A-Share Pre-Market Brief | 2026-08-20, Thursday

0. One-Sentence Summary

The strongest overnight catalyst is precious metals, but it is a "price event," not an "earnings inflection" — that distinction is the pivot of this entire edition. In the 8/19 US session GLD closed at 413.84, +3.84%, breaking above the whole 8/10–8/18 range (398.55–405.49); at a read time of 8/20 07:01 Beijing time, COMEX gold was 4,579.50 (+0.75%) and silver 67.095 (+1.93%). Yet the Shanghai Gold Exchange's Au99.99 closed 8/19 at RMB 945.22/gram, actually lower than 955.16 on 8/18, having been pinned inside the 940–956 range all through August — converting from $4,518/oz at the prevailing exchange rate, the overnight breakout corresponds to an RMB gold price of about 985 RMB/gram, a gap of roughly +4.2% above the 8/19 close.

But "the A-share market has not priced it" needs to be stated precisely: global gold miners already completed their re-pricing on 8/19 — GDX (the global gold-miner ETF) was +9.42% on the day and SLV +4.47%. The A-share market is not "unpriced by anyone"; it is one trading day behind global miners. That has two direct consequences: ① today's opening gap could be far larger than +4.2%; ② therefore the risk criterion set in §8-1 of this edition, "an opening gap of ≥ +6% puts you in chase territory," has a high probability of being triggered today — treat it as the base case, not as low-probability insurance.

But it is equally necessary to say what this is not: the year-on-year growth rate of the RMB gold price has already run its course. SGE average price YoY: Q1 +62.2% → Q2 +28.5% → Q3 through 8/19 +13.9%; and the Q2 single-quarter profit growth of the five gold miners has fallen in lockstep to +13%~+23% (山金国际 (Shandong Gold Mining International)'s ex-non-recurring is only +4.5%), matching the convergence in price growth precisely. Based on the 2025Q4 average price of RMB 948.2/gram, even if gold holds above 985, 2026Q4 YoY would be only +3.9%. So what to do today is trade the gap, not hold on an "earnings re-rating."

The second line is policy, and this edition classifies it as a sentiment line rather than a logic line. A State Council decree amends the Regulations on the Administration of Housing Provident Funds (effective 2026-09-20), newly allowing withdrawals for home renovation and property-management fees; of the 36 limit-ups market-wide on 8/19, 9 sat on this chain (25%: 6 in home furnishing / kitchen & bath / renovation materials + 3 in real-estate services), whereas over the prior three sessions (8/14, 8/17, 8/18) the chain was essentially flat (志邦家居 (Zbom Home Collection) cumulative +0.85%, 日出东方 (Sunrise New Energy) +0.15%, 万和电气 (Vanward Electric) −6.13%; only 罗普斯金 (Ropeok) rose 3.23% first on 8/18). But three hard pieces of evidence show it currently affects sentiment only: ① the text of the regulation says renovation of one's own home is allowed "up to a certain limit," and every locality's ceiling and eligibility rules are still unpublished, so the contribution to 2026 financials is zero; ② provident-fund withdrawal converts an employee's own accumulated savings into cash earlier — it is a liquidity release, not a subsidy; ③ the two names with the highest policy exposure did not hit limit-up (华帝 (Vatti)'s project channel is only 3.00%, 老板 (Robam) 10.62%), while limit-up name 日出东方 has 25.13% overseas revenue and only 11.26% from kitchen appliances, and 万和电气 has 41.66% exports — the ranking by benefit and the ranking by gain are almost exactly inverted, and the white-goods sector itself fell 1.59% on the day.

Style judgment precedes stock screening (the order demanded by yesterday's recap): overnight was "rate pressure easing + AI hardware falling on its own." S&P +0.21%, Nasdaq +0.16% closed green, the 30-year Treasury yield fell back to 5.189% and the 10-year to 4.641%, software IGV +0.83% while SMH −1.55% and the SOX −2.12% — yields falling, software up, hardware down, so "rates killing growth" does not hold tonight; what got re-priced is the deliverability of AI capex commitments. China ADRs strengthened in parallel (China Golden Dragon +1.51%, KWEB +1.64%, FXI +1.77%).

Pre-market state assessment (an observation priority, not trading advice): yesterday's high-ladder limit-up names and AI hardware are not on this edition's watch sequence; the reason to watch precious metals is "the gap event," not "an earnings re-rating"; the objects to watch on the policy chain are the names that did not hit limit-up, not the ones that did.

---

1. News Overview

Scan window: 2026-08-19 15:00 (A-share close) → 2026-08-20 07:30. Release times for every item have been checked.

Four "looks positive, actually isn't" corrections in this section

If these four are not called out, they lead directly to trading the wrong way today.

1. 恒瑞医药's interim report was misreported wholesale. Multiple outlets gave "H1 revenue RMB 15.761 billion +15.88%, attributable net RMB 4.450 billion +29.67%" — those are the 2025 interim figures (disclosed 2025-08-21). The actual 2026 figures are revenue RMB 15.456 billion (−1.94%) and attributable net RMB 4.465 billion (+0.34%). Check: 154.56 ÷ 157.61 − 1 = −1.93%, consistent with the API's YoY. 恒瑞 discloses its interim report every year on 8/20–8/21, so the same column at the same moment is extremely easy to mix up.
2. 恒瑞 buyback: the "RMB 1–2 billion" in the media roundups is the old plan, but the company also rolled out a new plan this morning — these two things must be stated separately. (Everything below is taken from the original exchange filings.)

• Old plan (interim 2026-124, implementation result): first disclosed 2025-08-21, implementation period 2025-08-20 to 2026-08-19, expired on 8/19. Actually repurchased 16.5910 million shares (0.25% of total share capital) for RMB 1.004 billion at an average price of about RMB 60.51/share, within a range of RMB 45.30–70.00 — only reaching the neighborhood of the lower bound of the RMB 1–2 billion range.
• New plan (interim 2026-123, first disclosure date 2026-08-20): approved by the board on 8/19, another RMB 1–2 billion buyback, self-owned funds, for the A-share employee share-ownership plan, price cap RMB 81.78/share, 12-month term, expected to repurchase 0.18%–0.37% of share capital, no shareholder-meeting approval required. The 2026 A-Share Employee Share-Ownership Plan (Draft) was also released the same day. The announcement further states: the controlling shareholder, actual controller, directors, senior management and shareholders above 5% have no plans to reduce holdings over the next 3 and 6 months.
• Net judgment: buyback bid support will not be interrupted, and that is the genuine positive this morning; the RMB 81.78 price cap is 55% above the 8/19 close of RMB 52.66, so it is not an execution constraint. But note that the old plan only executed to the lower bound, and whether the new one exceeds RMB 1 billion remains to be seen. This item overturns this edition's first-draft judgment that "buyback bid support disappears from 8/20"; the original filing governs.

3. 山金国际's "H1 attributable net +51.43%" is neither new information nor a beat. The company already issued an earnings pre-announcement on 2026-07-14 (attributable net RMB 2.35–2.48 billion), and the actual RMB 2.416 billion lands in the upper-middle of the range, with no surprise. The genuinely new information is the single-quarter figure hidden by the cumulative basis: Q2 attributable net only +13.4%, ex-non-recurring only +4.5%, gross profit only +8.9% — while the RMB gold price was still +28.5% YoY in that quarter.
4. 赛意信息's "RMB 20 billion cross-sector compute push" is a media framing. The original announcement (2026-024, 5/14) says the total transaction value of finance-lease business is expected not to exceed RMB 20 billion — that is a financing-channel authorization, not a capex commitment. Also: the first disclosure date of the RMB 5.079 billion server procurement is 2026-07-13, not May.

