A-Share Pre-Market Brief | 2026-08-06 Thursday

Data conventions disclosed up front (please read this first)

1. News time window: 2026-08-05 (Wednesday) 15:00 close → 2026-08-06 07:00. Anything outside the window is listed separately and down-weighted in Section 1.3 "Old-News Screening", and is not counted as a new catalyst for today.
2. Market-data convention: throughout this brief, A-share closing prices, percentage moves, turnover rates, turnover value, free-float market cap, trailing P/E and net main-force inflow are all as of the 2026-08-05 close, taken from the Tencent Finance single-stock snapshot (stock_zh_a_spot_tx, 5537 rows). The limit-up pool / broken-limit-up pool / limit-down pool / consecutive-limit-up ladder come from East Money (stock_zt_pool_em etc.). Industry percentage moves and industry net inflows come from the THS (Tonghuashun) industry taxonomy (90 industries) — industry net inflow and single-stock net main-force inflow are two different conventions and must not be added together.
3. Daily price limits differ by board, so every stock in this brief is tagged with its board: SH/SZ main board ±10%, ChiNext (30/301) ±20%, STAR Market (688) ±20%, BSE (starting with 8/9) ±30%. On ChiNext / STAR Market, +12% is not a limit-up — do not read a ±20% stock with ±10% intuition. The STAR Market and the BSE additionally have an RMB 500,000 investor-eligibility threshold.
4. Overnight overseas convention: "overnight" in this brief means the US session of 2026-08-05 (Wednesday), which closed at 04:00 Beijing time on 2026-08-06. Data come from the closing wrap-up and commodities wrap-up published by Sina Finance's US-stock channel in the early hours of 08-06.
5. "Net main-force inflow" is the East Money convention (large + extra-large orders). THS's "real-time fund flow" uses an active-buy/active-sell convention, and the two can point in opposite directions for the same stock (e.g. Grinm Advanced Materials: East Money +RMB 815 million, THS −RMB 331 million). The two conventions must not be mixed or added together.
6. Where gold and silver prices actually sit (the single most important premise in this brief): the overnight +4.1% in gold is a rebound after a deep correction, not a new high. Gold peaked in January 2026 (SGE Au99.99 month-end RMB 1163.95/gram, corresponding to roughly USD 5626.8/oz on COMEX); the current 4246.82 is still about 24.5% below that peak; silver fell from RMB 27530/kg in January to 14988, a halving that has not been repaired. Moreover, the SGE gold average price went Q1 1088.6 → Q2 990.4 → Q3-to-date 888.6, so even after the overnight gain the Q3 average is still materially below Q1/Q2. Every judgment on precious metals in this brief rests on this premise.
7. Items not obtained / left blank in this brief:
   • Single-day northbound net inflow: cannot be provided. Since August 2024 the exchanges have discontinued real-time and post-close disclosure of single-day northbound net buying; this is a disclosure gap, and this brief will not substitute any proxy data for it.
   • The 2026-08-05 Dragon-Tiger List: not obtained by press time; this brief lists no Dragon-Tiger seats and draws no "hot-money preference" inferences from them.
   • Unitree Robotics' final offer price: August 5 was the preliminary book-building day, and the offer price will be announced on August 6; it had not been disclosed by press time. The RMB 104/share cited here is a market estimate back-solved from the fundraising amount, not a confirmed price.
   • The August 6 A-share trading calendar (new listings / lockup expiries / trading halts and resumptions): a complete list was not obtained, so this brief does not include one.

⚠️ Data-sourcing and implementation notes for this run (internal only):

