A-Share Pre-Market Brief | 2026-08-31 Monday

Scan window: 2026-08-28 (Fri) 15:00 close → 2026-08-31 (Mon) 07:00 Beijing time. This window spans the entire weekend, and the weekend was the last two days before the statutory disclosure deadline for interim reports, so the information volume is far greater than an ordinary trading day's overnight window. Data read timestamps: A-share quotes are the 2026-08-28 closing snapshot (Tencent qt.gtimg.cn, each name carrying a trade timestamp, verified as the post-16:14 settlement caliber); the limit-up / failed-limit-up pools are the Eastmoney push2ex official pool, 8/28 caliber; the LHB (exchange trading-list disclosure) is the post-18:00 8/28 exchange caliber — last Friday's post-close piece went out at 15:30 and could not include it; this piece fills the gap; all interim-report financials come from the original interim-report PDFs on CNINFO / the exchanges (downloaded and parsed one by one, including income statements and notes, not second-hand paraphrase); the overnight U.S. market is the 2026-08-28 16:00 ET close (= 8/29 04:00 Beijing time).

Caliber boundaries of this piece (judge conclusion strength accordingly): ① Every "ex-non-recurring net profit" in this piece is taken from the original interim report; the "net profit" field of third-party aggregation databases is not used — the most important conclusions here are generated precisely by the gap between "attributable to parent" and "ex-non-recurring," and an aggregation database would miss all of them (see §0 and §6). ② The weekend disclosure batch is a "negative selection" sample: 944 companies disclosed in a cluster on 8/29–8/31, with a median net-profit YoY of −5.6% and only 48.0% posting growth. The strong-earnings names in this piece are exceptions within that weak sample, not representatives of the batch. ③ August official PMI is released today at 09:30 (the same minute as the open), later than this piece went out (07:00), so this piece contains no actual August PMI value. Baseline: July manufacturing PMI 49.2% (−1.1pct MoM). ④ ⚠️ The PE/PB fields from the quote vendor are not taken at face value in this piece; they are all self-computed and reverse-checked. This time we measured, on the same interface at the same moment, two names with different financial-caliber freshness: Tencent's PE for 长鑫科技 (ChangXin Memory, 688825) has already absorbed the 8/29 interim report, while its PE (TTM) of −411.77 for 沐曦股份 (MetaX, 688802) is still stuck at 2026Q1 and has not absorbed the 8/30 interim report; Tencent's PB of 19.79 for ChangXin does not reconcile with any net-asset caliber (the attributable-to-parent caliber should be 29.5x). All valuations in §3 and §5 of this piece are self-computed, with the formulas shown. ⑤ The limit-up/limit-down pools have been validated with tc == len(pool) (8/28 limit-ups 82/82, failed limit-ups 16/16); the limit-down pool returned tc=1 with an empty pool array, a known silent failure, so its details are not cited in this piece.

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

I. Interface failures

• Eastmoney push2/clist has now returned 502 Bad Gateway for a fourth consecutive full trading day. Affected: sector quotes, sector fund flows, whole-market paginated snapshots, ulist.np.
• Working substitute routes (same as last Friday; recommend making them permanent): ① quotes via Tencent qt.gtimg.cn; ② limit-up / failed-limit-up pools via push2ex; ③ sector fund flows via ak.stock_board_industry_summary_ths(); ④ LHB ak.stock_lhb_detail_em('20260828','20260828') succeeded this time, 63 rows — confirming the scheduling judgment that "last Friday's post-close piece failing to get it was not an interface failure but a 15:30 publication time earlier than the 18:00 disclosure." Do not raise an alert.
• The limit-down pool getTopicDTPool returned tc=1 with pool=[], i.e. the "nonzero tc but empty pool" silent failure; per the rule, it is not cited.
• Announcement route: Eastmoney np-anotice-stock is stable; concatenate art_code into https://pdf.dfcfw.com/pdf/H2_<art_code>_1.pdf to get the original. This piece downloaded and parsed 25+ interim reports and interim announcements in original form. WebFetch still returns only binary for PDFs; local pdfplumber works fine.
• The NFRA site nfra.gov.cn was rate-limited on this visit; three of the five supporting administrative measures could not be obtained in original form (see the open-question list in §2.2).

II. ⚠️ Search contamination was extremely severe this time — five hits in a row, and it must be logged

1. stcn.com/article/detail/960559.html "These Important Weekend Headlines May Move the Market" is entirely 2023-08-27 content (stamp duty halved, Evergrande resuming trading, Jiaxing lifting purchase limits), returned as 2026 weekend headlines. The same trap as "recurring column headlines hide the publication date" was hit again; the whole article has been discarded.
2. Secondary error: "the CSRC directed the Shanghai, Shenzhen and Beijing exchanges to further cut securities transaction handling fees from 8/28, with 30 brokers cutting commissions" is word-for-word from the 2023 article above. The entire item has been deleted, and it is therefore not used as the explanation for 锦龙股份 (Jinlong Development, 000712)'s 3 consecutive limit-ups. Without checking the publication date, this item would have become the false foundation of an entire "brokerage branch."
3. August PMI at "49.1%, −0.3pct MoM" was returned as already-published data. Reverse computation falsifies it immediately: that value implies a prior month of 49.4%, whereas the statistics bureau's official July figure is 49.2%; and 49.1/−0.3 matches August 2024 exactly. Judged to be 2024 data; discarded.
4. U.S. equities' 8/28 close as "Nasdaq +2.59%, SOX up over 6%" is the exact opposite of first-hand data (actual SMH −3.47%). Falsification method: stockanalysis's previous close (NVDA 227.98 / SMH 573.00 / MU 935.39) matched exactly, name by name, the Thursday closing prices recorded in last Friday's pre-market piece, and each was reverse-computed as "close/previous close − 1." Discarded.
5. "Nvidia fell on its commitment of $105 billion to support OpenAI's Ohio data center" was pinned to 8/28 — that event was actually 8/16–8/18. This piece therefore assigns no specific catalyst to Friday's U.S. semiconductor selloff.