---

2. Strongest Catalyst Branches, Descending

Branch notes

① Gold/Silver — why A and not S. My first-pass judgment last night was S; after fundamental verification it must come down one notch, for three verifiable reasons:

• The RMB gold price has decoupled from the USD gold price. SGE Au99.99 closed 8/19 at RMB 945.22/gram, down from 8/18, and was pinned in 940.72–955.75 all of August. The RMB has appreciated 5.17% against the dollar over the past year, systematically weakening transmission from the USD gold price to A-share miners.
• The YoY growth has already run its course. SGE average price YoY Q1 +62.2% → Q2 +28.5% → Q3 to date +13.9%; 2025Q4 average was 948.2, so even if gold holds above 985, 2026Q4 YoY is only +3.9%.
• The five names' Q2 single-quarter profit growth already reflects this: 山金国际 attributable +13.4% / ex-non-recurring +4.5%, 中金黄金 (China Gold International / Zhongjin Gold) +18.8%/+13.1%, 赤峰黄金 +20.5%/+22.8%, 湖南黄金 +19.6% (the last three derived from the midpoints of interim pre-announcements). The +80%~+129% growth of that Q1 batch cannot be extrapolated.

A fourth point (added before finalization, which weakens the "unpriced" claim): global gold miners already finished their move on 8/19 — GDX +9.42%, SLV +4.47%. The A-share market has not failed to price it; it is one trading day behind. So today is most likely a gap-up, not "buying an unreacted asset at a low."

Fifth point: this is not a trend high; it is a two-month high inside a rebound. GLD is still −16.5% from its one-year high (495.90,2026-01-29), SLV −43.2% from its high, GDX −16.0%. I therefore no longer use "gold is breaking out" to exempt the "low PE + high PB percentile" cycle-top warning — that warning still applies to this branch (PB 10-year percentiles: 中金 94.3%, 湖南 92.8%, 山金 91.3%). The silver line especially deserves care: it stands on a commodity that is 43% below its high.

So why still rank it first? Because it is the only branch with a hard catalyst today, and the A-share market genuinely has not moved. Short-term or medium-term: clearly short-term, and the execution window depends on the size of the opening gap. Earnings-delivery path: present, but decaying. One-day-wonder risk: high. Board distribution: all Shanghai/Shenzhen main board (±10%), so executability for readers is high.

② The provident-fund chain — downgraded from "logic line" to "sentiment line." Why: this positive does not affect any company's revenue within 2026. The regulation only takes effect 9/20, and every locality's withdrawal ceiling, frequency and documentation requirements (renovation contract or invoice) are still unpublished — until the rules land, any elasticity estimate is missing its most critical parameter, so this edition makes no estimate and cites no sell-side estimate either. Mechanically it lets employees monetize their own already-accumulated savings early — a liquidity release, not a subsidy — and only helps households that want to renovate but are cash-flow constrained. The demand side still points the other way: 2025 kitchen & bath retail sales −8.5%, finished-decoration units −22.1% (AVC, as cited in 老板电器's 2025 annual report), and all six companies had negative revenue growth in 2026Q1.

The most valuable information in this branch is not "who hit limit-up" but "who should have benefited and didn't":

Benefit and gain are almost perfectly inverted. Add that the white-goods sector was −1.59% on the day (Gree/Midea/Haier stayed out entirely), and the only possible conclusion is: what was bought on 8/19 was not "who benefits most" but "who can be pushed to limit-up." This edition therefore puts its recommendation slots on the two that did not hit limit-up.

③ Coke must be downgraded — this is a correction to yesterday's recap conclusion. Yesterday's recap listed coke as "tomorrow's main line," on the basis that "the price hike takes effect at midnight on 8/20." Three searches in this edition failed to confirm the existence of any "midnight 8/20 execution" event; the only verifiable facts are the 8/17–8/18 first-round push of RMB 50–55/tonne initiated by coking plants (Mysteel, southwest market), and a push by coking plants ≠ acceptance by steel mills. A harder point: 宝泰隆 PE −26.82, 美锦能源 (Meijin Energy) −12.73, 陕西黑猫 (Shaanxi Black Cat) −6.05, 大有能源 −11.04 — all four limit-up names are loss-making, so this branch currently has no path to earnings delivery at all. Combined with 宝泰隆's LHB share of turnover at 83.5% (heavily controlled), it is rated B, watch only.

---

3. Master Ranking of Single-Stock Catalyst Strength (descending by stock)

⚠️ Valuation-basis note (the single most important technical correction in this edition): East Money's default PE field is "forward PE = market cap ÷ latest single-quarter net profit × 4," not TTM. It annualizes 2026Q1 — a quarter in which the RMB gold price was +62.2% YoY — or the seasonally strongest Q1 for appliance makers, and systematically understates the true valuation of cyclicals and seasonal businesses. The table below gives both forward PE and rolling TTM PE; TTM governs.

---

4. Single-Stock Scoring Model (100 points)

Example (赤峰黄金 600988, total 82): authority 15 + directness 20 (gold price flows straight into gross profit) + earnings elasticity 10 (Q2 ex-non-recurring +22.8%, fastest in the group, but already converged from Q1's +101.6%) + fundamentals 14 (mined gold ≈100% of gross profit, ROE 27.14%, net cash, goodwill only RMB 16 million) + expectation gap 6 (already +10.50% over 4 sessions, partly priced) + durability 7 + trading characteristics 9 − risk −9 (PB 3Y percentile 91.8% + RMB 424 million Laos rare-earth impairment exposure) = 82.

Example (万和电气 002543, total 52): authority 12 + directness 8 + earnings elasticity 2 (H1 pre-announcement −76.56%~−84.16%) + fundamentals 5 (exports 41.66%, unrelated to the policy) + expectation gap 3 + durability 5 + trading characteristics 8 (seal ratio 0.479, 0 broken limits) − risk −15 (forward PE 17.68 is badly distorted, true TTM 48.69 and about to turn negative; sealed at 13:51, last in the whole market) = 52.

---

5. Top-10 Single-Stock Deep Dives

1赤峰黄金600988SH main board ±10% | Total 82 | Priority deep-dive · Chifeng Jilong Gold

• Related news: measured at 8/20 07:01, COMEX gold 4,579.50 (+0.75%); GLD +3.84% on 8/19, breaking the month's range. SGE closed 945.22 on 8/19; the overnight breakout implies about RMB 985/gram, a gap of roughly +4.2%.
• Catalyst logic: direct, hits the P&L.
• Stage of the industry branch: the branch overall is at "price gapped, A-shares have not reacted"; but the stock itself has risen 4 sessions in a row (+10.50%), putting it on the right-hand side within the branch.
• Fundamental verification: the purest-structured gold miner in the group — mined gold RMB 11.339 billion = 89.7% of revenue, gross profit RMB 6.635 billion ≈ the company's entire gross profit (the remaining segments are collectively loss-making). ROE 27.14%, highest in the group; cash RMB 6.331 billion vs. interest-bearing debt RMB 465 million, i.e. net cash; goodwill only RMB 16 million. Interim pre-announcement (7/15): H1 attributable net RMB 1.70–1.78 billion (+54%~+61%), with the text stating "the average selling price of gold rose about 43% year on year." This implies Q2 attributable +20.5% and ex-non-recurring +22.8%, the fastest of the group.
• Industry position: gold miner, with 71.1% of revenue from overseas.
• Technicals / sentiment: SH main board ±10%. 8/19 +1.21%, main-force +RMB 170 million.
• Risk (must be stated up front): PB 5.93,3-year percentile 91.8%; on 2026-08-08 it announced suspension of the Laos Mengkang rare-earth project, whose operating entity lost RMB 107 million in 2025 — the suspension removes a drag of about RMB 100 million a year (a positive) but leaves an asset-impairment exposure of about RMB 424 million, and whether a provision will be taken is still to be verified. One more thing to say plainly: this edition obtained no valuation-percentile data for the stock beyond PB history, and GLD is still −16.5% from its one-year high — so the "low PE + high PB" cycle-top warning is not exempted for this name either; it is merely held down for now by "gold is rising today."
• Final judgment: it takes the top watch priority in this edition because it has "the fewest discounts," not "the greatest elasticity." Its gold-price elasticity (+1% gold → attributable net +1.84%) ranks only third, but it has no hedging, no minority-interest leakage, no goodwill and no cash-flow blemish.

