A-Share Pre-Market Brief | 2026-09-18 Friday

News window: 2026-09-17 15:00 (prior session close) → 2026-09-18 07:00 (Beijing time). Market snapshot timestamp: single-stock prices/PB/turnover/fund flows are uniformly as of 2026-09-17 close (Eastmoney push2delay, timestamps checked name by name); interval returns are computed in-house from Tencent raw daily bars. Price limits: Shanghai/Shenzhen main board ±10% ｜ ChiNext (30x)/STAR Market (68x) ±20% ｜ BSE (8x/4x) ±30%. Every stock in this report is tagged with its board; limit-up/failed-limit-up is judged against its own limit band. Yesterday's limit-ups/limit-downs/failed limit-ups: official exchange-standard pools, validated with tc == len(pool) (ZT 47=47, DT 1=1, ZB 20=20). ⚠️ Valuation basis (unified in this report): every PE in this report is PE(TTM), rolled quarter by quarter as "prior full year − prior-year same period + current period" and reverse-checked. The market API's f162 is forward PE (H1 net profit ×2, annualized), which systematically understates at cycle peaks and has been dropped entirely from this report — the gap can be large; e.g. 先导基电 (Xiandao Jidian) forward PE is 82.91 while true PE(TTM) is 815.50 (10x off). Names whose rolling calculation was not completed are marked "not obtained". ⚠️ Read timestamps must not be mixed: A-shares are 9/17 15:00 Beijing; SHFE gold/silver/copper night session is 9/18 02:30; GDX/SLV/COPX are 9/17 16:00 ET; spot gold/silver are 9/17 19:59–22:07 Beijing. Wherever this report compares across markets, the timestamp is repeated in parentheses and no cross-timestamp subtraction is performed. Today's calendar event: 10:00 State Council Information Office press conference by MOHURD (see §1 item 12).

---

0. One-sentence summary

The hardest fact today is not a headline, it is a spread that can be computed within one and the same timestamp and one and the same currency: on 9/17 the A-share market killed precious-metal equities on a "Fed rate hike" narrative, −5% to −7%, while RMB metal futures closing the very same day barely fell (SHFE silver −1.51%, SHFE gold −0.59%, SHFE copper +0.71%) — the same-period equity-vs-commodity divergence is −4.56pp for gold, −4.49pp for silver, −3.08pp for copper.

But this branch carries a discount that must be stated up front: international gold and silver did rally hard after the close (spot gold +2.4%, spot silver +4.91%), yet A-share miners price their revenue off RMB metal prices, not USD prices. In RMB terms the overnight move was only SHFE gold +0.78%, SHFE silver +2.33%, SHFE copper +1.42% (9/18 02:30 night-session close). This 1.6pp gap has not been explained in this report (FX, internal-external spread, and contract-month differences are all possible), so the elasticity base is computed on the RMB figures throughout, not on +2.4%.

After completing the fundamentals drill-down, this report changed two of the draft's recommended slots: silver rallied hardest, yet all three major A-share silver names carry hard defects — 兴业银锡 (Xingye Silver & Tin, 000426) has the truest elasticity (silver is 61.97% of gross profit, 100% self-mined, no silver hedging) but its core Yinman mine pit has been suspended since 7/26 with no resumption announced; 盛达资源 (Shengda Resources, 000603) had both the company and its actual controller placed under CSRC investigation on 6/16 (suspected disclosure violations); 湖南白银 (Hunan Silver, 002716) gets only 2.94% of revenue from self-mined ore, with a silver gross margin of 5.05%, and the company states in its own interim report that rising silver prices cause cost inversion. → The silver elasticity is real; A-share has no clean vehicle for it. Today we express it through gold and copper, not silver.

The second item is a hard single-stock order that landed only after the close: 光启技术 (Kuang-Chi Technologies, 002625) announced at 19:06 a RMB 907 million metamaterial mass-production contract, equal to 80.4% of its 2026 H1 revenue (each prior contract was cross-checked — RMB 264 million on 1/8, RMB 612 million on 6/22 — confirming this is incremental rather than a cumulative-basis rehash); the stock closed −0.67% that day and sits −27.46% below its 60-day high.

Pre-market state call: strong offshore, weak onshore; the index will most likely gap up, but this is a continuation of the ebb, not a reversal. Yesterday's failed-limit-up rate was 29.9%, the median seal-order/turnover ratio for limit-up names was just 0.009, and yesterday's limit-up promotion rate was 10.1%. What a gap-up should be used for is selling yesterday's extended names and buying yesterday's unfairly-sold laggards — both will happen today at the same time.

---

1. News overview

1b. Out-of-window hard facts that take effect today

These items fall outside the news window but will affect pricing today, and no media outlet is mentioning them this morning. All come from company filing originals or exchange data.

---

2. Strongest positive branches, descending

Branch 1 expansion (one): the computable part of the divergence, and what it does not include

The "same-period divergence" column is the only strictly comparable quantity in this report: the A-share sector and the RMB futures are both 9/17 day session, same currency, same closing moment. Its meaning is: what A-shares sold off yesterday was not metal, it was the people holding metal equities — on the same day the metals themselves did not fall, and SHFE copper actually rose 0.71%.