• We very nearly stepped into the same trap as yesterday, in a stealthier version. The content scraped from the THS "Morning View" page (stock.10jqka.com.cn/zaopan/) called itself the "Morning View for August 6, 2026" and said "Nasdaq led with +2.59%, S&P +1.79%, Dow +1.71%" and "gold at USD 4071.66". That entire page was actually the August 5 morning edition — the page title's date auto-tracks the current day while the body is not updated. The tell is that those figures exactly match the August 4 US close (Nasdaq 26584.99, +2.59%). Taking it at face value would have shifted the whole brief's "overnight" by one trading day, and gold would have been written as 4071 instead of 4246 — which would have erased today's strongest catalyst outright. We ended up cross-checking against the original Sina US-stock-channel copy from the early hours of 08-06 (closing wrap-up published 04:08). → Suggest hardening this into a process assertion: whenever citing "overnight overseas", first back-check the publication date using the three major indices' closing levels; if the levels don't match that day, judge the page stale. This is the second form of the same failure class as yesterday's recap note on "the Sina daily-bar endpoint silently lagging by one day" — a lagging source throws no error, it just quietly hands you yesterday's numbers.
• Search engines failed systematically on the date "August 6": repeated queries either replied "August 6 has not arrived yet" or returned content from August 5 or even June 8. All the useful overnight information was obtained only by using WebFetch on the Sina US-stock channel listing page, getting the real URLs, and fetching each article. Search tooling is unreliable across a day boundary; early-morning news must go through the "listing page → get URL → fetch body" route rather than relying on search snippets.
• stock_board_industry_summary_ths returns 08-05 industry data and is usable (90 rows). But stock_board_industry_cons_ths does not exist as a function in this machine's akshare 1.18.79 (AttributeError); sector constituents were instead looked up from a hard-coded code list.
• The numeric columns in the Tencent snapshot stock_zh_a_spot_tx are string-typed. The first run of "net main-force inflow TOP25" sorted them as strings, producing a jumbled table mixing 998, 9971 and 98040 — which we nearly wrote into the report as a real fund-flow ranking. pd.to_numeric has been added and the sort redone. Suggest doing numeric coercion uniformly inside the data-fetch wrapper; this trap throws no error, it just hands you a fake ranking.
• East Money's limit-up pool stock_zt_pool_em(20260805) returned 103 rows normally, the broken-limit-up pool 43 rows, the limit-down pool 1 row → a broken-limit-up rate of about 29.5%. Note that yesterday's recap wrote "104 limit-ups, 28.8% broken rate", slightly different from what the endpoint returned this time (pool definitions are tweaked over time); this brief uses the 103/43/1 measured this run throughout and tags the convention in the body.
• stock_zt_pool_zbgc_em and stock_zt_pool_dtgc_em were killed by a 120-second timeout on first call (background task exit 143); a second, separate run with a narrower scope succeeded. These two endpoints are far slower than the limit-up pool; suggest calling them separately rather than chaining them with other endpoints in the same script.
• yfinance was not used at all this run — A-share data all via akshare, US data all via Sina Finance articles. Following the lesson recorded last time, yfinance was skipped outright, wasting no further round of trial and error.
• In Yicai's "Evening Announcements Roundup for August 5", the "code" column of the table is entirely empty, with only company short names. For every stock cited from that source, the code has been cross-checked against the Tencent snapshot by short name (this is the rule written down in the 8/5 pre-market brief, and it was executed this time). Actual result: 21 names checked one by one, all 21 correct (江苏银行 Bank of Jiangsu 600919, 鲁北化工 Lubei Chemical 600727, 江丰电子 KFMI 300666, 鸿远电子 Hongyuan Electronics 603267, 菱电电控 Lingdian Electronic Control 688667, 金诚信 JCHX Mining 603979, 中科环保 Zhongke Environmental 301175, 高铁电气 Gaotie Electrification 688285, 美的集团 Midea Group 000333, 伯特利 Bethel Automotive 603596, 千里科技 Qianli Technology 601777, 中国天楹 China Tianying 000035, 新宝股份 Donlim 002705, 大叶股份 Daye 300879, 爱丽家居 Aile Home 603221, 顺络电子 Sunlord Electronics 002138, 荣昌生物 RemeGen 688331, 百济神州 BeiGene 688235, 恒瑞医药 Hengrui Pharma 600276, 有研新材 Grinm Advanced Materials 600206, 中农立华 Zhongnong Lihua 603970). Against the lesson of the 8/3 pre-market run where all 5 codes in an aggregated article were wrong, this step is now zero-cost and high-yield; suggest keeping it in the process permanently.
• RemeGen's "H1 net profit RMB 4.7 billion" was independently cross-verified: Yicai gave "estimated H1 net profit attributable to parent of RMB 4.7 billion, swinging from loss to profit", while searches returned largely 2025 loss-narrowing figures, which is easy to misread. It was finally reconciled by an independent yardstick — the AbbVie licensing upfront of USD 650 million ≈ RMB 4.68 billion — matching the order of magnitude, which confirmed that the RMB 4.7 billion is 2026H1 and comes mainly from licensing income rather than product ramp. Without that cross-check this would easily have been written up as "swing to profit driven by product ramp", a wrong attribution.
• FII (601138) hit the limit again today, with net main-force inflow of RMB 4.266 billion, the largest in the market; Sina's AI-generated piece attributed it to "AI-server revenue growth above 230% + RMB 1–2 billion buyback + RMB 19.5 billion dividend + H1 pre-announcement of +93%–103%". Yesterday's recap already checked each of these: the earnings pre-announcement was 7/9, the dividend execution 7/23, the buyback proposal 7/27 — all old news. This brief therefore does not list FII as a news-driven name, citing it only on the flow side as evidence of where money is routing. This is the second consecutive day of refusing that same post-hoc attribution; recorded for the record.
• The driver behind Victory Giant Technology's +17.12% is the same story: North American cloud vendors raising capex (Google raised to USD 195–205 billion, Amazon to USD 220 billion, Meta USD 130–145 billion) is all old news from late-July earnings season; the direct trigger on 8/5 was a brokerage report. This brief treats it as continuation flow, not as a new in-window catalyst.
• The fundamentals sub-agent produced two key data points on the two memory names that the main agent could never have computed itself, and they rewrote the rankings: ① Longsys earned RMB 3.862 billion net in 2026Q1, yet operating cash flow was −RMB 2.875 billion; inventory jumped in a single quarter from RMB 11.678 billion to RMB 17.961 billion (50% of total assets), funded by RMB 5.05 billion of new long-term borrowings in that quarter to stockpile, while asset impairment booked in the quarter was only RMB 47 million. The first draft listed Longsys as "watch closely" on the grounds that "P/E 28x is the cheapest" — a textbook cyclical valuation trap, where the low P/E appears precisely at the earnings peak. It has been downgraded to "watch only". ② Longsys' P/B of 12.67 sits at the 81st percentile since listing (historical median only 5.95). Low P/E + high P/B is the standard shape of a cycle top, and looking at P/E alone will never reveal it. This is the same class of thing as the 8/5 pre-market note on "unconstrained readings under the 20cm regime": a single indicator will lie to you; you must find a second indicator capable of falsifying it. ③ A related correction to a popular misconception: GigaDevice's 1.80% stake in ChangXin Memory is carried under "investments in other equity instruments", so fair-value changes go through OCI and do not affect net profit (the company clarified this on 2026-05-26). So the roughly RMB 2.05 billion of non-recurring gains in GigaDevice's H1 is not a ChangXin realization; the specific asset cannot be confirmed for now, and this has been flagged in the body rather than force-fitted into a narrative. → Conclusion: this sub-agent is worth waiting for. It knocked one name from "watch closely" down to "watch only" on the strength of the cash-flow statement and the P/B percentile — two things the main agent has no time to pull pre-market.
• The precious-metals sub-agent overturned the first draft's within-branch ranking; this was the single most valuable correction today. The first draft ranked Shandong Gold 2nd on "purity + valuation" and gave it "priority deep-dive", reasoning that "pure gold, no dilution, higher elasticity purity than Zijin". After measuring against output and segment data, the sub-agent came back with: ① Shandong Gold's theoretical elasticity is 4.22x (highest in the field), but its measured 2026Q1 realization ratio is only 0.66x (worst in the field) — with the average gold price up 62.2% year on year, net profit attributable to parent rose only 40.9%. Four leaks: hedging price locks, lower self-produced gold output, minority interests taking 28.5%, and RMB 1.91 billion of finance costs. "Self-produced gold revenue is 8.10x net profit attributable to parent" — that bullish argument is itself another way of saying "profit is thinly stretched", and in the first draft I mistook it for an elasticity advantage. ② Zhongjin Gold: theoretical 1.96x, measured 2.08x — the only one of the four to over-deliver — already promoted to No. 1. ③ Worse still, one premise I had written exactly backwards: the first draft said "gold has broken out to a new plateau at 4246 dollars" and "valuations have not caught up with gold's new plateau". In fact gold peaked in January 2026 (roughly USD 5626.8 on COMEX), and the current 4246.82 is still 24.5% below that peak; silver halved from RMB 27530/kg to 14988. This is a rebound after a deep correction, not a new high. Moreover, SGE gold average prices went Q1 1088.6 → Q2 990.4 → Q3-to-date 888.6, so even with the overnight surge the Q3 average remains materially below Q1/Q2 — a sequential decline in Q3 results is a high-probability event, and these companies' low P/Es are precisely what the sky-high Q1 gold price propped up (P/E in the 16–33% low percentile over 5 years, while P/B sits in the 85–88% high percentile — the classic divergence at a cycle top). → This one must be hardened into a rule: whenever writing "some price has hit a new high / new plateau", you must first pull at least 12 months of that instrument's price series and back-check; you cannot go straight from "it rose 4.1% today" to "it is at a high". Had we not checked today, we would have produced a report that was directionally right but with the risk narrative completely inverted — writing "risk of chasing a rebound" as "risk of chasing at highs", two situations that call for entirely different responses. ④ The sub-agent proactively flagged checking whether "US July payrolls +44,000" was non-farm payrolls or ADP (8/5 was a Wednesday, and NFP normally prints on Friday). This brief had independently verified it as the ADP "little non-farm" measure, and the body wording is correct; recording here that the check was in fact performed.
• The Grinm Advanced Materials / Yunnan Germanium sub-agent likewise corrected two things: ① Grinm's interim report was disclosed after the close on 08-05, so the 8/5 limit-up happened before disclosure — yesterday's limit-up cannot be explained as earnings-driven; today is the first time the market prices this interim report. The first draft had the causality backwards; fixed. ② "Revenue +50.35%" ≠ semiconductor materials ramping: the platinum-group business was 41.84% of 2025 revenue but contributed only 7.58% of gross profit (gross margin 1.63%, trading/tolling in nature), and in 2026H1 platinum-group growth of +65% outpaced thin film's +63% — which is why the net margin is only 3% and the gross margin is still falling year on year. In addition, operating cash flow was −RMB 950 million (−RMB 225 million in the prior-year period) and the cash-to-profit ratio −5.18, with ex-non-recurring profit up only 16.98% (six-tenths of the increase in net profit attributable to parent came from non-recurring items). The first draft only wrote "net margin of about 3%, growth quality questionable" without unpacking the cause — a case of "spotting the anomaly but not chasing it to the bottom". ③ The most important point on Yunnan Germanium: it is widely treated as a "germanium-price beneficiary", but the company's own 7/14 pre-announcement attributes results to compound semiconductors driven by high-speed optical modules, and never mentions the germanium price anywhere; moreover, when germanium prices surged in 2025 its material-grade germanium gross margin collapsed from 29.00% to 4.32%. The Pass conclusion did not change, but the reason given in the first draft ("germanium prices rising, distant mapping") was wrong; the real reason is "germanium-price elasticity has already been falsified by its own financials". A wrong reason will collapse under client questioning even when the conclusion is right. Also, per the sub-agent's suggestion, its H1 pre-announcement of +148%~+261% is disclosed alongside, to avoid selective presentation.
• One convention conflict that must be tagged: Grinm's "net main-force inflow of RMB 815 million" on 8/5 is the East Money convention (large + extra-large orders); THS's "real-time fund flow" (active-buy/active-sell convention) showed −RMB 331 million the same day, the opposite direction. Yunnan Germanium's two readings are likewise opposite (East Money +RMB 578 million vs THS −RMB 1.580 billion). This is not a data error, it is two definitions. Everywhere the body cites net main-force inflow the convention is tagged; otherwise a reader can falsify it on the spot with the other convention.

---

0. Today in One Sentence

The strongest catalyst is a hard data point, not a statement: US July ADP "little non-farm" payrolls added just 44,000 jobs (vs 70,000 expected, prior revised down to 95,000, a 6-month low), released around 20:15 Beijing time on 08-05; after the print, spot gold closed up 4.1% at USD 4246.82/oz and spot silver closed up 4.2% at USD 62.0483/oz. This is the only catalyst in the whole field that "falls inside the time window, directly rewrites a commodity price, and has that price flow straight through to the income statement".