2盛达资源000603SZ main board ±10% | Total 74 | Watch closely · Shengda Resources

• Related news: measured at 8/20 07:01, silver 67.095, +1.93%, a stronger gain than gold's +0.75%; US-listed SLV +4.47% on 8/19.
• Catalyst logic: direct.
• Why this name and not 兴业银锡 for the silver slot (a reversal made before finalization): the first draft put 兴业银锡 (000426) here, on the grounds that "8/19 −7.34%, a bigger drop than the gold names = mispriced." During the mandatory single-name announcement re-sweep of every recommended stock before finalization, I found that 兴业银锡's own 8/19 announcement (2026-72) proposes acquiring control of 威领股份 (Weiling, 000631) for up to RMB 1.8 billion plus a partial tender offer — Weiling's attributable net assets at end-2025 were −RMB 32.05 million (negative), it lost money three years running, and it has already triggered delisting-risk warning. That is the reason for the day's −7.34% and main-force −RMB 263 million: not a mispricing, but the market pricing an RMB 1.8 billion investment into an ST company. 兴业银锡 has been moved to Pass.
• Fundamental verification: the highest silver purity in the group — in 2026H1, silver concentrate (gold-bearing) was 34.55% of revenue (gross margin 87.08%), lead concentrate (silver-bearing) 37.21% (73.59%), zinc concentrate (silver-bearing) 12.98% (70.64%), all three concentrates contain silver, together 84.7%. H1 revenue RMB 1.152 billion (+27.1%), attributable net RMB 390 million (+456.5%), gross margin 69.1%, ROE 12.1%.
• Announcement re-sweep result (clean): on 8/7 subsidiary 银都矿业 (Yindu Mining) received approval for a 900,000 tonne/year mining upgrade project, and on 8/4 subsidiary 鸿林矿业 (Honglin Mining) obtained a safety production permit for its tailings pond and formally began operations — both are positives on the capacity side; an abnormal share-trading fluctuation announcement was issued on 8/8.
• Expectation gap: medium to small. H1 results were already disclosed on 8/1, three-week-old news; the only new variable today is the silver price.
• Technicals / sentiment: 8/19 −5.38%, main-force −RMB 134 million. TTM PE 25.34, not cheap.
• Final judgment: it is a "purity for valuation" trade — more expensive than 兴业银锡, but with a clean silver-price transmission path and no off-balance-sheet risk like an RMB 1.8 billion investment in an ST company.

3华帝股份002035SZ main board ±10% | Total 74 | Watch closely · Vatti

• Related news: State Council decree amending the provident-fund regulations, effective 9/20 (Xinhua²).
• Catalyst logic: indirect, an industry-trend positive, not an order. Graded under iron rule 2, this is "marginal improvement on the demand side."
• Why it ranks ahead of every limit-up name: it has the highest policy exposure in the chain — the 2025 annual report's sales-model split shows offline 48.68% + online 33.82% = domestic retail 82.50%, project channel only 3.00% (lowest in the group), overseas 12.70%. The new provident-fund rules act on the C-end renovation of existing homes, so the lower the project-channel share and the higher the domestic retail share, the more direct the benefit.
• Fundamental verification: 2026Q1 revenue RMB 1.186 billion (−5.86%), attributable net RMB 101.1 million (−4.88%), ex-non-recurring RMB 96.7 million (−6.86%) — the smallest decline and the steadiest of the six. Gross margin 44.79%; cash RMB 1.254 billion; gearing 42.55%. TTM PE 14.21 (forward PE 10.40 is somewhat distorted because 25Q4 was a single-quarter loss of RMB 55 million), PB 1.12, lowest in the chain.
• Technicals / sentiment: 8/19 only +3.12%, no limit-up — no chase risk, but also a sign that money has not chosen it yet.
• Final judgment: "the one that should benefit" on the policy chain. But it must be said honestly: this policy will bring it no revenue at all within 2026; buying it is buying a rotation from loss-making small caps toward fundamentally sound names, and whether that rotation happens is precisely what §8-6 sets out to verify.

4老板电器002508SZ main board ±10% | Total 73 | Watch closely · Robam Appliances

• Related news: same as ③.
• Catalyst logic: indirect, industry trend.
• Fundamental verification: the outright kitchen-appliance leader (offline range-hood share 31.42%, gas cooktop 31.56%, first in the industry). Forward PE 12.66 and TTM PE 12.69 are almost exactly the same — it is the only name in the chain with no valuation distortion, which is itself an advantage on today's list. Gross margin 54.10%, highest in the group; gearing 28.71%, lowest in the group; cash RMB 1.489 billion. Channels: direct sales 48.85% + agency 37.85% = retail 88.12%, project only 10.62%, overseas only 0.86%.
• The negatives that must be stated alongside: 2026Q1 revenue −5.46%, attributable net −9.82%, ex-non-recurring −17.47%; full-year 2025 revenue −9.78%, attributable net −20.38%. The fundamentals are declining — just more slowly than peers.
• Technicals / sentiment: 8/19 only +4.32%, no limit-up, turnover rate only 2.46%, market cap RMB 15.5 billion — the same amount of money cannot push it 10%.
• Final judgment: it belongs to the "low-level watch list of names that did not hit limit-up," not to the follow-the-strength list. It is itself the subject of the §8-6 criterion: if it and 华帝 outperform 亿田智能 and 日出东方 today, money has started ranking by fundamentals; otherwise the entire policy chain gets downgraded.

5山金国际000975SZ main board ±10% | Total 72 | Watch closely (downgraded from priority deep-dive) · Shandong Gold Mining International

• Related news: the 2026 interim report was posted after the 8/19 close — revenue RMB 9.659 billion (+4.47%), attributable net RMB 2.416 billion (+51.43%).
• The basis problems this edition must call out (these were my first-version errors):

1. This is not new information. The company already issued an earnings pre-announcement on 2026-07-14 (attributable net RMB 2.35–2.48 billion, ex-non-recurring RMB 2.28–2.41 billion), and the actual RMB 2.416 / 2.385 billion lands in the upper-middle of the range — neither a surprise nor a blow-up.
2. The cumulative basis masked the single-quarter stall. Subtracting Q1 from H1 gives Q2 alone: attributable net RMB 1.022 billion (+13.4%), ex-non-recurring RMB 935 million (+4.5%), gross profit RMB 1.693 billion (+8.9%) — while the RMB gold price was still +28.5% YoY that quarter. The chain "gold price rises → earnings deliver" has already been falsified once by the company's own interim report.
3. More importantly: the gross margin actually rose 4.8pp sequentially (40.7% → 45.5%), which says the deceleration is not a price problem but a volume or product-mix problem. The interim report's product-level output and sales volumes were not obtained in this edition and are marked as not yet confirmable, which is the biggest blind spot.