What this column does not include (three limits that must be declared):

1. It does not include the overnight move. The overnight move is the two right-hand columns; it is additional information and cannot be added to the divergence.
2. It cannot be subtracted against GDX. GDX's +3.36% was read at 9/17 16:00 ET (= 9/18 04:00 Beijing), 13 hours after the A-share close, and is a USD-denominated basket of overseas miners. The "8.5pp gap" between it and the A-share gold sector's −5.15% blends time difference, currency difference and constituent difference — three layers of non-comparability. This report performs no subtraction on it and cites it only as directional corroboration.
3. The elasticity base is computed in RMB. There is a 1.6pp gap between spot gold +2.4% and SHFE gold +0.78%, whose cause this report has not established (FX, internal-external spread and contract-month differences are all possible). A-share miners price revenue off the RMB gold price, so the whole report computes elasticity on +0.78% / +2.33% / +1.42% and does not use the USD figures.

⚠️ This limit weakens the branch and must be acknowledged: on a USD gold price of +2.4% the overnight gap is substantial; on an RMB gold price of +0.78% the incremental overnight piece is in fact thin. The main force of this branch comes from that −4.56pp same-period divergence on 9/17, not from the overnight move.

Branch 1 expansion (two): the silver elasticity is real, but A-share has no clean vehicle

The trap in the phrase "silver is 70% of revenue": 湖南白银's silver revenue is RMB 6.126 billion, 70.26% of revenue, which looks like a pure silver play. But that is pass-through smelting and processing volume, with a gross margin of only 5.05%; genuinely self-mined ore is only 2.94% of revenue. Revenue share lies; gross-profit share does not.

Branch 1 expansion (three): why this is "repair" and not "reversal"

• Precious-metal equities are not at highs: 盛达 is −17.64% below its 60-day high, 兴业 −19.39%, 山东黄金 −13.30%, 湖南黄金 −12.56%, and all are down 6.8%~10.7% over the past 5 sessions.
• But on a 60-day view the gains are still considerable (盛达 +42.06%, 山东黄金 +38.94%, 赤峰黄金 +63.19%) — a round of money has already been made this year.
• One valuation warning, but note its scope: the three silver names (兴业/盛达/湖南白银) have PE(TTM) 5-year percentiles of 19.8%~40.9% but PB 5-year percentiles of 82.1%~90.5%, a gap of 42–62 percentage points. Low PE percentile + high PB percentile is the cyclical-top pattern, not cheapness — the denominator is cycle-peak earnings.

  ⚠️ The sample for this criterion is only three silver names. This report did not obtain 5-year PE/PB percentiles for 山东黄金, 山金国际 or 紫金矿业, so this conclusion is not extrapolated onto the recommended names. A reader wanting to apply it to gold equities must fill in the percentile data themselves.

• The biggest counter-evidence: the October hike probability is 55.4%, and CSC Financial explicitly warns that "the hiking cycle has begun, the tailwind period for US equities/commodities may be over". If this is the start of a hiking cycle, yesterday's de-rating was correct and today is only a bounce.

→ Give it size, not duration. A 1–3 day repair trade.

Branch 1 expansion (four): evidence grading, and why this report treats NPO and robots differently

A sharp question came up during the audit: NPO was denied any single-stock recommendation for "no filing-level evidence", yet the Tesla supply relationships of 三花/拓普 are equally unverified by this report — on what grounds do they occupy recommended slots? The challenge stands, and on that basis this report downgraded the robot branch from A to B+ and moved the two names into an explicitly labelled second tier (see §9). The difference between the two cases is set out here:

The difference is real, but not enough to support the "watch closely" label — so both names were downgraded, with the rationale written into the §9 table rather than a footnote.

Branch 4 expansion: the market will most likely read Huawei's NPO backwards

One line in the official description is widely overlooked: 5,500 Hi-ONE units (NPO optical engines) can replace roughly 48,000 800G optical modules, cutting power draw by over 550 kW.

• That is an 8.7 : 1 replacement ratio. For pluggable optical-module consumption inside Huawei's Ascend ecosystem, it is an unambiguous shrinkage signal.
• But it cannot be generalised into "optical modules are finished": the main customers of Innolight / Eoptolink are North American cloud providers (the Nvidia ecosystem), not Huawei Ascend. These are two different transmission channels.
• Be clear about the clock: the air-cooled / liquid-cooled versions of the Ascend 960 SuperPod launch in 2027 Q3/Q4 respectively; the substitution effect lands in 2027.
• The theme is not new: the OPEN NPO MSA was already launched on 2026-07-09, more than two months ago, over which period 60-day declines in the optical-module sector were Innolight −32.30%, Eoptolink −30.49%, Accelink −26.86%, HGTECH −39.06%. The only new information last night was that replacement ratio and the pulled-forward Ascend roadmap.

---

3. Master ranking of single-stock catalyst strength

The PE column is uniformly PE(TTM) (see the header basis note). Names whose rolling calculation was not completed are marked "not obtained"; forward PE is not used as filler.

---

4. Single-stock scoring model (100 points)

One rule correction produced by this audit: the draft scored 三花/拓普 under "supply-chain indirect = 5–8", but when the supply relationship is not confirmed by filing, this model's own basis puts it at 5 (bottom of the range), not the top. Both scores have been cut accordingly and their labels downgraded, see rows 4 and 5 of §3.

Item-by-item scores (key names)

Note: risk deductions cap at −15, but the actual deduction items for 盛达 and 兴业 already far exceed that cap. When the deduction is truncated by the cap, this report downgrades the label rather than only the score — which is why, despite scores of 58/38, their labels are "watch only" and "Pass" respectively.

---

5. Detailed analysis of the top 10 names

1光启技术002625Priority deep-dive ｜ 84 pts · Kuang-Chi Technologies · SH/SZ main board ±10%

Related news: 2026-09-17 19:06:24 (eiTime, after the close) ｜ filing 2026-072 "Voluntary Disclosure Announcement on the Signing of Metamaterial Product Mass-Production Contracts" ｜ Securities Times⁴ Core: wholly-owned subsidiary Kuang-Chi Advanced signed mass-production contracts with three customers totalling RMB 907 million, of which one customer accounts for RMB 862 million and the other two for RMB 45.2942 million combined.