The strongest branch is [precious metals], and it is a branch that was already running yesterday and got another shot of fuel overnight — the A-share precious-metals industry already ranked first across all industries on 08-05 at +7.72% (14 up, 0 down), and Zijin Mining (601899) had the second-largest net main-force inflow in the market at RMB 2.847 billion. Branch continuity and catalyst freshness hold simultaneously here, a rare combination in a pre-market list.

But first a premise that almost everyone gets wrong must be corrected — this is not a "gold at new highs" tape, it is a rebound after a deep correction. Gold peaked in January 2026 (SGE Au99.99 month-end RMB 1163.95/gram, corresponding to roughly USD 5626.8/oz on COMEX). The overnight 4246.82 is still about 24.5% below that peak; silver more so — SGE silver T+D fell from RMB 27530/kg in January to 14988 today, halved and not yet repaired. This brings a risk pointing in exactly the opposite direction, and this brief must put it up front: SGE gold average prices went Q1 1088.6 → Q2 990.4 → Q3-to-date just 888.6 (silver 21283 → 17944 → 14346). Even after the overnight surge of 8/5, the single-quarter Q3 average is still materially below both Q1 and Q2. In other words, a sequential decline in these companies' Q3 results is a high-probability event, and their current P/Es are precisely what the sky-high 2026Q1 gold price held down. Hence this brief's characterization of precious metals: "today's catalyst is real and not yet priced by the A-share market; but it appears as a rebound on a medium-term downward price curve, not the start of a new bull market." Those two sentences must be read together.

Within the precious-metals branch, this brief's ranking differs from mainstream market consensus, on the basis of segment data and output-derived measured realization ratios. Shandong Gold (600547), usually seen as "the purest gold elasticity", does have the highest theoretical gold-price elasticity in the field (self-produced gold revenue is 8.10x net profit attributable to parent, theoretical 4.22x), but in 2026Q1, with the average gold price up 62.2% year on year, net profit attributable to parent rose only 40.9% — a measured realization ratio of just 0.66x — hedging price locks, lower self-produced gold output, 28.5% minority interests and RMB 1.91 billion of finance costs, four leaks that ate the elasticity. Zhongjin Gold (600489), by contrast, is theoretical 1.96x and measured 2.08x — the only one of the four to "over-deliver". On that basis this brief promotes Zhongjin Gold to No. 1 and cuts Shandong Gold to No. 8. "Elasticity on the financial statements" and "elasticity reaching shareholders" are two different things.

The second branch, [memory], needs its character spelled out separately: delivered results are extremely strong, but the company's own next-quarter guidance came in below market expectations. SanDisk, after the US close on 08-05, reported FY26Q4 revenue of USD 8.965 billion (up 372% year on year, +51% quarter on quarter), net profit of USD 6.90 billion (versus a USD 23 million loss in the prior-year period) and a gross margin of 84.6% — the hardest financial delivery yet of the NAND price cycle. But what actually pushed the stock down was not those delivered results; it was the FY27Q1 guidance in the same release: revenue of USD 10.3–10.8 billion, below the sell-side consensus of about USD 11.16 billion. After the release, the stock fell more than 8% at one point in after-hours trading, then narrowed to about 4%. Two things must be separated here, and the first version of this brief conflated them: ① SanDisk's −5.40% in the regular session that day happened before the earnings release (the release came after the US close), and represents profit-taking after two days of large gains — it cannot be counted as the market's verdict on the report; ② the real reaction to the report is the after-hours move, and its trigger was guidance below consensus, not "results too good and already fully expected". To be fair, it should also be said: guidance of USD 10.3–10.8 billion against Q4's 8.965 billion is still +15%~+20% sequential growth — this is not a signal of a cycle top, it is "growth still there, but the slope falls short of the sell-side model". Combined with TrendForce having already cut its Q3 DRAM/NAND contract-price increase to +1318% / +1015% (Q2 was about 60%), the right grading for memory today is "delivered results confirmed, but forward guidance below expectations" — neither a simple positive nor a top.

On the flow side, 08-05 already said it plainly: the three largest net outflows were ChangXin Memory (688825) −RMB 2.542 billion, Eoptolink (300502) −RMB 1.991 billion and Innolight (300308) −RMB 1.627 billion, all optical-module / memory manufacturers; the largest net inflows were FII (601138) +RMB 4.266 billion, Zijin Mining (601899) +RMB 2.847 billion, Hengtong Optic-Electric (600487) +RMB 1.838 billion, GigaDevice (603986) +RMB 1.774 billion and Victory Giant Technology (300476) +RMB 1.523 billion. Money is leaving "optical modules with US export exposure" and moving into "non-ferrous + domestic equipment and materials + PCB/optical fiber + memory design".

Driver types: macro data (ADP) + commodity prices + overseas earnings delivery. Pre-market read: precious metals are likely to gap up with a base of real absorption; memory and compute hardware will most likely gap up too, but with the Nasdaq −0.83%, AMD −7.04% and SpaceX −13.61% overnight, the quality of absorption after the gap matters far more than the size of the gap. What most needs guarding against today is not missing out — it is chasing an already-priced-in gap up in memory and compute hardware.

---

1. News Overview

1.1 New Items Inside the Time Window (2026-08-05 15:00 → 08-06 07:00)

1.2 Impact-Level Definitions

• S: changes an industry trend, or creates direct, quantifiable order/price elasticity.
• A: clearly positive for one branch with multiple beneficiaries; the logic needs no long chain of inference.
• B: the positive holds but the chain is long and needs further verification, or it affects only a single company.
• C: mainly a sentiment stimulus, with weak durability.

1.3 Old-News Screening (outside the time window, not counted as new catalysts today)

This section exists because aggregated articles very easily stitch old news into "today's blockbuster". The publication date of each item below has been individually verified:

---

2. Strongest Positive Branches, in Descending Order

Supplementary notes on branches:

• Why precious metals rank above memory: the memory news is "bigger" (a report with +372% year-on-year growth), but the FY27Q1 guidance in SanDisk's same release (USD 10.3–10.8 billion) came in below the sell-side consensus (about 11.16 billion), and the stock fell more than 8% at one point after hours — the market has begun cutting its forward expectations for that chain. On the precious-metals side, the ADP print is new information that only appeared overnight, and gold rose 4.1% after that new information, which the A-share market has not yet priced. A pre-market ranking is about "the increment that has not been priced", not "the absolute magnitude of the news itself".
• Board distribution: representative names in the precious-metals branch are almost all on the SH/SZ main board (±10%), meaning limit-ups are judged at 10% and there is no investor threshold; the memory and semiconductor-equipment branches are heavily on the STAR Market (±20%, RMB 500,000 threshold) and ChiNext (±20%), where the same +12% is an ordinary up day on ChiNext but an impossible number on the main board. When reading any percentage move in this brief, look at the board column first.
• Whether there is a path to earnings delivery: precious metals yes (the gold price flows straight into selling prices, and 兴业银锡 Xingye Silver & Tin (000426) has already pre-announced H1 growth of 168.95%–197.86%); memory yes (but already delivered into the share price); AI compute hardware none inside the window (a Musk statement generates no revenue); semiconductor equipment and materials none inside the window (no new order announcements).

---

3. Master Ranking of Single-Stock Positive Strength

Sorted in descending order by stock. The board column determines the price limit: main board ±10%, ChiNext/STAR ±20%, BSE ±30%. All market data are as of the 2026-08-05 close.

---

4. Single-Stock Scoring Model

Total of 100 points, broken into the components below. This section gives the component breakdown for the top 8 on the list, so that readers can check whether the scoring was pulled off course by narrative.

Component breakdown for the top 8:

Notes on the scoring (two key deductions, both based on measured data from public financials, not subjective impressions):

1. GigaDevice's "expectation gap" gets only 4 points — not because its own results are poor (H1 pre-announcement of +1099%), but because the FY27Q1 guidance in SanDisk's same release, USD 10.3–10.8 billion, is below the sell-side consensus of about 11.16 billion, and the stock fell more than 8% at one point after hours. Overseas markets have already begun cutting their forward expectations for this chain, while the A-share market has not yet reflected it. (Note: that guidance is still +15%~+20% sequentially against Q4's 8.965 billion, so it is "slope below the model", not "cycle top".) This item pushed memory from "S grade" down to "A+ grade", and also kept GigaDevice out of the top three.
2. Shandong Gold's "risk deduction" of −6 is the heaviest on the list — its directness score of 20 (full marks) and earnings-elasticity score of 13 (theoretical 4.22x, highest in the field) are both fully deserved; the problem is transmission: in 2026Q1 the average gold price was +62.2% year on year while net profit attributable to parent rose only 40.9%, a measured realization ratio of 0.66x. Hedging price locks, lower self-produced gold output, 28.5% minority interests and RMB 1.91 billion of finance costs are four leaks, on top of a 62.57% debt ratio (highest of the four). This is the only name in this brief whose components are all respectable yet which is pushed down to No. 8 by transmission efficiency — and it is also the name in this whole brief most easily missed by anyone taken in by the "purity" narrative.