• Fundamental verification: mining and processing gross margin 80.53%, highest in the group; mined gold 83.81%; operating cash flow RMB 3.067 billion (127% of attributable net — good quality). But 46.8% of revenue (RMB 4.524 billion) comes from metals commodity trading at a gross margin of −0.61%, a loss-making business; investing cash flow −RMB 3.835 billion, and gearing rose from 18.65% to 27.33% in half a year.
• Related-party relationship that must be disclosed: 山金国际 is a controlled subsidiary of 山东黄金 (Shandong Gold), which holds 28.94%. Buying both = betting twice on the same set of mines.
• Gold-price elasticity: +1% gold → attributable net +1.37%, fourth of the five — because its mined-gold gross margin is already 83.8%, leaving limited room for margin improvement.
• Final judgment: downgraded from "priority deep-dive" to "watch closely." It is still the only gold name with actual interim numbers that did not rise on 8/19, so the gap logic holds; but the claim of "dual catalysts of earnings + gold price" does not stand — its results were pre-announced long ago, and the single quarter is decelerating.

6南都物业603506SH main board ±10% | Total 71 | Watch closely · Nandu Property Service

• Related news: the provident-fund regulation explicitly adds "paying property-management fees on one's own home" as a permitted withdrawal — property management is the only industry directly named in the text of this amendment.
• Catalyst logic: direct (named in the text), with the path being better property-fee collection rates → better cash flow and bad debts. Graded under iron rule 2, this is one notch harder than kitchen appliances' "marginal demand improvement."
• Fundamental verification: PB 1.86, market cap only RMB 2.2 billion. Interim report not yet disclosed. The specific magnitude of any collection-rate improvement cannot be quantified, and this edition makes no estimate.
• Technicals / sentiment: limit-up on 8/19, but seal orders RMB 24 million / turnover RMB 89 million = 0.273, with 1 broken limit — medium seal quality; turnover under RMB 100 million.
• Final judgment: hardest logic, smallest market cap. An RMB 2.2 billion market cap means high volatility and easy manipulation, so the risk scale should be widened accordingly. See the turnover criterion in §8-11.

7中金黄金600489SH main board ±10% | Total 68 | Watch only (downgraded from priority deep-dive) · Zhongjin Gold

• My first version listed it as a "priority deep-dive," with the core reason being "PE 12.82, the lowest in the sector." That reason is wrong and must be withdrawn.
• Valuation truth: East Money's f9 is forward PE — back-calculating, 1221.04 ÷ (23.81 × 4) = 12.82 ✓ — it annualized 2026Q1, a quarter in which the RMB gold price was +62.2% YoY. Using the midpoint of the company's own interim pre-announcement (H1 attributable net RMB 4.35 billion), Q2 alone is attributable net RMB 1.969 billion, only +18.8% YoY, ex-non-recurring +13.1%, a cliff-edge fall from Q1's +129%/+88%. On a rolling TTM basis using the pre-announcement midpoint, PE is 18.5×, not 12.8×. "The cheapest in the field" is an arithmetic illusion.
• A harder point: 2026Q1 operating cash flow was −RMB 2.168 billion while attributable net over the same period was +RMB 2.381 billion — net profit converted to negative cash, plugged by +RMB 3.762 billion of financing; inventories of RMB 17.226 billion are on the high side.
• PB 10-year percentile 94.3%, the highest of the five, and layered on the low-PE appearance, this is the most textbook case of the "low PE + high PB percentile" cycle-top pattern.
• Final judgment: the direction is right, but the valuation advantage does not exist and the cash flow has a hard flaw. Listed as watch only.

8中微公司AMEC / 688012STAR Market ±20% | Total 67 | Watch only

• Related news: H1 revenue RMB 6.691 billion (+34.89%), attributable net RMB 2.825 billion (+300.22%); also investing RMB 3.5 billion in Lingang Phase II.
• Decomposing the headline number (the core of this item): attributable ex-non-recurring is only RMB 1.123 billion, +108.36% YoY; non-recurring gains were RMB 1.703 billion, 60.3% of attributable net — of which gains on disposal of non-current assets were RMB 991 million and fair-value changes and disposals of financial assets RMB 999 million (the latter being unrealized mark-to-market gains, with "other non-current financial assets" rising on the books from RMB 2.172 billion to RMB 3.205 billion). The Q1 report already named after-tax net gains of about RMB 397 million from selling 拓荆科技 (Piotech) shares. Roughly 60% of the "+300%" comes from trimming stakes in peers and mark-to-market gains, and has nothing to do with how well etchers are selling.
• But the core business really is accelerating: Q2 alone revenue RMB 3.776 billion (+35.48% YoY, +29.55% QoQ), ex-non-recurring RMB 645 million (+168.24% YoY, +35.07% QoQ); H1 ex-non-recurring already reached 72.4% of full-year 2025 ex-non-recurring. Gross margin 39.74%, essentially flat YoY.
• Two contrary signals: contract liabilities of RMB 2.772 billion, −8.9% versus RMB 3.044 billion at end-2025 (in the equipment industry, contract liabilities are customer prepayments and usually lead revenue); accounts receivable of RMB 3.058 billion, +53.9%, materially faster than revenue's +34.89%.
• Valuation: PB 11.86; TTM PE 84.2× (reported) vs. 167.0× (ex-non-recurring) — that gap is created entirely by this period's non-recurring gains.
• Basis calibration on the RMB 3.5 billion Lingang Phase II: the announcement's "locally attributed sales revenue of RMB 3 billion once at full production" is a local assessment target in a government investment agreement, not a company revenue forecast, and no timetable to full production is given; against 2025 revenue of RMB 12.385 billion it is only about 24.2%. The first risk warning in the announcement is that "the relevant documents for this project have not yet been signed."
• Sequencing correction (important): AMEC's −6.61% on 8/19 happened at 15:00 Beijing time on 8/19, whereas LRCX −6.33% and AMAT −3.53% were at the 16:00 ET 8/19 close = 4 a.m. Beijing time on 8/20, 9 hours later. So that down candle in AMEC corresponds to the US 8/18 session, not 8/19; US equipment names fell in two consecutive sessions.
• Final judgment: the core-business quality is real, but the headline is inflated nearly 3×, ex-non-recurring PE is 167×, and contract liabilities turned down sequentially. Listed as watch only. One directional reminder to add: equipment revenue speaks to the supply side, not the demand side — "domestic equipment is selling well" means more domestic capacity in the future, which is information pointing negative for memory/wafer prices.