The dedicated "is this old news" check:

• 2026-01-08 filing RMB 264 million ｜ 2026-06-22 filing RMB 612 million ｜ 2026-09-17 filing RMB 907 million ← this one
• "Cumulative" in the filing refers to the total across these three customers in this batch, not a year-to-date cumulative. → Confirmed as incremental, and the largest of the three 2026 batches.
• Reference frame: from 2025-01-01 to 2026-06-22 Kuang-Chi Advanced signed cumulative mass-production contracts of RMB 3.749 billion; this single batch is ≈24% of that 18-month cumulative total.

Fundamentals (2026 interim report, disclosed 8/25): revenue RMB 1.128 billion (+19.59%), net profit attributable to parent RMB 319 million, net operating cash flow +RMB 167 million. → 907 ÷ 1,128 = 80.4% (reverse-computed).

Valuation: PE(TTM) 97.55, PB 6.15.

⚠️ The draft wrongly used forward PE 94.85 here (= RMB 60.479 billion ÷ RMB 319 million ×2); corrected to the TTM basis, 97.55.

Branch stage: no peer ladder, no sector anchor, no thematic companions — an isolated single-stock event, not dependent on sentiment, and with no concentrated money behind it.

Technical sentiment: closed 9/17 at RMB 28.07, −0.67%, turnover RMB 828 million. −27.46% below the 60-day high. Not an extended name, no risk of a locked limit-up.

Risks: ① a single customer accounts for RMB 862 million = 95.0% of this batch; ② a mass-production contract ≠ current-period revenue; delivery and acceptance timing were not disclosed in the filing, so this report cannot determine whether the revenue lands in 2026H2 or 2027; ③ PE(TTM) 97.55 — not a cheap valuation.

Final judgement: priority deep-dive. Highest evidence grade, most computable amount, cleanest expectation gap, safest position. An isolated event stock — give it a single-name position, not a sector position.

---

2山东黄金600547Watch closely ｜ 74 pts · Shandong Gold · SH/SZ main board ±10%

Related news: SHFE gold 2612 night session +0.78% (9/18 02:30); on 9/17 the A-share gold sector was −5.15% vs SHFE gold day session −0.59%, a same-period divergence of −4.56pp.

Correction regarding GDX: the draft said "GDX +3.36% vs A-shares −5.15%, an 8.5pp gap; after deducting the night-session SHFE gold +0.78%, the true oversold amount is about 4.6pp". That subtraction is wrong: 8.51 − 0.78 = 7.73, while 4.56pp is a completely different quantity that contains no GDX at all (A-share gold sector vs SHFE gold day session). Two numbers on different bases were stitched into one causal claim. Now corrected to:

• The computable oversold amount = −4.56pp (same day, same currency, same closing moment).
• GDX +3.36% (9/17 16:00 ET) serves only as directional corroboration; because of the 13-hour time difference, different currency and different constituents, it takes part in no subtraction.

Why it replaced the silver names as the primary pick in this branch: the sub-agent's interim-report drill-down confirms self-mined gold is 88.44% of gross profit, giving roughly +26.9% attributable-profit elasticity to a +10% gold price — higher than any of the three silver names' elasticity to silver. Its silver elasticity is near zero (silver is only a by-product of gold-concentrate smelting, refining capacity 50 tonnes/year), but what needs expressing today is "precious metals were unfairly sold", and gold is the cleaner vehicle.

Fundamentals (including the negatives that must be written out): PE(TTM) 27.67, PB 4.64, market cap RMB 151.5 billion. H1 revenue RMB 53.588 billion (−5.60%), attributable profit RMB 3.543 billion (+26.17%), operating cash flow +RMB 7.771 billion.

• 🔴 Goodwill of RMB 12.637 billion, 26.8% of RMB 47.137 billion of attributable net assets.
• 🔴 2026-06-02 filing: the subsidiary's Donghai Securities stake participating in the Dongwu Securities acquisition is expected to reduce 2026 total profit by RMB 705 million (unaudited).
• 🔴 Mined gold output was 19.10 tonnes, −22.70% YoY; sales volume −15.42%. This is a fact that has already occurred, not a risk — the +26.17% profit growth came entirely from the gold price, with volume a negative contributor. Applying the same yardstick this report applied to 兴业银锡 (shrinking output = disqualifier), this should carry the same disqualifying weight; this report's treatment is: 兴业's suspension is "future output may break", while 山东黄金's −22.70% "has already happened and has already been digested by the share price", hence the former was downgraded by label while the latter takes a heavy score deduction (−8) but keeps its label. The reason for the differential treatment is written here so readers can judge for themselves whether they agree.
• 🟠 Minority interests of about RMB 2.19 billion = 38.2% of total net profit, significantly diluting elasticity on an attributable basis; gearing 61.90%.
• 🟢 Share purchases by the controlling shareholder and parties acting in concert have been completed (2026-04-09).

Technical sentiment: closed 9/17 at RMB 32.86, −6.46%, turnover RMB 3.732 billion (largest among gold names). −13.30% below high, −7.67% over 5 sessions, +38.94% over 60 days.

Final judgement: watch closely. Best liquidity, computable same-period divergence, highest gold-price elasticity; the cost is a leader's smaller elasticity, plus goodwill, minority dilution and falling output — three long-term discount factors.

---

3紫金矿业601899Watch closely ｜ 73 pts · Zijin Mining · SH/SZ main board ±10%

Related news: SHFE copper 2610 night session +1.42% (9/18 02:30); LME copper touched $14,500/tonne, a record high (09-17 22:29); COPX +2.87% (9/17 16:00 ET).