III. ⚠️ Errors in second-hand reporting about the original policy texts (caught by sub-agents, very valuable)

• ifeng.com claimed that projects that had already obtained a pre-sale permit before the new rules take effect may have their individual housing loans handled under the old rules — the anchor in the original text is the "construction works planning permit," and it governs pre-sale conditions and fund supervision, not the timing of mortgage disbursement. The two differ substantially in coverage, and this error was copied and spread across multiple search summaries.
• The Xinhua release on development-loan tenors wrote only "up to 5 years / 7 years," omitting "in principle no more than 3 years / 5 years."
• ifeng.com's headline on the ChangXin Memory interim report gives a gross margin of 84.84%, which is wrong, off by 0.10pp from the income statement it relies on; the correct figure is 84.74% (self-computed and re-checked in this piece).
• "The income-to-debt-service ratio cap raised from 55% to 60%" appears repeatedly in search summaries but could not be found in any first-hand source or authoritative reprint; marked unverified and not cited.

IV. Assessment of sub-agent output (both were fed back in, and both overturned this piece's first draft)

• fundamentals-analyst (ChangXin + MetaX): overturned the first draft's judgment that "84.7% gross margin is above the industry's history and is questionable" — it pulled SK Hynix / Micron / Samsung statements for the same period; ChangXin's Q1 gross margin of 79.16% vs SK Hynix's 79.27%, a gap of 0.11pp, landing on the same point across accounting standards and across auditors. The historical rule that "DRAM has never had an 85% gross margin" has been voided by the peers themselves in this 2026 cycle. It also found that minority interest takes 27.8% of profit, that Tencent's PB does not reconcile, and that the forecast was a range rather than a point value — all three changed conclusions.
• fundamentals-analyst (property chain): found that the first draft had missed an entire document (the MOHURD et al. "Circular on Improving the Commodity Housing Sales System," in which the pre-sale condition itself mentions "structural topping-out"), and overturned the first draft's attribution for two limit-up property names (5i5j was actually earnings-driven and profitable in H1; 香江控股 (Xiangjiang Holdings) was actually driven by a compute-power related-party transaction, and its PB of 3.07 is a high, not a low).

---

0. One-Sentence Summary of the Day

What has to be priced at today's open is not one piece of news but an entire batch of information that landed over the weekend and that the market has had no chance to trade — 944 interim reports, plus a full set of real-estate system-reform documents released only at 19:08 after the close. And within that batch, the thing to read first is the gap between "attributable to parent" and "ex-non-recurring." This piece downloaded and cross-checked the original text of every key interim report, and the conclusion is that sorting by the "net profit" field of a third-party aggregation database produces an almost inverted ranking: 沐曦股份 (MetaX, 688802) headlines "net profit RMB 612 million, +429% YoY," while the original text shows ex-non-recurring net profit attributable to parent of −RMB 48.8595 million, with the core business still lossmaking — the source of that RMB 612 million is not government grants (only RMB 9.34 million) but a listed-equity holding costing about RMB 107 million that has marked up to RMB 994 million in half a year, with the company itself stating in the report that it accounts for "105.75% of total profit" and "sustainable: no"; 华峰测控 (Huafeng Test & Control / Accotest, 688200) shows +112.60% attributable to parent but only +39.06% ex-non-recurring, and its ex-non-recurring weighted ROE fell rather than rose YoY (4.81%→4.24%); conversely, 国科微 (Nationalchip, 300672) shows +872.78% attributable to parent but +1,917.55% ex-non-recurring, with operating cash flow 3.6x net profit, and 伟测科技 (Wintest, 688372) shows +57.61% attributable to parent but +171.33% ex-non-recurring. The strongest branch is therefore "domestic semiconductor design and testing where ex-non-recurring growth beats headline growth and cash flow covers profit," not the names with the biggest headline gains.

The single largest event of the day by size is 长鑫科技 (ChangXin Memory, 688825), and on it this piece corrects its own first draft. Interim report: revenue RMB 150.310 billion (+873.64%), ex-non-recurring net profit attributable to parent RMB 78.793 billion, operating cash flow RMB 131.156 billion (122% of consolidated net profit), ex-non-recurring weighted ROE 82.30%; and over the same weekend it announced the world's first mass production of LPDDR6, debuting on the Xiaomi 18 Fold. The first draft questioned the statements because "84.7% gross margin is above any period in DRAM industry history"; after verification that doubt does not hold: self-computed from the income statement (cost of sales RMB 22.934 billion ÷ revenue RMB 150.310 billion) it is indeed 84.74%, while SK Hynix's H1 gross margin was 81.63% and Micron's last two quarters 80.85%; broken down by quarter, ChangXin's Q1 79.16% against SK Hynix's Q1 79.27% is a gap of only 0.11 percentage points — two sets of books produced under different accounting standards and by different auditors land on the same point. The historical rule that "DRAM cannot have an 85% gross margin" has been voided by the peers themselves in this 2026 cycle.