9赛意信息300687ChiNext ±20% | Total 66 | Watch only · SIE Consulting

• Related news: announcement on the evening of 8/19 (2026-053) — two compute-service contracts signed with "Company W," RMB 6.450 billion including tax, equal to 311.11% of audited 2025 revenue, service term 60 months.
• Is it new information: a layered answer. On the capex side it is not new — the RMB 20 billion finance-lease facility on 5/14, RMB 833 million of procurement on 5/20, RMB 5.079 billion of procurement on 7/13, cumulative procurement authorization of RMB 5.912 billion; "we are going to buy a lot of GPUs to do compute leasing" was fully told long ago. The revenue side is new — this is the first compute-service sales contract of 2026, and until now the market only knew the company would spend money, not who would pay it.
• Basis check: audited 2025 revenue was RMB 2.07323 billion, and back-calculating, 64.5 ÷ 20.7323 = 311.11% ✓, so the amount and base are correct. But 311.11% is a cross-period basis of "5-year cumulative contract value ÷ 1 year of revenue." Amortized over 60 months, the annual average including tax is RMB 1.290 billion (= 62.2% of 2025 revenue), and excluding VAT RMB 1.217 billion (58.7%). The true annualized elasticity is +59%~+62%, not +311%.
• It is a framework contract, not confirmed backlog (this matters more than the amount): the announcement states that "the two parties shall confirm the specific content, quantity and term of the compute services in the form of orders." That is, the RMB 6.45 billion is a five-year ceiling, not locked-in backlog. This edition therefore lowered "directness of catalyst" from the "contract exists = 16 points" level, and uses the tape reaction in §8-12 to test whether the market sees through it.
• Economics: the revenue/procurement ratio is only 1.27× (64.5 ÷ 50.79). Gross margin, compute scale, unit price and customer identity are all undisclosed by the company, and this edition makes no estimate.
• Funding capacity is the tightest link: 2026Q1 cash RMB 345.8 million (−35.7% within the quarter), interest-bearing debt about RMB 780 million, operating cash flow −RMB 209 million. In announcement 2026-041 the company states that about 77.85% (≈RMB 3.95 billion) of the funding gap is "to be raised through a diversified mix of channels including equipment finance leasing and receivables factoring." Corroborating evidence: the 7/29 buyback plan was only RMB 30–60 million in size, and on 8/4 it specifically announced obtaining a "letter of commitment for a share-buyback special loan."
• Profit base: 2025 revenue RMB 2.073 billion (−13.45%), attributable net −RMB 107 million, ex-non-recurring −RMB 129 million (a loss); 2026Q1 revenue still −1.31%; PE (TTM) is negative. The H1 pre-announcement (7/6) puts attributable net at RMB 52.5–65 million, off an extremely low base.
• "Company W": the announcement gives only four generic descriptors and contains no wording about an exemption-from-disclosure procedure (whereas both of the company's prior procurement announcements explicitly stated that such a procedure had been followed). This edition draws no conclusion on compliance, but the net effect is that the counterparty's credit and the reality of demand behind an RMB 6.45 billion contract cannot be verified externally at all, while the first risk the announcement itself lists is "customer default risk… leading to idle servers."
• Final judgment: the amount is real, but the framing is inflated 5×, the company itself is loss-making, the funding gap is RMB 3.95 billion, and the customer is unverifiable, on top of AI capex being re-priced overnight. Listed as watch only.

10山东黄金600547SH main board ±10% | Total 64 | Watch only · Shandong Gold

• Related news: overnight gold breakout.
• Stage of the industry branch: pricing day 3. 8/17 +2.59%, 8/18 +1.69%, 8/19 +6.50% (main-force +RMB 313 million), +10.51% cumulatively over four sessions — the only name in the sector to react early.
• Its gold-price elasticity is in fact the highest in the group (+1% gold → attributable net +2.98%), but it carries three discounts:

1. Hedging. The 2026-07-10 announcement (interim 2026-037) shows the company conducting hedging of self-produced gold, and applying cash-flow hedge accounting from 2026-07-01, with the announcement itself stating that one effect is "smoothing the impact of short-term gold-price swings on net profit." It is the only one of the five to explicitly disclose hedging of self-mined gold, and the effective date falls exactly in the quarter of this breakout. The hedge ratio and exposure are undisclosed, still to be verified.
2. Minorities take 40.7% (2026Q1 minority interests RMB 993 million vs. attributable net RMB 1.446 billion), mainly due to consolidating 山金国际 and others.
3. Goodwill RMB 12.654 billion = 27.5% of attributable equity; gearing 62.57%, the highest of the five.

• One genuine bright spot: operating cash flow of +RMB 6.095 billion, 4.2× attributable net, the strongest in the group.
• Another signal: it is the only one of the five with no interim pre-announcement (confirmed from two sources: neither the 4,865 records in East Money's interim pre-announcement database nor the 60 records in its earnings flash database contain 600547) — under the disclosure rules, this hints that H1 attributable net YoY is most likely below 50%.
• Final judgment: it is the branch leader, and also the worst entry point today. The lesson from yesterday's recap was precisely "got the call right but expressed the position in the worst place" — listed as watch only, used as an observation gauge of branch strength (§8-2) rather than as a position.

---

6. Pass List

---

7. Intra-Branch Ranking

Branch 1: Gold / Silver (grade A)

Gold-price elasticity estimate (gold +1% → attributable net profit +X%), model: self-mined gold revenue × 1% × (1 − effective tax rate) × attributable share ÷ TTM attributable net; assumes unchanged output and unit cost, and ignores hedging.

• Hardest single name: 赤峰黄金 (600988) — the fewest discounts.
• Highest silver purity: 盛达资源 (000603) — all three concentrates contain silver.
• ⚠️ One reversal before finalization: 兴业银锡 (000426) was ranked 2nd in the first draft on the argument that "−7.34%, the biggest drop = a mispricing"; after the single-name announcement re-sweep, an announcement that day proposing an RMB 1.8 billion acquisition of an ST company (威领股份 000631) was found, giving the drop a clear company-specific cause, so it has been moved to Pass.
• Counterintuitive but important: the higher the gross margin, the lower the percentage elasticity to the gold price. 山金国际's mined-gold gross margin is already 83.8%, leaving limited room for margin improvement; the leverage is actually greater at 赤峰's 58.5% and 山东's 50.6% on self-mined gold.
• Who is passed: 湖南黄金 (halt + pseudo-elasticity); 晓程科技 (Xiaocheng Technology, 300139) with PE 114.8 and a 19.69% turnover rate — pure concept.
• ⚠️ Risk common to the whole branch: PB historical percentiles are broadly above 90% (中金 94.3%, 湖南 92.8%, 山金 91.3%), while PE looks cheap because of Q1 annualization — this is exactly the "low PE + high PB percentile" cycle-top pattern. That is precisely why this edition treats it as a short-term trading opportunity and not as a value re-rating.

Branch 2: The provident-fund chain (grade B+, a sentiment line)

• The core conclusion of this branch: the benefit ranking (华帝 > 老板 > 万和 > 志邦 > 日出东方) is almost perfectly inverted versus the 8/19 gain ranking. This edition therefore puts its recommendation slots on the top two that did not hit limit-up, and lists all six limit-up names as watch only or Pass.
• Seal-quality tiering (corroborating "intraday relay rather than consensus"): 罗普斯金 sealed in the 09:25 auction / 0 broken limits / 0.965 → 志邦 09:32 / 1 broken limit / 0.356 → 好太太 09:36 / 2 broken limits / 0.240 → 日出东方 11:01 → 亿田 13:07 → 万和 13:51 (last).

Branch 3: Coke / Coal (grade B, watch only overall)

• The fatal problem with the whole branch: all four limit-up names have negative PE, and not one can convert a coke price increase into verifiable profit. Given that the push was initiated by coking plants and acceptance by steel mills is unconfirmed, this is pure price speculation.

---

8. Today's Open Verification Signals

Every item is written in the falsifiable form "is X > threshold," with the read time noted. Yesterday's recap concluded that within the same report the verification points were right 16/17 while the recommendation slots were wrong 5/6 — the difference is that the verification points were forced into falsifiable form.

Auction signals (09:15–09:25)

1. [Can precious metals be executed] 09:25: is the opening gap on 赤峰黄金 (600988) and 山金国际 (000975) < +4%?

• < +4% → the gap has not been fully filled (overnight RMB gold implies about +4.2%).
• ≥ +6% → the gap has been exhausted in one move, and the day is chase territory. This is the single most important risk criterion in this edition. Given GDX was already +9.42% on 8/19, this scenario is the base case today, not a tail case.