It has one feature nobody else has: on 9/17 the copper sector was −2.37% while SHFE copper rose that day (+0.71%). Copper is the only one of the three metals where "the commodity rose and the equity fell" — the other two are "commodity fell slightly, equity fell a lot", which requires explaining the gap in magnitude; copper is opposite in direction and requires no explanation at all.

Fundamentals: PE(TTM) 12.14, PB 4.25, market cap RMB 821.7 billion — the cheapest and most liquid resource name in the candidate pool.

⚠️ This report did not perform a line-by-line interim-report drill-down on 紫金 (the sub-agent's scope was the four silver names). Self-mined share, cost curve and minority dilution are all unverified.

Technical sentiment: closed 9/17 at RMB 30.90, −3.95%, turnover RMB 9.523 billion, net main-force outflow RMB 1.183 billion (largest among resource names that day). −11.79% below high, −9.41% over 5 sessions.

Risk: the largest net main-force outflow at RMB 1.183 billion — institutions were cutting yesterday, and that selling pressure will not clear in a single day.

Final judgement: watch closely. Clearest divergence direction + lowest valuation + best liquidity, but the heaviest main-force selling yesterday.

---

4拓普集团601689Watch only (⬇ "watch closely" in draft) ｜ 66 pts · Tuopu Group · SH/SZ main board ±10%

Related news: 09-17 19:43, Tesla robot mass production accelerating, the team landed in Ningbo on 9/16 and began a new round of mass-production plant audits on 9/17 ｜ Leaderobot⁷

The most valuable observation in this branch: the 8 auto-parts names that hit limit-up yesterday have nothing to do with this catalyst.

Group	Names	9/17 move
Limit-up group (8 names)	众泰汽车 (Zotye Auto), 山子高科 (Shanzi High-Tech), 均胜电子 (Joyson Electronics), 北特科技 (Beite Technology), 冠盛股份 (Guansheng), 天海电子 (Tianhai Electronics), 克来机电 (Cenker), 金杯汽车 (Jinbei Auto)	All limit-up
Robot core group	拓普集团 (Tuopu Group) +4.36%, 五洲新春 (Wuzhou Xinchun) +3.65%, 三花智控 (Sanhua) +2.70%, 双环传动 (Shuanghuan) +1.85%, 恒立液压 (Hengli Hydraulic) +0.96%, 绿的谐波 (Leaderdrive) +0.94%, 蓝思科技 (Lens Technology) +0.39%, 汇川技术 (Inovance) +0.29%	Not one limit-up

Table scrolls horizontally

The sub-agent ran filing drill-downs on four of the limit-up group (800 filing titles each + full-text grep of the 2026 interim report), concluding:

⚠️ Precise statement: within the scope searched — namely each company's most recent 800 filing titles, plus the full text of the 2026 interim report — none of the four shows "Tesla" as its own customer. That search did not cover: 2025 and earlier annual-report texts, prospectuses / offering documents (except 北特, searched separately), exchange inquiry responses, or the complete set of investor-relations records. So this is "not found within the scope searched", not "does not exist in any filing". The draft's universal-negative wording has been corrected.

• 山子高科: all 6 occurrences of "robot" in the interim report are subsidiary names and business scope, revenue 0; the company already denied participating in the Nezha restructuring on 2026-09-15 — the very limit-up rationale the market is quoting has been denied by the company.
• 克来机电: all 36 occurrences of "robot" in the interim report are industrial robots / AMRs; the controlling shareholder's reduction plan is in progress.
• 均胜电子: it does have a 100%-owned embodied-intelligence robotics subsidiary (real), but 均普智能 (Junpu Intelligent, 688306), the group entity that actually does robotics, rose only 1.03% that day while 均胜 was limit-up at 10.02% — if the catalyst were really robotics, the pure-play not moving while the mapped name limits up is an internally inconsistent price structure.
• 北特科技: the only one with registration-document-level screw evidence, but the interim report states explicitly that "the screw product has not had a material impact on results", and PE(TTM) is 128.1.

→ What went limit-up yesterday was not "Tesla robots", it was low-priced restructuring/shell themes. On the core names this plant-audit catalyst delivered only 0.29%~4.36%. The humanoid-robot concept sector was +0.60% that day with net main-force inflow of only RMB 78 million, which matches.

That is the source of the expectation gap. But this report cut the label from "watch closely" to "watch only" for two reasons:

1. ⚠️ This report obtained no filing-level confirmation of 拓普's own Tesla Optimus supply relationship. The sub-agent's drill-down scope was the four limit-up names, excluding 拓普 and 三花. The claim is widely reported by industry media but was not verified by this report today. Under the §4 scoring basis, a supply-chain indirect relationship not confirmed by filing can score only 5 (bottom of the range) — and this report has already refused to recommend NPO concept names on exactly that ground. The same yardstick must be applied across the board.
2. The core group is down 20–26% across the board over 60 days and sits −22%~−40% below highs. This is a line inside a downtrend, and one plant audit is not enough to reverse it.

Fundamentals: PE(TTM) 30.34, PB 3.14, market cap RMB 76 billion.

Note: the draft used forward PE (拓普 37.19 > 三花 35.66) to conclude "拓普's quality is below 三花's". On the TTM basis the order reverses (拓普 30.34 < 三花 36.47), and that judgement has been withdrawn.

Technical sentiment: +4.36% on audit day, the strongest reaction in the core group; −35.74% below the 60-day high (second deepest in the core group).