But for ChangXin there are two things more worth telling than the gross margin, and both were missed in the first draft: ① 27.8% of the profit does not belong to listed-company shareholders — of RMB 107.550 billion consolidated net profit, minority interest is RMB 29.944 billion, leaving only RMB 77.605 billion attributable to parent; minority equity of RMB 136.027 billion already exceeds parent equity of RMB 134.722 billion (direct stakes in the two core subsidiaries are only 30.68% and 31.72%, with control via concert-party agreements). A model that runs PS on RMB 150.3 billion of revenue and PE on consolidated net profit will systematically overstate shareholder value. ② The true PB is 29.5x (3,978.0 ÷ 134.722), not the 19.79x shown by the quote vendor — the latter does not reconcile with any net-asset caliber. For comparison: Micron PB 10.46 and PE (TTM) 21.05, SK Hynix PE 7.26, Samsung 11.46, and the three have forward PEs compressed to 3.97–6.49x. The market is pricing the three memory giants at "cycle-peak profit × low multiple," and ChangXin at "cycle-peak profit × high multiple." These two pricings cannot both be right.

The driver type is earnings (Day 1) + policy (Day 1) on twin main lines, neither of which has been traded at all. The real-estate system reform was released after the close at 19:08 on 8/28, so today is pricing Day 1. And here the first draft missed a key document: besides the PBOC + NFRA "Opinions" (36 articles), MOHURD + the Ministry of Natural Resources + the NFRA also issued, the same day, the "Circular on Improving the Commodity Housing Sales System" (Jian Fang Gui [2026] No. 3), in which "for commodity housing projects sold on a pre-sale basis, the individual building shall have completed structural topping-out" — this raises the pre-sale threshold itself, which matters more than the credit-side "mortgages must be disbursed after completion filing." Stacking the two, developers' funding chain shifts from "foundation-level → pre-sale eligible → structural topping-out → mortgage received" to "topping-out → pre-sale eligible → completion filing → mortgage received," shifting the entire cash-flow curve right by one full "start-of-construction → topping-out" cycle. This is the institutional end of the high-turnover model. But it must not be read as "developer cash flows collapse": Article 13 of the "Opinions" simultaneously pushes the first principal repayment on development loans to after completion filing and extends development-loan tenors — this is a designed offset. The real outcome is more likely a slow ROE compression and share concentration toward central and local SOEs, not an acute rupture.

And for the two limit-up property-chain names on Friday, this piece has changed the attribution entirely — neither is a beneficiary of this policy. 我爱我家 (5i5j Holding, 000560)'s limit-up came on the first trading day after its interim report was released after the close on 8/27, and is earnings-driven (H1 net profit attributable to parent +RMB 80.8 million, +110.39% YoY, with ex-non-recurring RMB 88.0 million above the headline figure); note that it was profitable in H1 2026 — the "negative PE" shown by quote software is the TTM caliber dragged down by the Q4 2025 loss and does not mean a current-period loss. 香江控股 (Xiangjiang Holdings, 600162)'s 2 consecutive limit-ups are entirely unrelated to property: at 19:22 on 8/28 it announced that a wholly-owned subsidiary had signed a compute-power technical service contract of up to RMB 970 million, constituting a related-party transaction (the amount is 1.37x its entire H1 revenue), and on the same day the controlling shareholder pledged shares at a high level; moreover its PB of 3.07 is 9.6x that of 保利发展 (Poly Developments, 0.32) — it is by no means a "low-valuation property stock." Meanwhile, the true leaders did not move at all: 招商蛇口 (China Merchants Shekou, 001979) −0.28%, 保利发展 (Poly Developments, 600048) +0.19%, 万科 A (China Vanke A, 000002) +1.61%. The offshore market has already cast a vote on the policy: 贝壳 (KE Holdings, BEKE) rose +3.33% over the full U.S. trading day after the policy was published, versus FXI +0.77% and KWEB +0.84%, an excess of +2.56pct — the second-hand brokerage segment is the only link in the whole chain that is "written explicitly in the original text and requires no extrapolation" (Article 22 of the "Opinions" imposes no timing constraint on mortgages for re-traded homes, plus facilitation arrangements for "transfer with mortgage attached").

Pre-market state judgment: the index sits high while the sentiment structure is turning over; today will most likely gap up, but "whether the gap-up gets absorbed" matters far more than "how big the gap is." The Shanghai Composite closed 8/28 at 3,952.18 (−0.11%), only 0.96% below the 8/18 interim closing high of 3,990.30 and 1.21% below 4,000 — a gap-up today would move almost directly into the resistance zone. And one counter-signal that must be spelled out: of the 24 hottest consecutive-limit-up names on Friday, 10 issued abnormal-volatility or risk-warning announcements over the weekend, and three of them directly negate the narrative the market had assigned them: 深中华 A (China Bicycle A, 000017, 7 boards) disclosed on itself a rolling PE of 190.89x vs an industry 18.57x, an H1 gross margin of only 7.59%, and that "the company has no AI business revenue"; 千金药业 (Qianjin Pharmaceutical, 600479, 3 boards) disclosed that 22.05% of its H1 profit growth came from acquisition consolidation. This batch of announcements is the most direct coolant for today's high-flying names.

