampling error. The reason for avoidance now rests solely on A-shares themselves: on 9/3 GigaDevice −1.46%, Demingli −1.91%, Longsys −0.74% and Montage Technology −1.60% all closed down. This is "A-shares weakening on their own," not "weakness resonating across both markets" — materially weaker than this brief's initial wording.
2. ⚠️ Compute / optical modules (Innolight Technology 300308, Eoptolink 300502) — ⚠️ two errors in this brief's initial judgment, both corrected. ① AVGO reported after the close on 2026-09-02, so 9/3 was the first trading day after earnings, not the "second trading day." The −2.74% is the earnings reaction itself — which is actually a stronger signal ("above-consensus guidance still sold"), but the first draft's reasoning was wrong. ② U.S. AI servers and networking rose across the board over the same period (DELL +4.91%, HPE +5.04%, ANET +2.87%, SMCI +2.35%, NVDA +1.80%), so "the compute chain is weakening" does not hold. One supplementary key data point: optical-communications equipment maker CIEN was −10.36% on 9/3 — which supports avoiding optical modules better than AVGO does, while the control group of FN −0.09%, COHR −1.57%, LITE −2.67% and CRDO −0.64% shows optical communications is diverging rather than weakening chain-wide. The A-share chain closed down in step on 9/3 (Innolight Technology −1.14%, Eoptolink −0.50%).
3. ⛔ The elevated liquid-cooling / thermal-management cohort (Guangzhou Jointas Chemical 002909, Henan Feilong Auto Components 002536, Siquan New Material 301489, Jinfu Technology 003018) — Jointas and Feilong issued risk-warning filings on the evening of 9/3; on valuation, Jointas is at PE −94 (loss-making), Siquan at 258x and Feilong at 199x; stacked with the overnight weakness in the compute chain.
4. ⛔ Lab-grown diamonds (Henan Huanghe Whirlwind 600172) — no new catalyst inside the window, the same narrative having been used four times already in 2026; turnover 30.15%.
5. ⛔ The small-cap precious-metals names that have already gone wild (Baiyin Nonferrous 601212, Hunan Silver 002716) — same source as branch 1 but they must be cut apart: they rose the most yesterday, have the highest turnover and the weakest fundamentals. Baiyin Nonferrous is loss-making at PE −118.5; Hunan Silver was verified to be a silver smelter rather than a mining company (smelting is 97% of revenue at a gross margin of only 4.74%, so a rising silver price lifts revenue but not profit, while H1 already saw RMB 93.02 million of inventory write-downs from high-cost stock). Being bullish on precious metals is not the same as being bullish on these two — especially Hunan Silver, whose price exposure runs the wrong way.
6. ⛔ 兴业银锡 Xingye Silver & Tin (000426) — the only name in this brief downgraded because of an in-window filing. Its resource endowment is the hardest in the field (silver resources first in Asia and seventh globally), but the 9/3 evening filing shows that Yinman Mining, which contributes 46.7% of group revenue and 55.8% of ex-non-recurring profit, has had its mining area shut for more than 5 weeks since the 7/26 accident; that day only the concentrator restarted and the mining area is still under rectification. Stacked with a Q3 average silver price 22.7% below H1, Q3 faces a double hit to volume and price, and the interim report shows none of it. This is the classic "good company, wrong timing."
7. ⚠️ 晓程科技 Xiaocheng Technology (300139) — one correction to an initial judgment. The first draft wrote that "the reality of the gold business is doubtful, likely a concept stock"; verification shows that is wrong: Ghana gold-mine revenue has been 82%–94% of the total for two consecutive periods, gross margin 63%, overseas revenue 98.6% — the business is real. Its problem is a scale mismatch (half-year gold revenue of RMB 295 million vs an RMB 14.5 bn market cap, PB 12.3, half-year ROE 4.64%) and its money structure (net extra-large orders of −RMB 0.1 million on 9/3, zero institutional participation), plus a Q2 swing to an ex-non-recurring loss despite high gold prices. The conclusion is still Pass, but for a different reason — and that difference matters for judging whether it could see a fundamental inflection later.
8. ⚠️ Do not go heavy on "precious metals + insurance" at the same time — the two are opposite bets in the face of tonight's payrolls (strong payrolls → rising yields → good for insurers, bad for gold), so holding both is self-hedging.

④ The final one-sentence judgment

Today is a day for "front-running before the referee blows the whistle": the overnight gain in precious metals is real, but the data it is betting on will not be revealed until 20:30 tonight, after the A-share close — so any position should be sized as "pre-event" rather than "mid-trend." And within the branch, the real opportunity is not in the small caps that hit limit-up yesterday, but in the large caps like Shanjin International Gold, Shandong Gold Mining and Zijin Mining, where "the overnight positive belongs to them yet they did not rise yesterday." At the same time, remember two widely misstated premises: first, gold has not made a new high — spot is $4,475.70, still about −20% below the 2026-01-29 peak, which makes this a rebound after a deep fall rather than a breakout, and it was already giving some of it back before this went out; second, the gold miners' "high interim growth" is a Q1 legacy — all five saw Q2 profit fall sequentially, and the year-over-year gold-price dividend will go to zero in Q4. This is a rebound trade you can participate in, not an industry theme you can hold.