2. [Branch strength] 09:25: is the opening gap on 山东黄金 (600547) > that on 赤峰黄金 (600988)?

• Yes → money is still buying "the most famous" rather than "the cleanest structure," meaning sentiment diffusion rather than fundamental pricing, and the entire precious-metals branch drops a notch.

3. [Can the silver line hold] 09:25: does 盛达资源 (000603) gap up > +3%?

• Yes → the silver price (overnight +1.93%, SLV +4.47%) is genuinely transmitting into A-shares.
• Opens down or gaps up < +1% → abandon the silver line. 3b. [Silver-purity test — the key verification after this edition's reversal] Close: does 盛达资源 (000603) outperform 兴业银锡 (000426)?
• Yes → money is pricing on silver purity (盛达's three concentrates all contain silver and are 84.7% of revenue; 兴业's silver is about 40% of gross profit and tin about 35%), meaning genuine silver-price transmission.
• No (兴业 outperforms) → the driver is an oversold bounce rather than the silver price, meaning the market has not digested 兴业's RMB 1.8 billion acquisition of an ST company; the silver line should be treated as a sentiment line with a shortened holding window.

Sector signals (09:30–10:30)

4. 10:30 read: among the 13 names in the precious-metals sector (BK0732), is the number of advancers ≥ 9? (8/19: 4)

• ≥ 9 → the branch holds; < 6 → the overnight gold rise failed to transmit, falsifying this edition's first branch.

5. 10:30 read: is the main-force net in the gold sector (BK1617) > +RMB 800 million? (8/19: +RMB 432 million)

• Turning negative → distribution into the gap-up.

6. [Is the policy chain sentiment or logic — the single most important item in this edition] Close: do the gains of 华帝股份 (002035) and 老板电器 (002508) both beat 亿田智能 (300911) and 日出东方 (603366)?

• Yes → money is starting to rank by policy exposure and fundamentals, and the policy chain can continue.
• No (loss-making / mismatched small caps still lead) → confirmed as pure sentiment speculation, and the entire policy chain is downgraded to watch only.
• Supporting criterion: if 罗普斯金 (002333) stays locked at limit from the open, sentiment-driven leadership is unchanged.

7. 10:30 read: is the de-duplicated limit-up count across coke (BK1492) + coal (BK0437) ≥ 3?

• Baseline correction: after de-duplication 8/19 was 4, not 7. The two sectors' constituents overlap — 宝泰隆 (601011), 陕西黑猫 (601015) and 美锦能源 (000723) belong to both, while 大有能源 (600403) belongs only to coal. Using 7 as the baseline would make "down from 4 to 3" read as "still healthy."
• < 3 → the price-hike expectation has been realized and the branch is ebbing.

8. Close: is the main-force net in the coke industry > 0 for a second consecutive day? (8/19: +RMB 681 million)

Single-stock signals

9. Close: is the main-force net on 赤峰黄金 (600988) > +RMB 300 million? (8/19: +RMB 170 million)
10. Close: does 山金国际 (000975) outperform 山东黄金 (600547)?

• Yes → the market is pricing on "unreacted + actual interim numbers."
• No → leader inertia dominates, and this edition's stock-selection logic (pick the ones that have not risen) does not apply today.

11. Close: is 南都物业 (603506)'s turnover > RMB 150 million? (8/19: only RMB 89 million)

• No → an RMB 2.2 billion market cap lacks the liquidity to carry the money, so there is no durability.

12. [Quality of the 赛意信息 order] Close: does 赛意信息 (300687) seal its ±20% limit-up with a seal orders / turnover ratio > 0.3?

• Gap up then fade → the market has already seen through "60-month amortization + anonymous customer + the company itself is loss-making," consistent with this edition's judgment.

Risk signals

13. [Survival rate of yesterday's high-ladder names] Close: among the 36 limit-up names of 8/19, is the number of advancers today ≥ 18 (50%)?

• < 30% → the wealth effect has not been repaired, and all ladder-relay strategies fail.

14. [3-board risk positions] Close: do 金健米业 (600127) and 红四方 (603395) make a 4th board?

• Both fail → high-ladder names are confirmed to be disintegrating (this edition already judged their relay money exhausted from LHB net buys of +RMB 3 million / −RMB 17 million).

15. [Style confirmation] Close: does the SSE 50 outperform the ChiNext Index for a second consecutive day? (8/19 gap: 4.73pp)
16. [Has AI hardware stopped falling] Close: is the main-force net in the semiconductor sector > −RMB 10 billion? (8/19: −RMB 36.358 billion)
17. [恒瑞: earnings vs. buyback, which one does the market read] 09:25 auction: is 恒瑞医药 (600276)'s opening gap > +3%?

• > +3% → the market is reading "buyback + share-ownership plan," or reading the misreported "+29.67%." The latter is mispricing; do not participate.
• Opens down or flat → the market has seen through the −1.94% revenue, and the buyback positive is offset by the earnings.
• Criterion: if the innovative-drug sector (BK1106) rebounds in sympathy through the day, it is policy + buyback resonance; if only 恒瑞 rises while the sector does not follow, it is a single-point buyback effect and cannot be extrapolated.

---

9. Final Conclusions

① The 5 stocks most worth watching today

This section executes the mandatory consistency check proposed in yesterday's recap: every name must be tagged with which verification point in §8 it triggers and whether the direction is consistent. Names with conflicting directions may not enter the recommendation slots. Under that constraint, 山东黄金 (600547), 中金黄金 (600489), 罗普斯金 (002333) and 万和电气 (002543), though within the field of view, are all kept out of the recommendation slots because they trigger the reverse conditions of §8-2 / §8-6 or because of valuation distortion.

Note on position expression: the 5 names are spread across 2 unrelated branches (precious metals 3 / policy 2), not concentrated in a single high-β direction. The failure on 8/19 was exactly putting 3 robotics names into a list of 5 at the same time.

⚠️ Must be stated explicitly: 1 of the 5 (山金国际) has only "partial" directional consistency; this edition does not hide that but flags it and places it last. Its results were pre-announced 7/14 and Q2 ex-non-recurring is only +4.5%; it is included because of "no reaction on 8/19," not because of "an earnings beat." The other 4 are fully consistent in direction.

⚠️ One substitution must also be stated: the first draft's No. 2 was 兴业银锡 (000426), on the argument that "−7.34% is a mispricing." The mandatory single-name announcement re-sweep before finalization found that on 8/19 it announced a proposal to acquire, for up to RMB 1.8 billion, a company with negative attributable net assets, three straight years of losses, and an existing delisting-risk warning (威领股份 000631); the drop has a clear company-specific cause, "mispricing" does not hold, and it has been replaced by 盛达资源 (000603), which has a clean announcement record and higher silver purity.