The external read is "divergence," not "broad decline," and the direction of the divergence happens to favor memory and disfavor compute. U.S. equities on 8/28: SMH −3.47%, SOXX −3.20%, NVDA −4.57%, AMAT −4.29%, while SPY was only −0.23%, QQQ −0.65%, IGV (software) −0.74% — software is the longest-duration asset yet barely fell, which shows this was not a rate-driven multiple compression; more importantly, within semis: MU −0.27%, WDC −0.55% — the memory leg held, while compute and equipment were the disaster zone. Warsh was hawkish at Jackson Hole (saying inflation shows "no meaningful improvement in the underlying trend"), and the market pushed the September hike probability from about 40% to near 50%, the direction being a hike, not a cut; but U.S. broad indices barely reacted that day, so it should not be treated as today's dominant variable. Chinese assets, by contrast, were up: KWEB +0.84%.

In one sentence: today you buy the earnings that "landed over the weekend, hold up ex-non-recurring, and are covered by cash flow," and you avoid the earnings that are "loud in the headline but negative ex-non-recurring," the consecutive-limit-up names "whose narrative the company itself just repudiated over the weekend," and the "mislabeled" property chain.

---

1. News Overview

All times are Beijing time. Items marked (weekend / post-close) fall inside the scan window and today is pricing Day 1; items marked (out-of-window / priced for N days) are old news already traded, listed to mark pricing progress, not as fresh positives.

1.1 The Pricing-Progress Dimension

1.2 Name-by-Name Verification of "Incremental vs Already Disclosed"

For key names we re-checked the Eastmoney announcement list one by one (60 entries back per name, reaching March–May 2026) to confirm there was no earnings pre-announcement or express report before the interim report:

This table is the methodological core of this piece. We have repeatedly fallen into the trap of "treating a number already public in a pre-announcement as fresh news today." 澜起科技 (Montage Technology, 688008) is this session's textbook case: the weekend interim report looks like a big event, but attributable profit of RMB 1.997 billion (+72.33%) lands precisely inside the 7/17 pre-announced range of RMB 1.9–2.1 billion (+72.55%), so there is almost no expectation gap and this piece does not put it in a recommendation slot.

---

2. Strongest Positive Branches, Descending

2.1 Why "Memory" and "Compute" Must Be Separated

Friday's U.S. session provided a falsifiable control group:

Two inferences:

1. This was not a rate-driven multiple compression. Warsh's hawkishness did push up the September hike probability, but if this were duration pricing, IGV should have fallen harder than SMH; it fell only 0.74%, and broad indices barely moved. So Friday's U.S. semiconductor decline is a sector-specific problem and cannot be used as evidence that "global risk-off will transmit to A-shares."
2. Within semis, memory held. 长鑫科技 (ChangXin Memory, 688825) belongs to memory, not compute; mapping SMH −3.47% directly onto it is wrong. ⚠️ Self-imposed limit: AMAT speaks to equipment / the supply side, and its decline does not mean memory demand is weak; MU is just one stock on one day. This item is used only to show that "a wholesale mapping is invalid"; it is not a positive argument that "memory must be strong."

2.2 Real-Estate System Reform: A Business-Model Repricing, Not a Rescue Package

This was four documents from three ministries plus the CSRC on the same day, not the single "PBOC + NFRA opinion" most reports describe:

The change to the funding chain (this is the central diagram of this section):

• Old: investment reaches 25% / foundation level → pre-sale allowed (collect down payments) → structural topping-out → mortgage received
• New: structural topping-out → pre-sale allowed (collect down payments) → completion filing → mortgage received

The entire cash-flow curve shifts right by one full "start-of-construction → topping-out" cycle. This is the institutional end of the high-turnover model, not merely a technical adjustment to mortgage timing.

⚠️ But "developer cash flows collapse" is an over-reading. Article 13 of the "Opinions" and Article 16 of the "Development Loan Measures" simultaneously provide that the first principal repayment date on development loans is in principle also after completion filing, and development-loan tenors are extended (up to 7 years for completed-home sales). The two sides are a matched design. The real outcome is "longer capital lock-up + higher peak funding needs → ROE compression + rising capital intensity," a chronic bleed rather than an acute rupture.

Comparison of the three leaders' 2026 interim reports (all disclosed):

Caliber note: A-share developers do not separately disclose pre-sale vs completed-home sales amounts; this piece did not obtain them and does not fabricate them. We use contract liabilities / inventory as a proxy: contract liabilities are cash already received on sold-but-unrecognized units (protected by contract-level grandfathering), and inventory is saleable value still to be sold. The lower the ratio = the larger the share of value not yet sold and thus exposed to the new rules.

Conclusion: 万科 A (China Vanke A) is the most impaired (cash-to-short-term-debt 0.32, H1 loss of RMB 14.951 bn, the lowest pre-sold lock-in at 22.4%, and it is being deleted from the MSCI China Index after today's close); 保利发展 (Poly Developments) is relatively advantaged (cash-to-short-term-debt 1.92, high-quality ex-non-recurring profit of RMB 1.868 bn, PB 0.32); 招商蛇口 (China Merchants Shekou) has the worst profit quality of the three — of RMB 528 mn attributable profit, RMB 465 mn is non-recurring, leaving only RMB 63 mn ex-non-recurring.