---

VII. risk-auditor QA results — 14 items found, of which 1 engineering defect and 4 substantive factual errors, all corrected

⚠️⚠️ [An engineering defect that had to be fixed on the spot, or it would contaminate every future edition] The Treasury CSV columns were indexed positionally and were offset by 2 columns overall. The Treasury's daily_treasury_yield_curve now contains a non-standard column 1.5 Month, so f[5]/f[10] were picking up 4 Mo/5 Yr rather than 2 Yr/10 Yr. The first draft's 9/3 figures of "2Y 3.99 / 10Y 4.52" are exactly that day's 4 Mo/5 Yr, and the same holds for 9/2 — all four numbers across two days line up, which rules out picking the wrong date and confirms a column-selection bug. I reproduced and confirmed it with csv.DictReader keyed by column name: 9/3 is in fact 2Y 4.34 / 10Y 4.77 / 30Y 5.25. The fix: columns must be selected by name with a KeyError assertion; positional indexing is banned. Ironically, the wrong numbers actually understated this brief's own argument — the true 2Y of 4.34% sits roughly 71bp above the funds-rate midpoint, implying far stronger hike pricing than 3.99% would.

Four substantive factual errors (all rewritten in the main text with an audit trail):

1. "Memory was sold off" is a sampling error. Only the two HDD names (WDC/STX) and AVGO on its earnings day fell, while MU +0.22% and SNDK +0.10%, which are directly comparable to the A-share memory chain, were up, as were DELL/HPE/ANET/SMCI/NVDA. Yet the first draft self-graded this as "the best-evidenced avoidance of all." Using HDD to falsify DRAM/NAND is taking the wrong control group.
2. The AVGO earnings day was stated backwards. Broadcom reported FY26Q3 after the close on 9/2, so 9/3 is the first trading day after earnings, whereas the first draft wrote "the second trading day, with the positive fully priced." The direction of the conclusion is right but the reasoning was wrong, and the truth ("above-consensus guidance sold on day one") is in fact a stronger signal. Also missed: CIEN −10.36% on 9/3, which supports avoiding optical modules better than AVGO does.
3. "Shandong Gold Mining was the only decliner in the whole sector" is a universal-quantifier error. Zhaojin Gold (000506) was also −0.43% that day, and the Tonghuashun precious-metals sector itself was 12 up and 2 down — that number is written in this brief's own §2 sector table. This is the classic "two pieces of evidence fighting each other inside the same piece," rooted in my drawing a whole-sample conclusion like "the only one" from a hand-built list of 8 names.
4. Mixed sector calibers. The section opened with Tonghuashun precious metals +2.40% but then used Baiyin Nonferrous (which actually belongs to the industrial-metals sector, where it is the leading gainer) and Zijin Mining (not a precious-metals constituent) to argue about "sector internal structure" — and that is the very foundation of the whole brief's final stock-selection conclusion. Changed to state explicitly that this is a "self-built universe."

Other items corrected: SCFI is actually the 8/21 print (the first draft labelled it 9/3); ADP was released before the U.S. open on 9/2 and priced by that session (the first draft treated it as the 9/3 overnight catalyst); the gold level 4406 → 4475.70 (and it contradicted this brief's own GLD −19.5%); "Zijin has the lowest PE in the whole table" does not hold in a cross-branch table (four insurers at PE 4.36–6.59 are lower); the double standard in rate attribution (the "2–5bp is not enough to explain it" yardstick was applied to software but not to precious metals); Jindi Co. appeared out of nowhere; Henan Huanghe Whirlwind at PE −26.30 is equally loss-making yet was filed under the low-confidence category; the insurance "5 up, 0 down" is a small-sample illusion from a sector with only 5 constituents; the §3 ranking rule and the recommendation-slot coverage rule were not disclosed.

Parts the QA pass independently confirmed as accurate and not requiring change: all A-share 9/3 stock prices and opening gains (back-computed item by item and matching), the sector data, limit-ups 44 / failed limit-ups 33 / limit-downs 16 and the 42.9% failed-limit-up rate, all U.S. 9/3 closing data (every symbol's last_time is 9/3 ET, with no 9/2 contamination), all ADP / payrolls figures and release timings, and the key terms of JCET's placement. As well as this brief's most central assertion: "gold has not made a new high" holds, with the peak date of 2026-01-29 corroborated by three independent fields.

⚠️ A trap the QA pass specifically flagged for the next edition: searching "August 2026 nonfarm payrolls" returns an article claiming "August payrolls +22,000, unemployment 4.3%," which is in fact a 2025-09-06 article about 2025 data. When writing tonight's recap it will masquerade as "the result already published."

Four items still open, to be filled in intraday or after the close (in priority order):

1. ⚠️ Daily TCE values and historical percentiles for BDTI / VLCC TD3C — branch 2 and two recommendation slots currently run naked on "price action + low PE." Having just used "low PE + high PB = cycle top" to reject the precious-metals valuation argument, not applying the same test to tankers' low PE is the same double standard. If it cannot be filled in, the branch should be downgraded to "pending verification."
2. U.S. data or speeches on 9/3 (jobless claims, ISM services, Challenger, Warsh) — the real trigger for that overnight precious-metals jump.
3. Xingye Silver & Tin's filing on the restart of Yinman Mining's "mining area" — the only piece of information that is invisible in the interim report yet sufficient to change the direction of Q3.
4. Same-day 9/3 catalysts for shipping and insurance — insurance has now gone unobtained for a second consecutive trading day.

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