Final judgement: watch only. The expectation gap holds, the position is the best, the valuation is the lowest in the core group, but the single-stock supply evidence was not verified by this report, which is not enough to support a "watch closely" or higher label.

---

5三花智控002050Watch only (⬇ "watch closely" in draft) ｜ 65 pts · Sanhua Intelligent Controls · SH/SZ main board ±10%

Related news and branch analysis: same as ④. Differences: only +2.70% on audit day, a weaker reaction than 拓普, so the degree of "not priced in" is higher; but PE(TTM) 36.47 is above 拓普's 30.34, so it is more expensive; market cap RMB 145.8 billion, with the best cash-flow quality and scale in the core group. Technical sentiment: −28.89% below the 60-day high, −20.66% over 60 days. Risks: as with 拓普, this report obtained no filing-level confirmation of its Tesla supply relationship; the 60-day trend is down; Tesla's robot mass-production plans have been pushed back repeatedly in the past. Final judgement: watch only. The degree of "not priced in" is the highest, but the evidence gap is identical to 拓普's, so the label was cut in step. If choosing one of the two within the core group: 拓普 wins on position and valuation, 三花 wins on scale and cash-flow quality.

---

6兴业银锡000426Watch only ｜ 58 pts · Xingye Silver & Tin · SH/SZ main board ±10%

It should have been number one: the sub-agent's interim-report drill-down confirms mined silver revenue of RMB 2.511 billion, 58.64% of revenue and 61.97% of gross profit, with mining at 99.88% of revenue (pure self-mined ore, zero smelting), and hedging only on tin, not on silver (silver exposure is fully naked). H1 mined silver output was 158.08 tonnes, +20.4% YoY. Silver +10% → attributable-profit increment +11.4%, the truest elasticity of the four. Financials are the healthiest: ROE 19.61%, operating cash flow +RMB 2.113 billion, gearing 38.02%, goodwill only RMB 193 million. PE(TTM) 19.36. The technical position is also the best: −1.68% over 60 days (lightest trapped profit), −10.60% over 5 sessions (deepest drop), −19.39% below high.

Disqualifier:

🔴 The core Yinman Mining pit has been suspended since the underground accident of 2026-07-26 and no resumption announcement has appeared. The 9/3 filing (No. 2026-81) approved resumption only of the "ore-dressing and tailings system". On 7/28 the company itself said surface ore stock stood at 350,000 tonnes, enough to supply the concentrator for about two and a half months — counting from 7/26, the buffer runs out around mid-October.

⚠️ Basis warning on the "44.8% of H1 attributable profit" figure (audit finding, must be stated): the numerator, RMB 1.014 billion, is Yinman Mining's total net profit (including minorities), while the denominator is the listed company's attributable net profit of RMB 2.263 billion — the two are on inconsistent bases and should not be divided directly. Moreover, the two main mines the company names — Yinman RMB 1.014 billion + Yubang RMB 444 million = RMB 1.458 billion — account for only 64.4% of the RMB 2.263 billion, and the source of the remaining roughly RMB 800 million is not accounted for in this report. "44.8%" should therefore be treated as an order-of-magnitude reference rather than a precise share, and this report did not obtain 兴业's ownership percentage in Yinman. If ownership is materially below 100%, the magnitude of this disqualifier needs recomputing.

An honest note on duration (audit finding): this report defines the branch as a 1–3 day repair trade, while the suspension risk does not bite until mid-October. Strictly speaking, an output risk that only takes effect in Q4 does not disqualify a 1–3 day price-elasticity trade. This report still downgrades it, on the grounds that repair trades frequently turn into positions people cannot let go of, and once this name goes from 3 days to 3 weeks it runs straight into the output break. This is conservatism about holding-duration spillover, not a claim that it cannot rise today. A reader who strictly enforces the 1–3 day discipline may promote it back to a watch slot.

Two further risks: controlling shareholder Inner Mongolia Xingye Group's holding of 363,320,020 shares is entirely pledged (100%); mined tin output was already −23.2% in H1. The part not affected: subsidiary Yubang Mining (H1 revenue RMB 1.022 billion, net profit RMB 444 million, the largest single silver-mine resource in Asia) is unaffected by this accident.

Final judgement: watch only. The elasticity is real; the sustainability of the output base is the question mark. If a pit-resumption announcement appears, it immediately becomes the best name in this branch — a clear, waitable trigger (§8⑦).

---

7盛达资源000603Pass(⬇⬇ a recommended slot in the draft) ｜ 38 pts · Shengda Resources · SH/SZ main board ±10%

🔴 On 2026-06-16 both the company and actual controller Zhao Mantang were placed under investigation by the CSRC (case nos. 0142026038 / 0142026039), both for "suspected violations of information disclosure rules", pointing to the non-operating fund occupation by the controlling shareholder and its affiliates disclosed on 2025-04-29. More than three months on, the investigation's conclusion and penalty consequences are unknown.

The business side is actually clean: 88.28% self-mined ore, gross margin 69.15%, ex-non-recurring / attributable 98.0% (the cleanest profit of the four), no goodwill, no material litigation, naked silver exposure. H1 revenue RMB 1.152 billion (+27.08%), attributable profit RMB 390 million (+456.46%). PE(TTM) 26.57, PB 6.03.