⚠️ Key judgment: under this regime, "relatively advantaged" does not equal "beneficiary." The three leaders' half-year ROEs are already compressed to 0.5%–1.0%, and the new rules demand longer capital lock-up. The institutional winners are the central and local SOEs able to secure long-term lead-bank funding, but they win by "surviving and consolidating share," not by "improving profit." This piece therefore puts no developer in the recommendation slots of §9.1.

The only beneficiary link requiring no extrapolation is second-hand housing. Article 22 explicitly states that "where the home purchased is a re-traded home, entrusted payment goes to the borrower's counterparty account" — no timing constraint on second-hand mortgages; the regulator's Q&A separately confirms "facilitation arrangements such as transfer with mortgage attached for second-hand transactions." The offshore market has already voted: BEKE rose +3.33% over the full U.S. trading day after the policy became public, an excess of +2.56pct over FXI. (Cautious phrasing: a single day's excess return cannot prove causation, but it is the hardest market evidence available.)

⚠️ One item still not obtained in this section that directly affects the conclusion: the full original text of the "Individual Housing Loan Administrative Measures (Trial)" (Jin Gui [2026] No. 5), in particular whether its supplementary provisions contain project-level transition clauses (the NFRA site was rate-limited on this visit). If those measures contain a separate project-level transition arrangement, the industry's mortgage-collection cadence over the next 1–2 years would be entirely different and this section's judgment on developers would need to be rewritten. Marked as "no reliable data available" and listed as today's most important policy detail to track. Also, the widely circulated figure "the income-to-debt-service ratio cap raised from 55% to 60%" could not be found in any first-hand source or authoritative reprint and is not cited in this piece.

---

3. Overall Single-Stock Positive-Strength Ranking (descending by name)

Boards and price limits: Shanghai/Shenzhen main board (60/00) ±10%; ChiNext (30) ±20%; STAR Market (68) ±20%; BSE (8/4) ±30%. "Ex-non-recurring quality" is the core column of this table — a qualitative read of "ex-non-recurring growth ÷ headline growth," used to detect water in headline profit. The valuation column is self-computed in this piece; quote-vendor fields are not used.

Why 沐曦股份 (MetaX, 688802) is still listed at No. 15: because nearly every financial terminal today will display it as "net profit +429% YoY," and Tencent's interface still gives it a PE (TTM) of −411.77 that has not absorbed the interim report. Listing it and spelling out that ex-non-recurring is negative is more valuable than leaving it off.

---

4. Single-Stock Scoring Model (100 points)

---

5. Detailed Analysis of the Top 10 Names

51 长鑫科技688825STAR Market ±20% | Memory | Watch closely · ChangXin Memory

• Related news: ① interim report 8/29 (Sat), original PDF¹; ② LPDDR6 mass production 8/29, Securities Times².
• Results (all from the original text, checked item by item): revenue RMB 150.310 billion (+873.64%), cost of sales only RMB 22.934 billion (+70.76%); gross margin 84.74% (self-computed: 1 − 22.934/150.310; prior-year period 13.00%); by product DDR 88.35%, LPDDR 82.60%; total profit RMB 117.145 billion; consolidated net profit RMB 107.550 billion, attributable to parent RMB 77.605 billion, ex-non-recurring attributable RMB 78.793 billion; operating cash flow RMB 131.156 billion (+2,985.64%); parent net assets RMB 134.722 billion; total assets RMB 468.078 billion; ex-non-recurring weighted ROE 82.30%.
• ⚠️ On the 84.74% gross margin — this piece corrects its own first draft. The first draft questioned the number on the grounds that "the DRAM industry's historical peak is about 60–70%." After verification that doubt does not hold:

Company	2026 Q1 gross margin	2026 Q2 gross margin	2026H1
长鑫科技 (ChangXin Memory)	79.16%	87.59%	84.74%
SK Hynix	79.27%	83.20%	81.63%
Micron (fiscal-year mismatch)	74.41%	84.56%	80.85%
Samsung (diluted by non-memory)	61.19%	69.56%	65.89%

Table scrolls horizontally

The Q1 comparison is nearly perfect: ChangXin 79.16% vs SK Hynix 79.27%, a gap of 0.11pct — two sets of books produced under different accounting standards (CAS vs K-IFRS) and by different auditors land on the same point. The historical rule that "DRAM has never had an 85% gross margin" has been voided by the peers themselves in this 2026 cycle. All four "could this be fake" channels have been checked: government grants (other income of RMB 2.180 billion sits below the gross-profit line, and asset-related grants use the gross method and do not net against fixed-asset original cost), inventory write-down reversals (3 orders of magnitude too small), depreciation policy (see below), and low-base capacity ramp (holds, and is the genuine reason). The only accounting "helper" is that depreciation lives are set at the long end: the actual annualized depreciation rate on machinery is 11.92%, close to the 8-year end of its disclosed range (5–8 years); switching to 5 years would take the gross margin only from 84.74% to 79.28%, i.e. the depreciation caliber explains at most 5.4pct — not the whole distance to 84.7%.