② The 3 strongest branches today

③ Directions not worth chasing today

1. Memory / CPO / optical modules / semiconductor equipment — day 5 of the ebb; overnight STX −7.87%, WDC −6.87%, LRCX −6.33%, AMAT −3.53%. Note the distinction: what fell was mainly HDD and equipment, while DRAM's MU was only −0.39% and NVDA −0.99%, but stay out until main-force outflows stop.
2. The entire robotics chain — 8 names in the chain hit limit-down on 8/19, main-force −RMB 37.974 billion; 崧盛股份 (300899) terminating its robot harmonic-reducer contract is added negative evidence; 绿的谐波 (±20%) was −17.67% on 8/19 without hitting limit-down, so there is still room to fall.
3. Coke / Coal (yesterday's recap's "main attack" direction, downgraded in this edition) — the price hike was initiated by coking plants and acceptance by steel mills is unconfirmed; all four limit-up names have negative PE; 宝泰隆's LHB flows are 83.5% of turnover.
4. The six home-furnishing names that hit limit-up on 8/19 — benefit and gain are inverted, and among them 万和电气 (exports 41.66%, H1 pre-announcement −76%~−84%), 日出东方 (kitchen appliances only 11.26%, H1 −11.3%), 志邦家居 (H1 projected loss) and 亿田智能 (attribution mismatch, 5 straight quarterly losses) all lack fundamental support.
5. High 3-board names — 金健米业 (600127) LHB net buy only +RMB 3 million, 红四方 (603395) net buy −RMB 17 million.
6. Limit-up names with ≥ 4 broken limits on 8/19 — 尚太科技 (001301) 10 times, 永顺泰 (001338) 7 times, 万向德农 (Wanxiang Denong, 600371) 5 times, 我爱我家 (000560) 4 times, 亨迪药业 (301211) 4 times.
7. 恒瑞医药 (600276) — do not chase "because of the interim report," but do not short "because the buyback expired" either. The interim report is actually revenue −1.94%, attributable net +0.34% (the misreported +15.88%/+29.67% are 2025 figures); but this morning it simultaneously announced a new RMB 1–2 billion buyback plan and an employee share-ownership plan, so buyback bid support has not been interrupted. The two point in opposite directions; listed as watch only.
8. 湖南黄金 (002155) — halted from 8/20, not tradable.

④ Final one-sentence judgment

What is tradable today is a gap, not a re-rating: the overnight USD gold breakout implies a roughly +4.2% gap in the RMB gold price while A-share precious metals were still falling on 8/19 — but the YoY growth of the RMB gold price has converged from Q1's +62.2% to +13.9%, and the Q2 single-quarter profit growth of the five miners has fallen in unison to +13%~+23%, so treat this as a short-term event rather than holding on an earnings re-rating; the first choice is 赤峰黄金 (600988), the one with the fewest discounts, rather than 山东黄金, which has the highest elasticity but just activated hedging; on the silver line, recognize purity — 兴业银锡 (000426) has the most extreme divergence but about 1/3 of its gross profit comes from tin, and the truly pure silver exposure is 盛达资源 (000603). On the policy chain, only the names that did not hit limit-up — 华帝股份 (002035) and 老板电器 (002508) — because the benefit ranking of those six limit-up names on 8/19 is inverted versus their gain ranking; coke and AI hardware are not on today's watch sequence.

---

⚠️ Risk warning: this list is pre-market information organization 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 industry-chain mapping errors; it must not be used directly as a basis for trading.

Data-Pull and Run Log (not sent to clients)

1. The nine traps this edition saw through (by importance)

0. The "mispricing" call on 兴业银锡 is the wrong conclusion this edition came closest to publishing, stopped by risk-auditor before finalization. The first draft had it at No. 2 in the recommendation slots, arguing "silver was +1.93% overnight while it was −7.34% on 8/19 — divergence in both direction and magnitude = a mispricing." The audit found: that same day (8/19) the company announced (2026-72) a proposal to acquire control of 威领股份 (Weiling, 000631) for up to RMB 1.8 billion plus a partial tender offer. On my re-check of the original filing, Weiling's attributable net assets at end-2025 were −RMB 32.0512 million (negative), it lost money for three straight years 2023–2025 (−RMB 290 / 412 / 454 million), revenue shrank from RMB 1.141 billion to RMB 320 million, and it has already triggered delisting-risk warning; there are also two subsidiary safety-incident announcements dated 7/28 and 7/31. The −7.34% and the main-force −RMB 263 million have a clear company-specific cause; "mispricing" does not hold. It has been moved to Pass, and the silver slot went to 盛达资源 (000603), which has a clean announcement record and higher silver purity. Root cause (worth retaining): I selected a single name using "sector divergence" but never ran a single-name announcement re-sweep on it. The divergence at the sector level was real, but it does not automatically transmit to every constituent — any "mispriced / oversold" argument must first rule out negative company-specific announcements on the day, otherwise "divergence" only means I did not see the reason. Before finalization a mandatory single-name re-sweep was run on all 6 candidate recommendations (赤峰, 盛达, 华帝, 老板, 山金, 南都), and all came back clean.

1. Forward PE was taken for TTM PE — this is the most systematic error in this edition, and the entire first version was wrong on it. East Money's f9/f162 is market cap ÷ latest single-quarter net profit × 4. Verified by back-calculation name by name: 中金黄金 1221.04÷(23.81×4)=12.82 ✓, 万和 56.29÷(0.796×4)=17.68 ✓, 老板 155.25÷(3.066×4)=12.65 ✓. It annualized 2026Q1 — a quarter in which the RMB gold price was +62.2% YoY — and the seasonally strongest Q1 for appliance makers. After correction: 中金黄金 12.82 → 18.5, 万和电气 17.68 → 48.69, 华帝 10.40 → 14.21, 日出东方 84.64 → 36.49. The first version's two core recommendation rationales — "中金黄金 is the cheapest in the field" and "万和电气 has the healthiest valuation" — are both void.

2. 山金国际's "H1 +51.43%" is a cumulative basis that masked the single-quarter stall. Extracting Q2: attributable net +13.4%, ex-non-recurring +4.5%, gross profit +8.9%, while the RMB gold price was still +28.5% YoY that quarter. And results were already pre-announced 7/14 (RMB 2.35–2.48 billion), with the actual RMB 2.416 billion landing in the upper-middle of the range — not a beat at all. The first version wrote it up as "dual catalysts of earnings + gold price" and ranked it first, the most serious misjudgment in this edition; it has been lowered to No. 5 and explicitly tagged as only "partly" directionally consistent.

3. 恒瑞医药's interim report was misreported wholesale. The media's RMB 15.761 billion/+15.88% and RMB 4.450 billion/+29.67% are 2025 figures (disclosed 2025-08-21); the actual 2026 figures are RMB 15.456 billion (−1.94%) and RMB 4.465 billion (+0.34%). The way it was caught was arithmetic back-calculation: 154.56÷157.61−1 = −1.93%, consistent with the API's YoY, proving the media were quoting last year's same report. 恒瑞 discloses its interim report every year on 8/20–8/21 — a "recurring column + fixed timing" type of trap.

4. The 恒瑞 buyback — this one I got wrong myself and then dug back out myself, and the process is worth recording. The first draft inferred from "the plan was first disclosed 2025-08-21 with an implementation period to 2026-08-19" that "buyback bid support disappears from 8/20, a marginal negative." The premise of the inference was correct (the old plan did expire on 8/19, with RMB 1.004 billion actually repurchased at an average price of RMB 60.51), but the conclusion was wrong — while re-checking 600276 against the exchange announcement feed before finalization, I found that on 8/20 the company simultaneously disclosed a new buyback plan (interim 2026-123, RMB 1–2 billion, cap RMB 81.78) and the draft 2026 employee share-ownership plan. "The old plan expired" is a fact; "bid support disappears" is an inference I added, and that inference went into the conclusion unverified. The text has been rewritten throughout as "two lines pointing in opposite directions," and the label upgraded from Pass to watch only. Lesson: for timing-type facts like "X expires / is terminated," you must take one more step and check "is there a successor plan," or one correct fact will yield a conclusion pointing the wrong way.

5. 湖南黄金 is halted from 8/20. The stock was −0.69% over 4 sessions and would have entered the first tier on a purely quantitative ranking. Confirmed by measurement against the exchange announcement feed. And even if tradable it should not be given gold-stock elasticity — externally purchased non-standard gold is 88.4% of revenue at a gross margin of only 0.076%, and antimony is 45.5% of gross profit.