But two other problems are equally present:

• The silver share cannot be broken out. The company discloses by concentrate type rather than by metal, and silver is "priced separately" within lead and zinc concentrates. The sub-agent checked three investor-relations records dated 9/4, 9/8 and 9/14; institutions asked about "expected silver metal output this year" every time, and all three times the company replied identically: "please refer to the company's 2026 annual report". The absolute tonnage is not confirmable for now; the back-solved "silver revenue of about RMB 430 million, roughly 37% of total revenue" is an estimate, not a disclosed figure.
• Main-mine depletion: from the IR record itself — Yindu Mining's Bairendaba "has been mined for 20 years… the current actual production scale is about 500,000–600,000 tonnes per year" (design 900,000 tonnes/year). Incremental ore is "expected to complete construction in 2028", and the only near-term increment is the Honglin copper-gold mine, which contributes gold and copper, not silver.

Final judgement: Pass. A company under CSRC investigation for "information disclosure violations" that simultaneously refuses to disclose output of its core metal — put those two facts together and any fundamental reasoning built on it lacks credible foundations. It was −6.92% yesterday, deepest in the sector, and technically looks most oversold, but this report will not trade regulatory uncertainty for a 1–3 day repair move.

---

8湖南白银002716Pass40 pts · Hunan Silver · SH/SZ main board ±10%

Item	Figure
Silver revenue / share of revenue	RMB 6.126 billion / 70.26%
Silver gross margin	5.05%
Self-mined ore (mine products) as % of revenue	2.94% (gross margin 68.77%)
Smelting and other as % of revenue	97.06% (gross margin 4.74%)

Table scrolls horizontally

The strongest evidence is what the company wrote itself (from the interim report's "risk analysis of derivative positions"):

"The company's current lead and silver smelting capacity exceeds the supply from its own raw-material production, requiring external purchases of lead concentrate, silver concentrate, crude lead and other inputs… when purchased raw materials have arrived but not been priced, a rise in market prices will produce cost inversion… it is a risk type with exposure on both the upstream and downstream sides."

That is: on the raw-material side a rising silver price is cost pressure, and the company itself lists it as a risk rather than a positive. The company also actively hedges silver with futures, further flattening the exposure.

Financials: PE(TTM) 60.37 (highest of the four), PB 5-year percentile 90.5%; operating cash flow −RMB 1.377 billion (OCF/net profit −8.2×); inventory rose from RMB 1.555 billion to RMB 2.808 billion (+80.6%) in half a year, equal to 69% of attributable net assets, while cash fell from RMB 1.277 billion to RMB 546 million over the same period; unrecovered losses of −RMB 2.769 billion exceed 1/3 of paid-in capital, with no distribution for 2025; the only one with a recent 5% shareholder reduction record.

The one possible counter-case: if the RMB 2.808 billion of inventory is mostly silver and not fully hedged, a +10% silver price theoretically corresponds to about RMB 280 million of potential revaluation. But (a) inventory is measured at the lower of cost and net realisable value, so a price rise does not book a gain; (b) the company already hedges; (c) the supplier prices the raw material. The sign cannot be determined from public information. Conversely, if silver falls, that RMB 2.8 billion of high-cost inventory is a clear impairment risk.

Final judgement: Pass.

---

9山金国际000975Watch only ｜ 66 pts · Shanjin International · SH/SZ main board ±10%

Differentiating fact: all 13 precious-metal names fell on 9/17, and 山金国际 was the only one in the sector with net main-force inflow (+RMB 31 million), despite the share price being −5.66%. Fundamentals: PE(TTM) 18.29, PB 4.36, market cap RMB 69.4 billion — the cheapest tier among gold names.

Note: the forward PE of 14.35 cited in the draft has been dropped; the TTM basis is 18.29.

Technical sentiment: −14.49% below high, −3.29% over 5 sessions. Risks: this report did no interim-report drill-down on it; self-mined gold share, cost curve and minority dilution are all unverified; and it carries in-group homogeneity risk with 山东黄金. Final judgement: watch only. The money evidence is there; the fundamental evidence is missing.

---

10双环传动002472Watch only ｜ 64 pts · Shuanghuan Driveline · SH/SZ main board ±10%

Differentiating fact: −7.26% over 60 days, the only single-digit decline in the robot core group; PB 2.97, the lowest in the core group. Technical sentiment: +1.85% on audit day, −22.35% below high. Risks: no filing-level order; as with 拓普/三花, not verified at filing level by this report; PE(TTM) not obtained. Final judgement: watch only. If this line really starts today, it is the "catch-up + low valuation" position, but it will not be the one that moves first.

---

6. Pass list

---

7. Intra-branch ranking

Branch 1: Precious metals (full sample of 13 names, not one rose yesterday)

• Hardest: 山东黄金 (Shandong Gold, 600547). Truest elasticity but unusable: 兴业银锡 (Xingye Silver & Tin, 000426). Highest trading recognisability but a regulatory problem: 盛达资源 (Shengda Resources, 000603).

Branch 2: Humanoid-robot core chain (the whole branch has been cut from A to B+, with no name in the first tier)

---

8. Opening verification signals for today

⚠️ An honest note on "independence" (audit finding): the draft claimed the seven verification points did not overlap. That does not hold. ① and ③ belong to the same precious-metals branch and ③ is downstream of ① — once the gold price gives back, sector breadth almost certainly fails at the same time. The table below marks each point's dependency, and the draft's two points that treated "net main-force inflow" as independent evidence have been deleted — that indicator is back-solved from the day's price action and order-size split, is near-tautological with price direction, and does not constitute independent evidence.

Auction and open (09:15–09:30)

• Gap-up benchmark: FTSE A50 futures night session +0.36% (read at 05:19), SOX +3.14% (9/17 16:00 ET close), S&P futures −0.13% (read at 9/17 18:53 ET, the overnight gain already given back). The three readings are at different timestamps and cannot be added. An index gap-up of 0.3%–0.8% is normal; over 1% is overheated sentiment.
• Risk pattern: yesterday's failed-limit-up rate was 29.9%, median seal-order/turnover 0.009. If the number of limit-ups cannot exceed 20 before 10:00 after a gap-up, the gap is pure offshore mapping with no endogenous relay.