• Profit quality: ex-non-recurring (RMB 78.793 billion) is above the headline figure (RMB 77.605 billion), with non-recurring items at −RMB 1.188 billion; operating cash flow is 122% of consolidated net profit, and after capex free cash flow is about RMB 109.6 billion; cash plus trading financial assets of RMB 182.258 billion vs interest-bearing debt of about RMB 140.55 billion gives net cash of about +RMB 41.7 billion; liability-to-asset ratio 42.16%. Impairment charges for the period were only RMB 299.6 million (0.026% of total profit), and asset impairment losses narrowed from −RMB 796 million a year earlier to −RMB 276 million, with the notes stating that "market prices for some inventories recovered, making net realizable value higher than carrying value" — a cycle-confirmation signal, not a deterioration signal.
• ⚠️ Two items missed in the first draft that change the valuation conclusion:
  1. Minorities absorb 27.8% of the profit. Of RMB 107.550 billion consolidated net profit, minority interest is RMB 29.944 billion; minority equity of RMB 136.027 billion already exceeds parent equity of RMB 134.722 billion. The reason appears in the risk disclosures: direct stakes are only 30.68% in ChangXin Xinqiao and 31.72% in ChangXin Jidian (voting control via concert-party agreements). The company consolidates 100% of the revenue and profit while enjoying only about three-tenths of the economics; these two new production lines are ramping, and as their share of profit rises the attributable share will fall further.
  2. The true PB is 29.5x, not the quote vendor's 19.79x. Self-computed: 3,978.008 ÷ 134.722 = 29.5x (parent caliber); including minority equity it is 14.7x. Tencent's 19.79 reconciles with neither caliber and is not used.
• Valuation (all self-computed): market cap RMB 3,978.008 billion (67.884 billion shares × 58.60); PE (TTM) 48.6x; PE (H1 annualized) 25.6x; PE (Q2 annualized) 18.8x; PB 29.5x. For comparison: Micron PE (TTM) 21.05 / PB 10.46, SK Hynix 7.26, Samsung 11.46; the three have forward PEs compressed to 3.97–6.49x. Even on the most aggressive Q2-annualized 18.8x, it is still nearly 3x Micron's forward 6.49x.
• Quarterly path (derived by subtracting the Q1 statements from H1): revenue Q1 RMB 50.800 billion → Q2 RMB 99.510 billion (+95.9%); gross margin 79.16% → 87.59%; attributable profit RMB 24.762 billion → RMB 52.843 billion (+113.4%). Q2 is the entire substance of this interim report; extrapolating from Q1 would badly understate it.
• Technical sentiment and holdings: closed 8/28 at 58.60 (−0.88%), turnover rate 5.02%; +8.58% over 20 days, −5.18% from the range high. Listed 2026-07-27, about one month ago; free float is only 6.63%, amplifying two-way volatility. The greenshoe was fully exercised on 8/25, adding 1.003 billion shares, but those shares are locked for 12–36 months and do not add to near-term float (last Friday's pre-market piece described it as "adding float"; corrected here per the original announcement); the first lock-up expiry window is 12 months out.
• ⚠️ Three pieces of counter-evidence that must be stated together:
  1. PB 29.5x, while peers are being priced on "low multiples."
  2. The only remaining growth fuel is price: gross fixed-asset cost rose only +5.3% in the half, and capex of RMB 21.594 billion is −10.5% YoY; H1 volume growth came from utilization recovery and process shrink, not new capacity. Meanwhile, on TrendForce's caliber, 3Q26 server DRAM contract prices are +1318% QoQ, sharply converged from +9095% in 1Q26 — this is "converging gains," not "falling prices," and the two must be distinguished, but the second derivative has already turned. For a company with an 85% gross margin and a cost base dominated by depreciation, operating leverage works just as violently in reverse.
  3. Other registrations: the effective tax rate is only 8.19%, and the parent's high-tech enterprise qualification expired at the end of its 2023–2025 term with the company stating it "expects to complete renewal before the annual tax settlement" (an assumption not yet realized); subsidiary ChangXin Memory was placed on the U.S. Department of Defense's 1260H "Chinese military companies" list on 2026-06-08; the company has no controlling shareholder and no actual controller; finished-goods inventory went from RMB 4.595 billion to RMB 9.503 billion (+107%), while 88.8% of revenue goes through outright distributor sales, leaving channel inventory invisible.
• Final judgment: watch closely. The logic is the hardest of the day and entirely untraded; but PB 29.5x + 27.8% of profit going to minorities + a small float mean it suits "watch the absorption, not the gap." If it opens sealed at the limit or gaps up more than 8%, the odds on chasing are already poor.

52 国科微300672ChiNext ±20% | Semiconductor design | Watch closely · Nationalchip

• Results (original text, first disclosed 8/29, no July pre-announcement): revenue RMB 1.378 billion (+85.81%); attributable to parent RMB 196 million (+872.78%); ex-non-recurring RMB 175 million (+1,917.55%); operating cash flow RMB 703 million (+484.77%); gross margin 40.61%.
• Profit quality: among the best on this board. Ex-non-recurring growth (+1,917.55%) far exceeds headline growth (+872.78%), so non-recurring items are a drag rather than a contributor; operating cash flow is 3.6x net profit — what it earns is cash.
• Q2 alone: attributable profit Q1 RMB 55 million → Q2 RMB 141 million, +156.3% QoQ, accelerating rather than peaking.
• Technical sentiment: closed 8/28 at 168.18 (−1.70%), turnover rate 3.18%; +29.87% over 20 days, but −41.63% from the range high since 7/1 — it has run and it has retraced; it is not at the high. PB 9.10.
• Risk: the semiconductor sector saw net outflows of RMB 16.470 billion on Friday, the largest in the market; ChiNext's ±20% amplifies volatility.
• Judgment: watch closely. Dual verification by ex-non-recurring profit and cash flow + fully incremental + Q2 acceleration is the cleanest combination in the "earnings-driven" group.