---

5. Top 10 Stock Deep-Dives

The ranking basis for precious metals in this section is the "measured realization ratio", not "purity + valuation". Each name is tagged with its theoretical elasticity (computed from the leverage of self-produced metal revenue over net profit attributable to parent) and its measured realization ratio (single-period year-on-year growth in net profit attributable to parent ÷ year-on-year growth in the average gold/silver price for the same period) — where the two disagree, the measured figure governs.

51 中金黄金 Zhongjin Gold60048987 pts · priority deep-dive · SH/SZ main board ±10%

• Related news: at about 20:15 on 08-05, US July ADP payrolls added 44,000 jobs (vs 70,000 expected, prior revised down to 95,000, a 6-month low) → overnight spot gold +4.1% at USD 4246.82/oz (Sina Finance, Commodities Wrap-Up²).
• Bullish logic: direct, affecting both revenue and profit. Mined-gold selling prices follow spot while costs are relatively rigid, so a higher gold price directly expands unit gross profit.
• Branch stage: day two of the build-up. The precious-metals industry ranked first among THS's 90 industries on 08-05 at +7.72% (14 up, 0 down), with 盛达资源 Shengda Resources (000603) at limit-up and 兴业银锡 Xingye Silver & Tin (000426) +9.07% — a limit-up ladder already exists, but there is no consecutive-limit-up anchor yet.
• Fundamentals check (based on segment data from public financials, cross-verified with output) — this is the entire reason it ranks first:
  • Its delivery of gold-price elasticity is the hardest of the four: theoretical elasticity 1.96x (computed from the leverage of self-produced gold revenue over net profit attributable to parent), 2026Q1 measured 2.08x — the only "over-delivery", with no sign of a hedging drag.
  • 2026Q1 revenue RMB 22.588 billion (+52.0%), net profit attributable to parent RMB 2.381 billion (+129.2%), ex-non-recurring +88.4%, gross margin up to 20.47%.
  • The mining segment contributes 82.6% of gross profit (only 25.6% of revenue, at a 53.9% gross margin); smelting is 89.6% of revenue but at a gross margin of only 3.84%, a volume business.
  • Effective tax rate of just 16.7% and a debt-to-asset ratio of 43.01%, both the lowest of the four. FY2025 net profit attributable to parent RMB 4.934 billion (+45.7%), operating cash flow RMB 8.375 billion, free cash flow RMB 5.343 billion.
  • 2025 output: mined gold 18.40 tonnes, mined copper 66,200 tonnes, electrolytic copper 417,000 tonnes.
• Industry position: the central-SOE gold platform under China National Gold Group, with solid resource endowment.
• Technical sentiment: closed at RMB 24.25, +8.94% (close to but not at the ±10% main-board limit), +11.55% over 5 days, +21.98% over 20 days. Net main-force inflow RMB 300 million, turnover rate 2.41% (low). Still −41.5% from its 52-week high.
• Risks: P/B 3.58 sits at the 88.1st percentile of the past 5 years; on "market cap per tonne of mined gold" it is RMB 6.39 billion per tonne, the most expensive of the pure-gold names (山东黄金 Shandong Gold (600547) is RMB 2.73 billion per tonne), which says its "delivery advantage" is already substantially priced; the 2026Q1 operating cash flow of −RMB 2.168 billion is seasonal (2025Q1 was also −RMB 1.960 billion, turning positive at RMB 8.375 billion for the full year) and is not a new deterioration signal; the split of gold versus copper revenue inside the mining segment is estimated in this brief as "output × realized price" rather than disclosed by the company, so if the actual copper share is higher, its gold elasticity should be revised down.
• Final verdict: priority deep-dive. In a branch where the catalyst is real but transmission efficiency varies by company, it is the only one that has delivered its theoretical elasticity intact into shareholders' hands.

52 紫金矿业 Zijin Mining60189986 pts · priority deep-dive · SH/SZ main board ±10%

• Related news: same as 5.1, plus overnight LME copper +0.3% at USD 14110.5/tonne (at highs), zinc +1.5%, tin +1.3%.
• Bullish logic: direct. It is the only A-share name with large-scale self-produced gold and self-produced copper at the same time, and gold and copper both rose overnight.
• Fundamentals check — here is one piece of market consensus that needs correcting:
  • "Zijin is a copper company" does not match the 2026 financials. By FY2025 mine-segment gross-profit structure: gold 51.4% > copper 43.4% > silver 2.5% > zinc 2.2%; in 2026Q1 the tilt toward gold goes further, to gold 54.4% / copper 39.6%. On mine gross profit, gold has already overtaken copper as the largest profit source, and high copper prices are an add-on rather than the main story.
  • Best earnings predictability in the field: all-metal theoretical elasticity 1.62x vs 2026Q1 measured 1.57x, an error of under 3%. The mine gold gross margin of 64.6% and mine copper gross margin of 61.0% computed in this brief match the company's annual-report disclosures exactly, which says the model is usable for this company.
  • FY2025: revenue RMB 349.079 billion (+15.0%), net profit attributable to parent RMB 51.777 billion (+61.6%), ROE 27.91% (highest of the four), operating cash flow RMB 75.430 billion, free cash flow RMB 44.448 billion, debt-to-asset ratio down to 51.37%. 2026Q1 net profit attributable to parent RMB 20.079 billion (+97.5%).
  • 2025 mined gold 90 tonnes, at a cost of RMB 275 per gram of gold (below 山东黄金 Shandong Gold (600547)'s RMB 388 per gram on a cost-of-sales basis) — low cost = a thick safety cushion, low elasticity.
  • Trailing P/E 14.70, the lowest on the whole list.
• Technical sentiment: closed at RMB 34.10, +6.16%, +6.07% over 5 days, +24.63% over 20 days. Net main-force inflow of RMB 2.847 billion ranks 2nd in the market, while the turnover rate is only 2.09% — a large inflow paired with extremely low turnover is chip accumulation rather than speculative money. Turnover value RMB 14.445 billion. It is −24.1% from its 52-week high (the shallowest drawdown of the four).
• Risks: on gold-price elasticity alone it is only 0.80x, the bluntest of the four — it is too large, so pure gold elasticity gets diluted; P/B 4.60 at the 85.5th percentile over 5 years, market cap at the 94.3rd percentile (an outright high).
• Final verdict: priority deep-dive. It is the choice for "certainty" rather than "elasticity": the best financial quality, the lowest valuation, the most predictable results, and the only twin gold-copper engine of the four. For pure gold elasticity choose 中金黄金 Zhongjin Gold (600489); for downside protection choose this one.

53 兴业银锡 Xingye Silver & Tin00042684 pts · priority deep-dive · SH/SZ main board ±10%

• Related news: overnight spot silver +4.2% at USD 62.0483/oz, LME tin +1.3% at USD 56718/tonne.
• Bullish logic: direct, and it is the only precious-metals name on this list that has already written the price elasticity into a formal earnings pre-announcement. The company disclosed its 2026 interim pre-announcement: net profit attributable to parent of RMB 2.14–2.37 billion, +168.95%~197.86% YoY; ex-non-recurring net profit attributable to parent of RMB 1.69–1.92 billion, +110.87%~139.57% YoY, with the company explicitly attributing this to higher selling prices for its core silver and tin mine products (Securities Times¹⁰, ifeng Finance¹¹).
• Branch stage: day two of the build-up. Silver's overnight gain (+4.2%) slightly exceeded gold's (+4.1%), and silver also carries industrial attributes.
• Fundamentals check: trailing P/E 24.16, free-float market cap RMB 64.4 billion. The pre-announcement has already walked the full chain from "prices rise" to "profit delivered", which is the watershed between a concept stock and a genuine beneficiary, and is why this brief does not need to rely on elasticity estimates for this name. Net main-force inflow RMB 366 million, turnover rate 5.48%.
• Industry position: a leading domestic silver-tin polymetallic miner, with core subsidiary Yinman Mining as the main profit source. Yinman's expansion project has broken ground, but it will take several years to reach design capacity and does not constitute near-term output growth.
• Technical sentiment: closed at RMB 36.30, +9.07% (approaching the main-board limit), +12.21% over 5 days, +9.24% over 20 days — the lowest position within the precious-metals branch.
• Risks: the company previously had a subsidiary work-safety accident (1 fatality), so follow-up regulatory action bears watching; as with 盛达资源 Shengda Resources (000603), the current silver price is still below its H1 realized average, so there is sequential pressure in H2 (this brief did not obtain its H1 realized average silver price, so this item cannot be confirmed for now).
• Final verdict: priority deep-dive. "Price catalyst inside the window + earnings pre-announcement already landed + smallest 20-day gain" stacked together make it the best risk-reward on this list. Its advantage is that you do not have to believe any elasticity model — the company has already reported the numbers.