Verification points

⚠️ One thing this report did not do: the thresholds above carry no base rates. It should have stated "over the past 20 trading sessions, how often did this sector have ≥3 names up more than 5%" — if the historical hit rate is already 60%, then that verification point is testing the index rather than the branch. The back-calculation was not done within the delivery window; readers should discount the thresholds accordingly.

Risk signals (cut exposure on any two)

1. Gap-up then dive: the index gaps up >0.8% and turns red before 10:00.
2. Lone-survivor pattern: only 1–2 precious-metal names rise sharply while the rest diverge.
3. Anchors do not follow: 山东黄金 / 紫金矿业 clearly lag small-cap high-beta names (speculative money rather than institutional allocation).
4. Yesterday's strong themes keep ebbing: 澳弘电子 (Aohong Electronic, 605058) breaks its 5th limit-up at the open, 万向德农 (Wanxiang Denong, 600371) breaks its seal — these two are the pillars of yesterday's sentiment.
5. 金帝股份 (Jindi, 603270) opens sharply lower on its huge unlock (46.67% of free float) and drags down the small-cap robot ball-screw names.
6. US futures keep weakening intraday.

---

9. Final recommendations

① The 5 names most worth watching today (split into two tiers with different evidence grades)

Four changes versus the draft (all triggered by the fundamentals drill-down and the QA pass):

1. 盛达资源 removed from a recommended slot over the CSRC investigation → Pass;
2. 兴业银锡 cut to watch only over the core mine suspension (with a duration note);
3. 三花/拓普 cut from "watch closely" to a second-tier trial position because the Tesla supply evidence was not verified by this report — the same yardstick used to refuse recommending NPO concept names;
4. PE switched entirely to the TTM basis, on which the draft's "拓普's quality is below 三花's" judgement was withdrawn (on TTM, 拓普 30.34 < 三花 36.47 — the reverse order).

First tier ｜ evidence meets the bar (filing-level or computable within one timestamp)

Second tier ｜ ⚠️ evidence below the bar · trial position (supply relationship not confirmed at filing level in this report)

Dimension and independence check: ① order (single-stock filing) ｜ ② precious-metal price ｜ ③ industrial-metal price (different commodity) ｜ ④⑤ industry event (2 slots, capped and already downgraded). Of the five verification points, ①②④⑤⑦ are mutually independent and ③ is marked as downstream of ①. The draft's two verification points based on "net main-force inflow" have been deleted (that indicator is near-tautological with price and does not constitute independent evidence).

② Today's three strongest branches

The humanoid-robot branch has been dropped out of the third slot because the single-stock evidence grade is insufficient (see §2 expansion (four)).

③ Directions not worth chasing today

④ Final one-sentence judgement

Today is a day of "offshore handed us a gap-up and yesterday's mis-selling handed us a hole", but after walking both of the most tempting paths, less survives than one would think: silver, which rallied hardest, has no clean A-share vehicle (an investigation, a suspension and a smelting business take one name each), yesterday's limit-up auto-parts names have nothing to do with Tesla robots within the filing scope searched, and the robot core names' own supply evidence was not verified in this report either — so they merit only a second-tier trial position. Only three names actually clear the evidence bar: 光启技术 with a filing-level hard order, 山东黄金 with the −4.56pp same-period divergence and the highest gold-price elasticity, and 紫金矿业 as the only "commodity up, equity down" case. Two more things to remember: first, the RMB gold price rose only 0.78% overnight, not the 2.4% of the USD basis, so the elasticity base must be the smaller one; second, a 55.4% October hike probability hangs overhead and the relay structure behind a 29.9% failed-limit-up rate has not repaired, so give size and not duration — and place no bets on the property chain before the 10:00 MOHURD press conference.

---

⚠️ Data-fetch failures, correction log and open items for this edition (internal)

A. Data-fetch failures

1. push2his.eastmoney.com blocked across the board (every kline request RemoteDisconnected), which also made akshare.stock_zh_a_hist unusable. Fallback channel: Tencent web.ifzq.gtimg.cn, working normally. All "past 5/20/60 sessions, distance from high" figures were computed in-house from that channel's raw daily bars. push2delay / push2ex / datacenter-web are fine.
2. First fetch of the official limit-up pool failed silently: writing sort as the URL-encoded fbt%3Aasc returned tc=47 with pool=[]. Switching to the unencoded fbt:asc fixed it (47=47). A textbook "tc non-zero but pool empty" case, caught by the validation rule.
3. Sina's offshore quote endpoint hq.sinajs.cn returned Forbidden; gold/silver/copper daily bars were switched to akshare.futures_zh_daily_sina + CNBC quote, cross-checked across two sources.
4. CNBC's chart endpoint ts-api.cnbc.com/harmony/app/bootstrap-charts returned 404 (path retired); only quote is usable.
5. All three reportNames of Eastmoney's unlock endpoint RPT_LIFT_* return "report configuration does not exist"; switched to akshare.stock_restricted_release_summary_em / _detail_em.
6. sc crude 2611 night session showed −5.28%, badly inconsistent with WTI (−0.69%), likely a continuous-contract rollover or prior-settlement basis issue; this report used no crude oil data at all.