6. 亿田智能's limit-up was mis-attributed. It rose on the RMB 1 billion compute-center construction contract from the evening of 8/18, not on the provident fund. Decisive counter-evidence: 火星人 (Marssenger, 300894) — also an integrated-stove maker, 100% domestic retail, essentially identical policy exposure — was −6.97% the same day. If the driver were the provident fund, 火星人 would have no reason to move the other way.

7. "兴业银锡 = a silver stock" is the second near-miss "湖南黄金-style" error I almost made myself. I only checked its business mix after writing it into a recommendation slot, and the result is that in FY2025 mined silver was 39.17% of revenue, mined tin 29.70%, mined zinc 17.57%; split by gross profit, silver is about 40% and tin about 35%. It is a genuine silver stock (silver really is the largest and highest-margin line), but about 1/3 of its gross profit is driven by the tin price, and this edition has no tin-price data. Directional consistency was changed from "consistent" to "partly consistent," and §8-3b was added to test, via "does 盛达资源 outperform 兴业银锡," whether the driver is really the silver price or an oversold bounce. Lesson: any name whose thesis rests on a commodity price must have its product-level gross-profit split checked before it enters a recommendation slot — 湖南黄金 (gold is 93.7% of revenue but only 52.1% of gross profit) had just taught the same lesson, and I nearly repeated it in the same edition.

8. Yesterday's recap's "coke price hike takes effect at midnight on 8/20" could not be confirmed. Only the 8/17–8/18 first-round push of RMB 50–55/tonne initiated by coking plants (Mysteel) can be confirmed, and no primary basis for a "midnight 8/20 execution" was found; moreover all four limit-up names have negative PE. The branch has been downgraded from "main attack" to "watch only."

2. Full-scope basis checks

⚠️ One item that did not pass: the industry/concept sector clist request returned only 100 records when pz=600 was requested (declared total=496 / 504). This edition uses only the ranking information from its top 100 in descending order, and draws no conclusion whatsoever about the tail of the decliners list, avoiding the "one-sided pagination bias" already recorded in this repository.

3. Data-pull failures

• push2.eastmoney.com and push2his.eastmoney.com were unavailable across the board (the former 502, the latter ConnectionError/ProtocolError). Quotes and sectors were rerouted to push2delay.eastmoney.com (200 throughout); multi-day candles for single stocks were rerouted to Tencent's web.ifzq.gtimg.cn.
• datacenter-web's RPT_LICO_FN_CPD returns success:false when filtered with REPORT_DATE; the actual field is REPORTDATE (no underscore).
• CNBC's quote API has no data for CL.1/GC.1; you must use @CL.1/@GC.1 (with the @ prefix).
• 财联社's cls.cn/telegraph is JS-rendered and WebFetch cannot retrieve items; substituted with the exchange announcement feed + authoritative media roundups.

4. Freshness checks (guarding against "silently a day stale")

• All US data had last_time verified: .SPX 2026-08-19T16:38:31-0400, MU/STX/WDC/LRCX all 16:00:00-0400, SMH/IGV/GLD 16:10–16:15 ET → confirmed as the 8/19 close basis.
• @GC.1 2026-08-19T19:01:36-0400 = Beijing 8/20 07:01, a reading about 10 minutes before generation, with the read time noted in the main text.
• GLD's single-day +3.84% was cross-checked against two independent sources, CNBC and stockanalysis, with a closing price of 413.84 in both; the break above the 8/10–8/18 range (398.55–405.49) was confirmed on one-month daily bars.
• A third cross-check: $4,518/oz ÷ 31.1035 × 6.7854 (Bank of China conversion rate) = RMB 985.5/gram, while SGE 945.22 × 1.0384 = RMB 981.4/gram, a 0.4% difference — the USD gold price, GLD and the SGE corroborate each other across three sources.
• A sequencing error has been corrected: AMEC's −6.61% on 8/19 (15:00 Beijing) preceded the 8/19 US close (04:00 Beijing on 8/20) by 9 hours, so it cannot have been driven by the latter; it corresponds to the US 8/18 session.

5. Not done / not obtained in this edition

• Absolute self-mined gold output, 2026 production guidance and all-in sustaining cost (AISC) for all five gold miners could not be obtained, and are marked "cannot yet be confirmed." The elasticity ranking substitutes "self-mined gold revenue share + segment gross margin," which is logically equivalent but not a substitute for production guidance — if any of them raises output substantially in 2026, volume growth could drive earnings independently of price and change the ranking. This is the biggest known blind spot.
• 山东黄金's hedge ratio and locked price range are undisclosed, so "how much the hedging weakens the elasticity" is only flagged qualitatively, not quantified.
• Tin price data could not be obtained, while tin contributes about 35% of 兴业银锡's gross profit. So all that can be said about the stock is that the silver price benefits about 40% of its gross profit; no composite price elasticity can be given, as stated explicitly in the main text.
• 山金国际's product-level output and sales volumes in the interim report could not be obtained, so the cause of the Q2 deceleration (volume or mix) cannot be settled.
• TTM PE for 兴业银锡, 西部黄金, 南都物业 and 好太太 cannot be computed because their interim reports are undisclosed, so they are marked "cannot yet be confirmed" in the tables and were given no valuation credit.
• None of the six home-furnishing companies discloses a split of retail into "renovation of existing homes vs. new-home fit-out," which is the only hard indicator for judging policy benefit — a sector-wide blank.
• No estimate was made of the magnitude of the provident-fund rules' revenue pull for the industry, because every locality's implementation rules (withdrawal caps / frequency / documentation) are unpublished, leaving the most critical parameter missing. Explicitly: no fabricated estimate, and no citation of any sell-side estimate.
• The gross margin, compute scale and customer identity behind 赛意信息's RMB 6.45 billion contract are all undisclosed by the company, so this edition makes no post-depreciation gross-margin derivation.
• The specific targets behind AMEC's RMB 953 million of "disposal of long-term equity investments" in H1 are not itemized in the interim report (the Q1 report named only 拓荆科技 at RMB 397 million); the portion newly added in Q2 remains to be verified.

5b. The other four items stopped by risk-auditor before finalization

6. Sub-agent usage and this edition's self-corrections

Three fundamentals-analyst agents (gold chain / home-furnishing chain / compute-order chain) and one risk-auditor were launched. All three fundamentals agents overturned key conclusions of my first version:

• The gold agent found the forward-PE trap, 山金国际's Q2 stall, 山东黄金's hedging, and the convergence of RMB gold-price YoY growth;
• The home-furnishing agent found that 万和电气's TTM is 48.69 and about to turn negative, that benefit and gain are inverted, and that 亿田's attribution is mismatched (the 火星人 counter-evidence);
• The compute agent found that AMEC's non-recurring share is 60.3%, that 赛意's "RMB 20 billion" is a finance-lease facility, and that 赛意 was actually loss-making in 2025.

risk-auditor then overturned 2 of the second version's recommendation slots (兴业银锡, and the characterization of 恒瑞) — see §5b. In the end, 3 of the five names in §9① were swapped in after the first draft.

The first version of the report was rewritten wholesale according to these three verifications, and 3 of the five recommendation slots in §9① were replaced. This is worth retaining: the screening condition "the catalyst has happened and the price has not reacted" is itself correct, but it only screens for "worth looking at," not directly for "worth buying" — it must pass a further check on single-quarter basis and valuation basis. The first version did exactly that: after screening out the correct branch, it selected the wrong stocks using the wrong valuation basis (forward PE) and the wrong earnings basis (cumulative rather than single-quarter).