54 兆易创新 GigaDevice60398680 pts · watch closely · SH/SZ main board ±10%

• Related news: SanDisk FY26Q4 revenue USD 8.965 billion (up 372% YoY, +51% QoQ), net profit USD 6.90 billion (versus a USD 23 million loss a year earlier), gross margin 84.6%, full-year revenue above USD 20 billion; the same release gave FY27Q1 guidance: revenue of USD 10.3–10.8 billion and non-GAAP EPS of USD 44.00–46.00, below the sell-side consensus of about USD 11.16 billion (Yahoo Finance³, Sina Finance, Overseas Headlines⁴).
• Bullish logic: indirect (industry trend), not a direct order. SanDisk's results prove the reality and magnitude of the NAND price cycle, and GigaDevice benefits through ASP uplift in its own product lines. SanDisk is not its customer and there is no supply relationship between them — this must be stated plainly, or it will be read as "an overseas leader's good news transmitting directly".
• Branch stage: build-up, but top signals have appeared. The memory chain rose as a group on 08-05 (兆易创新 GigaDevice (603986) +8.26%, 普冉股份 Puya Semiconductor (688766) +11.35%, 江波龙 Longsys (301308) +8.59%, 佰维存储 Biwin Storage (688525) +7.14%, 德明利 Demingli (001309) +6.06%, 北京君正 Ingenic (300223) +7.74%).
• Fundamentals check (verified against public financials): China's full-category memory design leader. By product in the 2025 annual report: memory chips RMB 6.566 billion (71.3% of revenue, 42.84% gross margin), MCU RMB 1.910 billion (20.8%), sensors 4.2%, analog 3.6% — about 71% of revenue is directly exposed to memory price increases, with the other 29% as a cushion.
  • Growth has been independently confirmed by its own pre-announcement (disclosed July 10, which is old news outside the window but still valid as a fundamentals check): 2026H1 revenue of about RMB 11.5 billion (up 177% YoY), net profit attributable to parent of about RMB 6.9 billion (+1099% YoY), ex-non-recurring about RMB 4.85 billion. The 2026Q1 gross margin has already risen from 40.22% in 2025 to 57.08% — this is pure price elasticity, not cost reduction.
  • Profit quality is the biggest plus: 2026Q1 operating cash flow +RMB 1.783 billion > net profit attributable to parent of RMB 1.461 billion (a 122% conversion rate); cash and equivalents RMB 14.694 billion; debt-to-asset ratio only 8.13% with almost no interest-bearing debt; inventory of RMB 3.401 billion is just 12.2% of total assets and up only 10.9% sequentially — no all-in stockpiling gamble.
  • One deduction that must be made: of the RMB 6.9 billion of H1 net profit attributable to parent, about RMB 2.05 billion is non-recurring (fair-value changes on securities investments), leaving RMB 4.85 billion ex-non-recurring. Note: the company clarified on 2026-05-26 that its 1.80% stake in 长鑫科技 ChangXin Memory (688825) is carried under "investments in other equity instruments", with fair-value changes going through OCI and not affecting net profit — so that RMB 2.05 billion is something else, and the specific asset cannot be confirmed for now; it must not be presented as "a ChangXin listing payoff".
  • Trailing P/E 94.09 (historical median about 96, currently at the 48.9th percentile), P/B 10.70, free-float market cap RMB 257.7 billion. Net main-force inflow of RMB 1.774 billion ranks 4th in the market, turnover value RMB 26.179 billion.
• Price position: this stock and 江波龙 Longsys (301308) both peaked on 2026-06-29 (GigaDevice's high was RMB 840.00); it closed at RMB 385.44 on 8/5, −54.1% from the peak; it hit limit-down (−10.00%) on August 3 and then rebounded on 8/4–8/5. −36.10% over 20 days — the current rebound starts from a halved level, and the "price increases are peaking" expectation has already been priced violently once.
• Expectation gap (this is the key deduction): negative. ① SanDisk's FY27Q1 guidance of USD 10.3–10.8 billion is below the sell-side consensus of about 11.16 billion, and the stock fell more than 8% at one point after hours before narrowing to about 4% — note that this is not "results too good and fully expected", it is forward guidance below the model; at the same time, that guidance is still +15%+20% sequentially against Q4's 8.965 billion, so it is not evidence of a cycle top. (Its −5.40% in that day's regular session came before the report and is profit-taking, not counted as a reaction to it.) ② TrendForce has already cut Q3 contract-price increases from Q2's roughly 60% to DRAM +1318% and NAND +10~15%, citing consumer-side affordability peaking — prices are still rising, but the slope drops by about two-thirds, and memory stocks price the acceleration of price increases, not their absolute level; ③ 长鑫科技 ChangXin Memory (688825) closed −1.27% on 08-05 with the largest net main-force outflow in the market at RMB 2.542 billion — the domestic DRAM manufacturer being sold most heavily on the very day the memory chain surged as a group is the most discordant signal inside the branch.
• Final verdict: watch closely, no priority deep-dive. This is the cleanest balance sheet in the memory chain, with downside protection and a cushion from 29% non-memory revenue. But "big news, share price already run" is memory's real state today. If it gaps up more than 5% today without a matching expansion in volume, this brief leans toward avoiding rather than chasing. What is really worth watching is not how much SanDisk earned, but that its next-quarter guidance came in below the sell-side consensus — which says overseas markets are cutting forward expectations for the entire NAND chain.

55 赤峰黄金 Chifeng Jilong Gold60098878 pts · watch closely · SH/SZ main board ±10%

• Related news: same as 5.1.
• Bullish logic: direct, pure gold elasticity, a growth-stage miner (developing mines both at home and abroad).
• Fundamentals check: trailing P/E 21.64, free-float market cap RMB 68 billion, total market cap RMB 77.6 billion. Net main-force inflow RMB 232 million. This brief did not run a segment-level elasticity check on it (the measured work covers only the four names 中金黄金 Zhongjin Gold, 紫金矿业 Zijin Mining, 山东黄金 Shandong Gold and 盛达资源 Shengda Resources), so its realization ratio cannot be confirmed for now — which is why it sits at No. 5 despite flawless logic.
• Technical sentiment: closed at RMB 40.84, +8.19%, +7.87% over 5 days, +30.40% over 20 days — the highest position within this branch.
• Final verdict: watch closely, not listed as a priority deep-dive. It is already up 30% over 20 days, and the risk-reward of chasing is clearly worse than 兴业银锡 Xingye Silver & Tin (000426) or 中金黄金 Zhongjin Gold (600489). If it gaps up and then falls back today, that would be a better observation point.

56 山金国际 Shanjin International00097577 pts · watch closely · SH/SZ main board ±10%

• Related news: same as 5.1.
• Fundamentals check: trailing P/E 18.29, tied with 中金黄金 Zhongjin Gold (600489) for the lowest tier among mainstream gold names; free-float market cap RMB 61.1 billion, total market cap RMB 67.2 billion. Net main-force inflow RMB 167 million, turnover rate only 2.97%. Likewise no segment-level elasticity check was run.
• Technical sentiment: closed at RMB 24.22, +7.36%, +7.88% over 5 days, +24.52% over 20 days.
• Final verdict: watch closely. Low valuation plus low turnover; its logic overlaps heavily with 中金黄金 Zhongjin Gold (600489) but it lacks the backing of a measured realization ratio. If you need to limit the number of positions inside the gold branch, Zhongjin Gold has the more complete evidence chain.