53 圣邦股份300661ChiNext ±20% | Analog chips | Watch closely · SG Micro

• Results (original text, 8/29, no pre-announcement): revenue RMB 2.595 billion (+42.70%); attributable to parent RMB 420 million (+109.28%); ex-non-recurring RMB 356 million (+164.56%); operating cash flow RMB 382 million; weighted ROE 7.61%; total assets +69.24%, net assets +84.61% (H-share issue).
• Quality: ex-non-recurring growth is 55pct above headline growth, and operating cash flow (RMB 382 mn) exceeds ex-non-recurring net profit (RMB 356 mn) — two independent confirmations.
• Q2 alone: attributable profit Q1 RMB 124 million → Q2 RMB 297 million, +139.8% QoQ.
• Risk: PE 102.40, PB 8.03; net assets jumped 84.61% on the H-share issue, which will dilute subsequent ROE — the same structural issue as 华峰测控 (Huafeng Test & Control), except that SG Micro's ex-non-recurring growth is enough to cover the dilution and Huafeng's is not (see §6).
• Judgment: watch closely.

54 伟测科技688372STAR Market ±20% | Chip testing | Watch closely · Wintest

• Results (original text, 8/29, no pre-announcement): revenue RMB 1.072 billion (+69.00%); attributable to parent RMB 159 million (+57.61%); ex-non-recurring RMB 146 million (+171.33%); operating cash flow RMB 547 million, 3.4x net profit.
• Quality: ex-non-recurring growth is nearly 3x headline growth, the exact opposite shape to 华峰测控 (Huafeng Test & Control). Valuation is also the lowest in this group: PE 57.53, PB 4.89.
• Q2 alone: Q1 RMB 71 million → Q2 RMB 88 million, +24.8% QoQ.
• Risk: asset-heavy model, capex and depreciation pressure; its single-day −3.75% on 8/28 was the weakest among the candidates.
• Judgment: watch closely. Best value on the two dimensions of "ex-non-recurring quality + valuation."

55 万辰集团300972ChiNext ±20% | Value snacks | Watch closely · Wanchen Group

• Results (original text, 8/29, no pre-announcement): revenue RMB 34.836 billion (+54.26%); attributable to parent RMB 1.210 billion (+156.58%); ex-non-recurring RMB 1.141 billion (+153.17%); operating cash flow RMB 1.811 billion (1.5x net profit); weighted ROE 57.52%.
• Why a snack stock sits on a mostly-semiconductor board: it is one of the companies in this batch of 944 with the best simultaneous combination of "ex-non-recurring growth × ROE × low valuation," and it belongs to no hot narrative. PE (TTM) 18.42x is the only low-valuation name among the strong-earnings stocks here; it is −11.04% over 20 days, a divergence between price and earnings.
• ⚠️ Key counter-evidence: Q2 attributable profit of RMB 580 million was −7.9% QoQ from Q1's RMB 630 million — the +156.58% cumulative H1 figure masks the Q2 sequential decline, and the growth rate most likely peaked in Q1.
• Judgment: watch closely.

56 我爱我家000560Shenzhen main board ±10% | Property brokerage | Watch only · 5i5j Holding

• ⚠️ This piece corrects two items from its first draft: ① the first draft wrote "negative PE (lossmaking)" — it was profitable in H1 2026: H1 attributable profit +RMB 80.8 million (+110.39%), ex-non-recurring +RMB 88.0 million (above the headline figure), gross margin 13.05% (+3.18pct), liability-to-asset ratio down to 59.57%. The "negative PE" in quote software is the TTM caliber (a full-year 2025 loss of RMB 96.6 million, mostly in Q4). This is a company returning to profit, not a company in loss. ② The first draft attributed its limit-up to property policy — wrong: the policy was released only at 19:08, whereas it disclosed its interim report after the close on 8/27, opened 8/28 at 2.49 (+3.75%) and then ran to the limit — it was earnings-driven.
• Policy benefit logic (the cleanest link in the whole chain): second-hand brokerage contributes 59.8% of its gross profit; Article 22 of the "Opinions" imposes no timing constraint on mortgages for re-traded homes while new-home mortgages are pushed out 12+ months, a pure relative-competitiveness gain; the regulator separately confirmed "transfer with mortgage attached" facilitation.
• ⚠️ But the trend is deteriorating, and this is the most important finding: the H1 doubling of profit came 100% from stemming losses at Xiangyu (asset management) (segment gross profit −RMB 145.3 million → +RMB 40.0 million), while core brokerage revenue fell 7.68%, its gross margin fell from 22.82% to 19.55% (−3.27pct), and gross profit fell 20.9%. The company advertises "YoY growth in Beijing/Shanghai second-hand transaction counts and rising market share" — transaction growth and revenue decline holding simultaneously means average commission revenue per transaction is falling. For "40-year mortgages lifting transaction volume" to convert into profit, the decline in value per transaction must be solved first, and this policy cannot solve it.
• Position: limit-up on 8/28, +18.39% over 5 days, right at the top of its range; it already has 3 limit-ups in August and is +18.4% cumulatively from 8/5 to 8/28. Annualized ROE about 1.7%.
• Judgment: watch only. The logic is right and the direction is right, but the position is the worst and the core business is deteriorating.