B. Errors found and corrected by QA (executed by risk-auditor; this section is the correction list) 7. 🔴 Arithmetic error (corrected): the draft wrote in §5② / §9 that "an 8.5pp GDX gap, less the night-session SHFE gold +0.78%, means true mis-selling of about 4.6pp". 8.51 − 0.78 = 7.73, not 4.6; and 4.56pp is a completely different quantity that contains no GDX. This was the number-one rationale for the second recommended slot. Now changed to: use only the same-timestamp, same-currency −4.56pp, with GDX as directional corroboration only and not part of any subtraction. 8. 🔴 Mixed valuation bases (corrected): the draft mixed forward PE with PE(TTM), and the ★ marks happened to cluster on the downgraded silver names, systematically making downgraded names look expensive and recommended names look cheap. Actual TTM vs forward PE differences: 先导基电 815.50 vs 82.91 (10x), 山金国际 18.29 vs 14.35, 紫金 12.14 vs 10.49, 拓普 30.34 vs 37.19, 三花 36.47 vs 35.66, 光启 97.55 vs 94.85. The whole report is now on TTM, with "not obtained" written where conversion was impossible. 9. 🔴 A judgement withdrawn because of the basis correction: the draft used forward PE to say "拓普's quality is below 三花's"; on TTM the order reverses, and the judgement has been withdrawn. 10. 🔴 Double standard (corrected): the draft refused to recommend NPO names for "no filing-level evidence" while letting 三花/拓普, on exactly the same evidence grade, occupy 2 of the 5 recommended slots. Both have been cut to a second-tier trial position with directness scores cut to the bottom of the range (5 points), with the rationale written into the §2 expansion (four) and the body of the §9 table. 11. 🔴 Universal negative beyond the search scope (corrected): the draft wrote "not one of them has ever shown 'Tesla' in any filing", whereas the actual scope was 800 filing titles + the 2026 interim-report text, excluding prior-year annual reports, prospectuses and inquiry responses. Changed to "not found within the scope searched", with the uncovered document types listed. 12. 🟠 Inconsistent currency basis (corrected): the draft cited USD spot gold +2.4% for the catalyst throughout, switching to the RMB +0.78% only where the GDX subtraction was unfavourable to it. The whole report has been converted to the RMB basis, with the unexplained 1.6pp gap stated in all three of §0/§2/§9. 13. 🟠 Sub-sample extrapolation (corrected): the draft extrapolated the three silver names' "low PE percentile + high PB percentile" onto the entire precious-metals branch, recommended names included. The scope has been limited, with a statement that 5-year percentiles for 山东黄金/紫金/山金国际 were not obtained. 14. 🟠 Verification points not independent (corrected): the draft claimed all seven were non-overlapping, whereas ①③④ all belong to precious metals and are conditionally dependent, and two points using "net main-force inflow" as the criterion are near-tautological with price. Those two have been deleted, ③ has been marked as downstream of ①, and the 拓普/三花 criterion has been switched from relative strength to an absolute one (not breaking below the 9/17 close) — the original criterion would have passed even with both down 5%~6%. 15. 🟠 A ratio computed on inconsistent bases (warning added): in "Yinman is 44.8% of H1 attributable profit", the numerator is the subsidiary's total net profit (including minorities) and the denominator is attributable net profit, so they cannot be divided directly; and Yinman 10.14 + Yubang 4.44 = RMB 1.458 billion is only 64.4% of the RMB 2.263 billion, with the remaining roughly RMB 800 million unaccounted for. A basis warning has been added in §5⑥, noting that 兴业's ownership percentage in Yinman was not obtained. 16. 🟠 Duration mismatch (note added): using a suspension risk that only bites in October to disqualify a 1–3 day trade does not close logically; and 山东黄金's already-realised −22.70% output decline only cost points without a label cut. The reasons for the differential treatment are now written in §5⑥/§5②, for readers to judge. 17. 🟠 Subtraction across read timestamps (corrected): the §8 gap-up benchmark mixed 05:19 (A50) with 18:53 ET (S&P futures); §0 stitched the day-session divergence together with the overnight move in one sentence. Timestamps have been annotated point by point and the sentences split.

C. Items explicitly unverified in this report and flagged in the body 18. 拓普/三花/双环's own Tesla supply evidence — the sub-agent's drill-down scope was only the four limit-up names. This is the core gap for the two second-tier names. 19. Interim-report drill-downs on 紫金矿业, 山金国际, 赤峰黄金 and 湖南黄金; PE(TTM) was not obtained for these four other than 山金国际 and 紫金. 20. Independent readings of 湖南黄金's "antimony price" and 洛阳钼业's "cobalt price". 21. Delivery and acceptance timing for 光启技术's RMB 907 million contract (not disclosed in the filing). 22. All §8 thresholds lack base rates (the pass frequency over the past 20 trading sessions should have been back-calculated); not done because of the delivery window, and declared in the body. 23. The cause of the 1.6pp gap between spot gold +2.4% and SHFE gold night session +0.78% (FX / internal-external spread / contract-month differences). 24. 兴业银锡's ownership percentage in Yinman Mining.

D. Process 25. All sub-agents were run and heavily adopted: fundamentals-analyst×2 (four silver names, four auto names), risk-auditor×1. Of the 17 red flags raised by QA, all 6 high-severity items were corrected, 6 medium-high items were corrected or given warnings, and of the 3 medium-low items (threshold base rates, minor PB drift, QA disclosure) the first two were not done for time reasons and have been declared. 26. Timing of the Huawei Ascend release: Cailianshe flash timestamps were 16:38–18:04 (after the close), but the actual time of Wang Tao's keynote could not be confirmed, and media previews already existed pre-market on 9/17. The body does not argue "released after the close therefore completely unpriced"; it uses price-volume evidence instead (Ascend concept −0.50%, net main-force outflow RMB 348 million).

---

⚠️ Risk warning: this list is a pre-market information review and watchlist 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.