57 佰维存储 Biwin Storage68852576 pts · watch closely · STAR Market ±20% · RMB 500 · 000 threshold

• Related news: SanDisk's FY26Q4 report, same as 5.4.
• Bullish logic: indirect. Embedded memory plus in-house controllers, with an added edge-AI growth story.
• Fundamentals check: trailing P/E 27.83, the lowest tier among mainstream memory-chain names (兆易创新 GigaDevice (603986) 94.09, 普冉股份 Puya Semiconductor (688766) 126.31, 北京君正 Ingenic (300223) 105.90); free-float market cap RMB 109.9 billion. Net main-force inflow RMB 745 million.
• Technical sentiment plus one very important regime reading: closed at RMB 233.07, +7.14%. Biwin rose only 7.14% under the STAR Market's ±20% regime, and 江波龙 Longsys (301308) rose only 8.59% under ChiNext's ±20% — the two memory-chain names not constrained by ±10% both failed to reach 10% in a "free pricing" state; the same day 兆易创新 GigaDevice (603986) on the main board rose 8.26%, also short of the 10% cap. In other words, the memory chain's gains yesterday were not held down by the limit board — money simply only took them that far. This stands in sharp contrast with the semiconductor-equipment chain (正帆科技 Genetech (688596) sealed the 20cm limit outright, 中微公司 AMEC (688012) +12.67%, 拓荆科技 Piotech (688072) +12.14%).
• Risks: −41.18% over 20 days, a very large prior drawdown; the STAR Market's RMB 500,000 investor threshold means some readers cannot buy it.
• Final verdict: watch closely. Its valuation is the most reasonable in the memory chain, but the "only +7% under a 20cm regime" reading above says the memory branch's flow intensity yesterday was clearly weaker than the semiconductor-equipment branch's. That is the second reason this brief does not rank memory first.

58 山东黄金 Shandong Gold60054775 pts · watch closely · SH/SZ main board ±10%

This is the name where this brief diverges most from mainstream market opinion, and it is worth explaining separately why.

• Related news: same as 5.1.
• The mainstream view (which was also this brief's initial leaning): a pure gold name, no dilution from other metals, higher gold-price elasticity purity than 紫金矿业 Zijin Mining (601899), and therefore the first choice for buying the gold price.
• The measured data do not support that view:
  • Profit purity really is the highest of the four — in FY2025, self-produced gold revenue of RMB 38.396 billion was only 36.8% of revenue yet contributed 91.0% of gross profit (at a 50.6% gross margin); the other 61% of revenue is purchased gold, small gold bars and trading, at gross margins of just 0.80%–2.28%, making almost nothing. So any approach that measures 山东黄金 Shandong Gold (600547) by revenue growth is simply wrong.
  • Theoretical elasticity really is the highest in the field: self-produced gold revenue is 8.10x net profit attributable to parent, so in theory every 1% rise in the gold price drives net profit attributable to parent +4.22%.
  • But the measured realization ratio is only 0.66x. In 2026Q1 the SGE average gold price was up 62.2% year on year while net profit attributable to parent rose only 40.9%. The FY2025 measured realization ratio was also only 1.40x, far below the theoretical 4.22x.
  • Four leaks (all verifiable): ① gold hedging price locks (the company has announced gold hedging operations, and several financial media outlets ran pieces on "the shadow of hedging in a gold-price surge", noting Q1 net profit up 41% while the share price halved); ② self-produced gold output fell year on year in 2026Q1, worsening fixed-cost absorption; ③ minority interests take 28.5% of net profit (FY2025 minority interests RMB 1.888 billion / net profit RMB 6.627 billion); ④ finance costs of RMB 1.910 billion and a debt-to-asset ratio of 62.57%, the highest of the four.
  • 2025 mined gold 48.89 tonnes (+5.89%), at about RMB 388 per gram on a cost-of-sales basis, higher than 紫金矿业 Zijin Mining (601899)'s RMB 275 per gram — high cost means high elasticity, but only if that elasticity can be transmitted.
• Valuation: trailing P/E 25.87, the most expensive of the four; P/B 4.17 (38.1st percentile over 5 years, the only one of the four not in a high percentile). In a scenario where the gold price falls back to the Q3 average, its P/E jumps to 33.6x — high elasticity cuts both ways.
• Technical sentiment: closed at RMB 28.95, +6.91%, +5.93% over 5 days, +18.31% over 20 days. Net main-force inflow RMB 224 million, turnover rate 2.70%. −55.8% from its 52-week high, and −28.0% year to date in 2026.
• Final verdict: watch closely, downgraded from "priority deep-dive". It is not a bad company; it is the textbook case of "elasticity on the financial statements, not in shareholders' hands". If this leg of the gold rebound can persist, its theoretical elasticity is still the highest in the field — but an upgrade must wait until either the hedging exposure is disclosed or a quarter's realization ratio comes back above 1.5x. Until then, for the same bet on the gold price, 中金黄金 Zhongjin Gold (600489) has a far more complete evidence chain.

59 湖南黄金 Hunan Gold00215574 pts · watch closely · SH/SZ main board ±10%

• Related news: same as 5.1.
• Bullish logic: direct. Twin gold and antimony businesses.
• Fundamentals check: trailing P/E 22.02, free-float market cap RMB 38.6 billion. Net main-force inflow RMB 213 million, turnover rate 4.35%. No segment-level elasticity check was run, so the realization ratio cannot be confirmed for now.
• Technical sentiment: closed at RMB 24.68, +7.12%, +8.20% over 5 days, only +7.73% over 20 days — the second-lowest position in the precious-metals branch, behind only 兴业银锡 Xingye Silver & Tin (000426).
• Final verdict: watch closely. Its low position is its biggest advantage; antimony-price logic is not in sync with gold and is an extra variable (which may add elasticity or may be a drag).

510 盛达资源 Shengda Resources00060373 pts · watch closely (risk grade raised) · SH/SZ main board ±10%

• Related news: overnight spot silver +4.2% at USD 62.0483/oz.
• Bullish logic: direct. And its "silver purity" is backed by a disclosed figure, not a concept — the 2026 interim report states explicitly: mined silver metal is 70.38% of non-ferrous mining and beneficiation revenue and 62.14% of total revenue.
• Fundamentals check (the real part): 2026H1 revenue RMB 1.152 billion (+27.1%), net profit attributable to parent RMB 390 million (+456.5%), ex-non-recurring +417.1% (in line with net profit, with no padding from non-recurring items), gross margin 69.15%, the highest of the four; operating cash flow RMB 573 million / net profit attributable to parent RMB 390 million = 1.47x, a normal cash conversion. Silver concentrate gross margin 87.1%, lead concentrate 73.6%, zinc concentrate 70.6%. Non-ferrous mining and beneficiation revenue +58.9% with gross profit +98.7% — genuine volume-and-price resonance.
• Fundamentals check (the part that must be discounted, and the reason for raising the risk grade):
  1. Output barely grew: 2026H1 mined silver metal output rose only 8.82% year on year, with lead −13.67% and zinc −16.43%. The earnings surge is close to 100% the silver price, not capacity expansion. The moment the price softens, the earnings lose their support.
  2. The sharpest point — the current silver price is about 20.1% below its own H1 realized average: the SGE silver 2026H1 average was RMB 19556/kg, versus about RMB 15617/kg now (including the overnight +4.2%). If Q3 holds at this price, 2026H2 results will most likely decline materially sequentially, while the market is right now pricing the +456% of H1.
  3. The +456% measured elasticity is a low-base illusion: 2025H1 net profit attributable to parent was only RMB 70 million, and it is that tiny denominator that magnifies the number. On the new base of RMB 390 million per half year, forward static elasticity falls back to about 1.31x, which is nothing special among the four.
  4. Small scale: trailing net profit attributable to parent of RMB 852 million is one-seventh of 中金黄金 Zhongjin Gold (600489)'s and one seventy-second of 紫金矿业 Zijin Mining (601899)'s; FY2025 free cash flow was only RMB 168 million (capex ate 81% of operating cash flow). The recycled new-energy metals segment runs a gross margin of −6.7%, bleeding continuously.
• Valuation: trailing P/E 23.59, P/B 5.35 at the 87.6th percentile of the past 5 years (highest of the four).
• Technical sentiment: limit-up +10.00% to close at RMB 29.15, +16.60% over 5 days, +32.08% over 20 days. Net main-force inflow RMB 230 million, turnover rate 7.67%. Already a member of the limit-up ladder but not on consecutive limit-ups. Still −60.4% from its 52-week high.
• Final verdict: watch closely, but with the risk grade raised to the highest within the branch. The silver purity is real, the H1 results are real, and the cash conversion is real too — the problem is the volume-price mismatch: zero output growth while the silver price has already fallen back to 20% below its H1 realized average. It has the greatest H2 sequential-decline pressure of the four, and is simultaneously the one with the highest P/B percentile and a limit-up already in hand yesterday. Right logic, expensive position, and the direction still pointing down — not recommended as the first choice within the branch.

6. Pass List

---

7. Within-Branch Rankings

7.1 Precious Metals (first branch)

The ranking basis has been changed from "purity + valuation" to "the directness with which a gold-price rise is delivered into results" — that is, the efficiency and credibility of transmission from price increases into net profit attributable to parent. Where theoretical elasticity and the measured realization ratio disagree, the measured figure governs.

• The hardest single name: 兴业银锡 Xingye Silver & Tin (000426) — the only one to have walked the full "prices rise → profits rise" route with a formal earnings pre-announcement, relying on no elasticity model at all. If you must pick one within pure gold, it is 中金黄金 Zhongjin Gold (600489) — theoretical 1.96x, measured 2.08x, the only over-delivery of the four.
• Highest trading visibility: 紫金矿业 Zijin Mining (601899) — RMB 2.847 billion of net main-force inflow against a 2.09% turnover rate is the cleanest flow signal in the market.
• The most important conclusion in this branch: "theoretical elasticity" and "measured realization ratio" must be looked at separately. 山东黄金 Shandong Gold (600547) has theoretical elasticity of 4.22x (highest in the field) and measured only 0.66x (worst in the field), a gap of more than 6x; 中金黄金 Zhongjin Gold (600489) is theoretical 1.96x and measured 2.08x. If you look only at "whose self-produced gold revenue carries the greatest leverage over profit", you will pick precisely the one you should not.
• Who is following: 西部黄金 Western Gold (601069) and 恒邦股份 Hengbang (002237) (+4.41%) are back-row catch-up names.
• Who is a Pass: no Pass in this branch — every name has a real gold or silver business, and the differences are only in transmission efficiency and price position. That is corroborating evidence for ranking it as the first branch.
• The number-one risk shared across the branch (stated once more): SGE gold average prices went Q1 1088.6 → Q2 990.4 → Q3-to-date 888.6, and silver 21283 → 17944 → 14346. Even after the overnight surge, the single-quarter Q3 average is still materially below Q1/Q2 — a sequential decline in these companies' Q3 results is a high-probability event, and their current low P/Es are exactly what the sky-high Q1 gold price propped up. All four have P/E in the 16–33% low percentile over 5 years while P/B sits in the 38–88% mid-to-high percentile (three of them above 85%), and that divergence is a hallmark of a cycle top.

7.2 Memory (second branch)

• The hardest single name: 兆易创新 GigaDevice (603986) — flows, position and balance sheet are all there, and its 29% non-memory business (MCU / sensors / analog) provides the only downside cushion in the chain.
• The most important conclusion in this branch, ① "the leader is absent": 长鑫科技 ChangXin Memory (688825), the chain's genuine domestic manufacturer, closed lower on the day the sector surged and was net sold to the tune of RMB 2.542 billion. A branch with no anchor following along should have the durability of its limit-up ladder discounted.
• The most important conclusion in this branch, ② "江波龙 Longsys (301308) is not the best choice, even though it looks cheapest": Longsys has the highest product overlap with SanDisk (about 90% of revenue is NAND-related) and pre-announced H1 net profit attributable to parent of RMB 9.2–11.0 billion at a trailing P/E of only 28.39 — but its P/B is as high as 12.67, at the 81st percentile since listing (historical median only 5.95). Low P/E + high P/B is the classic feature of a cyclical top: the market prices it at an extremely low P/E precisely because it does not believe this earnings level is sustainable. More substantive is the balance sheet: inventory of RMB 17.961 billion is 50% of total assets, and in 2026Q1 it earned RMB 3.862 billion net while operating cash flow was −RMB 2.875 billion (all the profit settled into inventory), funded by RMB 5.05 billion of new long-term borrowings in the quarter to stockpile, with only RMB 47 million of asset impairment booked in Q1 — almost no buffer reserved against price declines. The same inventory is profit when prices rise and impairment when they fall; this played out once already in 2023 (a full-year loss of RMB 828 million). This brief therefore downgrades it from "watch closely" to watch only.
• Who is a Pass: 长鑫科技 ChangXin Memory (688825) (chip structure), 普冉股份 Puya Semiconductor (688766) and 北京君正 Ingenic (300223) (their logic is not on the NAND main line).

7.3 Semiconductor Equipment / Materials (fourth branch, but the highest trading intensity)

• The hardest single name: 北方华创 Naura (002371) — down only 8.49% over 20 days against an average 20%–30% decline across the whole chain, the only resilient one, which says its chip structure is the most stable.
• Highest trading visibility: 正帆科技 Genetech (688596) (20cm limit sealed, the only semiconductor name in the entire market to max out under a ±20% regime).
• The branch's fatal problem: there is no new catalyst whatsoever inside the window. Yesterday's rise was capital reallocation away from FCC exposure, and today is day two, which must run on new money rather than momentum. Hence everything except 北方华创 Naura (002371) is rated "watch only".

---

8. Opening Verification Signals for Today

8.1 Call-Auction Signals

• Precious metals: if 中金黄金 Zhongjin Gold (600489), 紫金矿业 Zijin Mining (601899) and 兴业银锡 Xingye Silver & Tin (000426) gap up 2%–4% with normal auction volume, that is healthy absorption; if they gap up more than 6% (near limit-up), the risk-reward of chasing today is very poor and it is better to wait for a pullback. A locked limit at the open or a gap up of 8% or more would mean sentiment is already overheated, and the risk shifts to gapping up and fading.
• Memory: a 3%–5% gap up is reasonable for 兆易创新 GigaDevice (603986) and 佰维存储 Biwin Storage (688525); a gap up beyond 8% (especially for Biwin in the 20cm bracket) calls for high alert — because SanDisk's next-quarter guidance is already below the sell-side consensus, so a large A-share gap up would mean marking prices up just as overseas forward expectations are being cut. 江波龙 Longsys (301308) is already rated "watch only" (inventory of RMB 17.961 billion at 50% of total assets, operating cash flow −RMB 2.875 billion); it is used here only as a reading on branch strength, not as a participation suggestion.
• Semiconductor equipment: if 正帆科技 Genetech (688596), 中微公司 AMEC (688012) and 拓荆科技 Piotech (688072) gap up and then fall into the red within 30 minutes, yesterday can essentially be judged a one-day wonder.

8.2 Sector Signals

• Confirmation signal: the precious-metals sector produces 3 or more fast limit-ups and 紫金矿业 Zijin Mining (601899) rises on expanding volume (the anchor follows) → the branch's continuation is established.
• Ladder signal: whether 盛达资源 Shengda Resources (000603) (limit-up yesterday) can post a second consecutive limit-up; if it does and new first-time limit-ups follow, the ladder is healthy.
• Divergence signal: if yesterday's limit-up names such as 盛达资源 Shengda Resources (000603), 有研新材 Grinm Advanced Materials (600206) and 云南锗业 Yunnan Germanium (002428) break their limits en masse, yesterday's 29.5% broken-limit rate is deteriorating further today.
• Contrarian signal: if 长鑫科技 ChangXin Memory (688825) keeps falling on expanding volume today, the memory branch's "no anchor" problem is confirmed, and the durability of everything else in the chain should be marked down across the board.

8.3 Single-Stock Signals

• Bullish confirmation: 中金黄金 Zhongjin Gold (600489) / 紫金矿业 Zijin Mining (601899) rise on volume without breaking their limits, with large-order net inflows continuing; 兴业银锡 Xingye Silver & Tin (000426) seals the limit quickly on the back of its pre-announcement.
• Realization-ratio verification (the single most valuable observation today): if 山东黄金 Shandong Gold (600547) is clearly weaker than 中金黄金 Zhongjin Gold (600489) and 山金国际 Shanjin International (000975) today, the "0.66x vs 2.08x" realization gap is confirmed by the market and this brief's downgrade holds; if 山东黄金 Shandong Gold (600547) instead leads, the market is trading "theoretical elasticity" rather than "measured delivery", which means this leg of the precious-metals move is sentiment rather than earnings, and its durability should be marked down across the board.
• Bearish warning: if 兆易创新 GigaDevice (603986) gaps up and fades while main-force flow flips from net inflow to net outflow, the memory branch can be judged finished for the day.
• Crowding-out risk: if new, lower-positioned names within the precious-metals branch post fast limit-ups (such as 湖南黄金 Hunan Gold (002155) or 西部黄金 Western Gold (601069)), they may divert money away from the already-extended 赤峰黄金 Chifeng Jilong Gold (600988) (+30.40% over 20 days) and 盛达资源 Shengda Resources (000603) (+32.08% over 20 days).

8.4 Risk Signals

• Gap up then plunge: with the Nasdaq −0.83%, AMD −7.04% and SpaceX −13.61% overnight, the probability that A-share compute hardware (胜宏科技 Victory Giant Technology (300476), 生益科技 Shengyi Technology (600183), 沪电股份 WUS Printed Circuit (002463), 深南电路 Shennan Circuits (002916), 工业富联 FII (601138)) gaps up and then plunges is not low. This is the category most in need of guarding against today.
• A lone limit-up: if only 1–2 precious-metals names hit the limit while the sector as a whole weakens, this is money testing the water rather than a trend.
• The anchor does not follow: if 紫金矿业 Zijin Mining (601899) closes lower today, every small name in the precious-metals branch must be re-rated.
• The specific risk in 盛达资源 Shengda Resources (000603): it already sealed the limit yesterday, is +32.08% over 20 days and sits at the 87.6th P/B percentile over 5 years, while the current silver price is about 20.1% below its 2026H1 realized average and output rose only 8.82% year on year. If it gaps up, pushes higher and then falls back today, that is the first ebb signal to appear within the branch.
• Yesterday's strong themes ebbing: semiconductor equipment (正帆科技 Genetech (688596), 中微公司 AMEC (688012), 拓荆科技 Piotech (688072)) and PCB (胜宏科技 Victory Giant Technology (300476), 生益科技 Shengyi Technology (600183)) have both risen for consecutive sessions with no new catalyst inside the window, and are the two most likely to ebb today.
• Index risk: turnover across the two exchanges was RMB 2.68 trillion on 08-05, with a broken-limit rate of about 29.5% (103 limit-ups / 43 broken / 1 limit-down). A broken-limit rate approaching three-tenths says yesterday's money-making effect was not as good as the index gain suggests.

---

9. Final Recommendations

9.1 The 5 Stocks Most Worth Watching Today

Where this brief differs from the mainstream ranking: 山东黄金 Shandong Gold (600547) is usually listed as the first choice for buying the gold price. But the measured data show that in 2026Q1, with the average gold price up 62.2% year on year, its net profit attributable to parent rose only 40.9% (a realization ratio of 0.66x), as hedging price locks, lower self-produced gold output, 28.5% minority interests and RMB 1.91 billion of finance costs ate up what is theoretically the highest 4.22x elasticity in the field. This brief accordingly places it 8th with a "watch closely" rating, and puts 中金黄金 Zhongjin Gold (600489) at No. 1. Likewise, 盛达资源 Shengda Resources (000603) drops out of the top five because "output up only 8.82% + the silver price already 20.1% below its H1 realized average". The top 3 all still fall in the precious-metals branch and are all on the SH/SZ main board (±10%, no investor threshold); Nos. 4 and 5 are memory names, kept to preserve exposure to the second branch, but with their negative expectation gap explicitly flagged.

9.2 The 3 Strongest Branches Today

9.3 Directions Not Worth Chasing Today

1. AI compute hardware (PCB/CCL/servers) — 胜宏科技 Victory Giant Technology (300476), 生益科技 Shengyi Technology (600183), 工业富联 FII (601138), 沪电股份 WUS Printed Circuit (002463), 深南电路 Shennan Circuits (002916). Reason: the only new item inside the window is Musk's statement that SpaceX will "use only NVIDIA processors", which generates no revenue at all; meanwhile the same night saw AMD −7.04%, SpaceX −13.61% and the Nasdaq −0.83%. These names already surged on 08-05 (胜宏科技 Victory Giant Technology (300476) +17.12%, 生益科技 Shengyi Technology (600183) +9.41%, 工业富联 FII (601138) limit-up), so the probability of gapping up and plunging today is higher than that of continuation.
2. The extended names in semiconductor equipment/materials — 正帆科技 Genetech (688596), 中微公司 AMEC (688012), 拓荆科技 Piotech (688072), 有研新材 Grinm Advanced Materials (600206), 微导纳米 LEADMICRO (688147). Reason: yesterday's gains were huge but there is zero new catalyst inside the window, and the driver was a one-off capital reallocation away from FCC exposure. Day two must run on new money, otherwise it is a one-day wonder. (北方华创 Naura (002371) is kept under separate observation for its resilience, down only 8.49% over 20 days.)
3. Optical modules — 中际旭创 Innolight (300308), 新易盛 Eoptolink (300502), 光迅科技 Accelink (002281). Reason: the FCC draft is not finalized and the companies' response is merely "no document has been issued, no comment for now", so the uncertainty is unresolved rather than resolved; the three were net sold by main-force money to the tune of about RMB 4.35 billion yesterday. Do not treat "it has fallen a lot" as a positive.
4. High-flying pure themes — 云南锗业 Yunnan Germanium (002428), 爱丽家居 Aile Home (603221), 光智科技 Guangzhi Technology (300489). Reason: valuation disconnected from results + no news inside the window + acceleration at highs. Yunnan Germanium needs a separate note: it is often treated as a "germanium-price beneficiary", but that chain has already been falsified by its own financial data — in 2025, the year germanium prices surged, its material-grade germanium gross margin collapsed from 29.00% to 4.32%, with full-year revenue +38.89% while net profit attributable to parent was −62.06% and ex-non-recurring showed a RMB 6.81 million loss. The wording of the company's July 14 earnings pre-announcement attributes results to "compound semiconductor materials seeing both volume and price rise on high-speed optical-module demand", and never mentions the germanium price anywhere; and compound semiconductors are currently only 12.93% of its revenue. A market cap of RMB 54 billion against trailing net profit attributable to parent of RMB 19.05 million gives a trailing P/E of 2835x and P/B of 36.93x.
5. "Unitree concept stocks" — Reason: Unitree Robotics is not yet listed, and the minority-holder / supply-chain lists circulating in the market mostly lack support from company announcements. The offer price is only announced on 8/6, and the correct way to participate is the 8/10 subscription.

9.4 The Final One-Sentence Judgment

Today is a day of "the loudest news is not the branch you should buy, and the most on-the-nose name is not the stock you should buy". SanDisk's report with +372% year-on-year growth is the loudest, but the next-quarter guidance in that same release (USD 10.3–10.8 billion) came in below the sell-side consensus (about 11.16 billion) and the stock fell more than 8% at one point after hours — overseas markets have already begun cutting forward expectations for this chain. The increment that genuinely falls inside the time window and has not yet been priced by the A-share market is the +4.1% in gold and +4.2% in silver brought by ADP at just 44,000 — but it must be said at the same time that this is a rebound after a 24.5% decline from the January 2026 peak, not a new high, and that the SGE gold Q3 average (888.6) is still materially below Q1 (1088.6) and Q2 (990.4), so a sequential decline in these companies' Q3 results is a high-probability event. On that premise, this brief's landing point is not something as coarse as "buy precious metals", but "buy the one that can actually turn the gold price into net profit attributable to parent": Zhongjin Gold has theoretical elasticity of only 1.96x yet measured delivery of 2.08x, while Shandong Gold, which looks purest with theoretical elasticity as high as 4.22x, measured only 0.66x — four leaks (hedging price locks, lower self-produced gold output, 28.5% minority interests, RMB 1.91 billion of finance costs) ate its elasticity inside the financial statements, and it never reached shareholders. That is also the thing most worth remembering today: in a branch with an extremely hard price catalyst, the real disagreement is not "whether to buy", but "who can turn price into profit".

⚠️ 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 timeliness gaps or errors in supply-chain mapping; it must not be used directly as a basis for trading.