5.7–5.10 Remaining Candidates (bullet form)

---

6. Pass List

The two tiers of Pass rationale (important — the two meanings are entirely different): Tier 1, "the logic does not hold" (MetaX, Huafeng Test & Control, China Bicycle A, Xiangjiang Holdings, Qianjin Pharmaceutical, Fengfan Power, Jinlong Development, Jinjian Cereals & Oils, Shangluo Electronics): the evidence shows the reason the market assigned them is itself wrong or has been denied by the company — these Passes carry high confidence. Tier 2, "position too high / already priced" (5i5j Holding, Montage Technology, Ganfeng, Yuanjie, Poly, China Merchants Shekou): the logic is fine, the price has simply already reflected it or the direction is slightly negative — this only says "this is not a good entry point," not that they will fall. Historical reconciliations repeatedly show that baskets passed on "high position" grounds still tend to outperform; do not read this as bearish.

---

7. Intra-Branch Ranking

7.1 Memory / DRAM

Hardest name: 长鑫科技 (ChangXin Memory, 688825). Who gets passed: 澜起科技 (Montage Technology) (the expectation gap has gone to zero).

7.2 Domestic Semiconductor Design / Testing (ex-non-recurring-verified)

Hardest name: 国科微 (Nationalchip, 300672). Who gets passed: 华峰测控 (Huafeng Test & Control) — the only name in this branch with "the brightest headline and the worst quality," and today's MSCI inclusion gives it extra attention, so the risk of buying it by mistake is the highest.

7.3 Property Chain (must be tiered; cannot be bought wholesale)

Hardest name: 我爱我家 (5i5j Holding, 000560) (cleanest logic, but the worst position). Most stable fundamentals: 保利发展 (Poly Developments, 600048). Who gets passed: 万科 A (China Vanke A) (policy and passive flows point in opposite directions), 香江控股 (Xiangjiang Holdings) (mislabeled; actually a compute-power related-party transaction).

---

8. Opening Verification Signals for Today

8.1 Call-Auction Signals

• 长鑫科技 (ChangXin Memory, 688825): the single most important name. A gap-up of 0–5% with moderate volume = healthy, watch the absorption; a gap-up of >8% or a sealed limit = the odds are already poor (market cap of RMB 3.98 trillion; this is not a size that seals boards on sentiment day after day). ⚠️ With a float of only 6.63%, the auction is extremely sensitive to small amounts of money; do not treat the auction gap as full-day pricing.
• 国科微 (Nationalchip, 300672) / 圣邦股份 (SG Micro, 300661) / 伟测科技 (Wintest, 688372): watch whether they can hold above the opening price after gapping up — for earnings-driven names, gapping up and fading means money is using the good news to distribute.
• Property chain: watch whether 我爱我家 (5i5j Holding, 000560) keeps sealing the limit, or 保利发展 (Poly Developments, 600048) and 招商蛇口 (China Merchants Shekou, 001979) start moving on volume. If only the already-extended theme names move while the leaders do not, it is sentiment, not policy pricing.
• High-flying consecutive-limit-up names (深中华 A China Bicycle A 000017, 锦龙股份 Jinlong Development 000712, 风范股份 Fengfan Power 601700, 千金药业 Qianjin Pharmaceutical 600479): risk announcements were issued over the weekend, so watch above all whether they gap up and fade or open down outright — this group is today's most likely "auction spike, then plunge at the open."

8.2 Sector Signals

• Whether semis can produce "3 or more fast limit-ups": if so, the weekend earnings have offset Friday's RMB 16.470 billion of net outflows; if only ChangXin rises with no supporting tier, it is a lone-runner move with weak durability.
• Whether the core holding trades on volume: whether ChangXin's turnover rate can clearly exceed Friday's 5.02%. Price up sharply without turnover expanding = a rally on no volume.
• Whether memory and compute diverge: if memory rises while compute (optical modules / PCB / servers) does not follow, §2.1 is verified; if the two move together in both directions, the market is still trading "semiconductors" as a whole and this piece's branch split should be down-weighted.
• Property-chain tier: whether 3 or more non-lossmaking names hit the limit. Only lossmaking names hitting the limit = pure theme trading.

8.3 Single-Stock Signals

• 长鑫科技 (ChangXin Memory, 688825): whether it can close above the opening price while turnover expands (turnover rate >5.02%) — the watershed between "earnings accepted" and "distribution on good news."
• 国科微 (Nationalchip, 300672): whether it can break above Friday's 3.18% turnover rate on volume; it is still 41.63% below the range high, the largest technical room in this group.
• 万辰集团 (Wanchen Group, 300972): it is −11.04% over 20 days with excellent results, so if it does not rise today, the market is pricing "the growth rate has peaked" ahead of "the earnings beat" — this verification is about the market's current pricing preference and carries a lot of information.
• 保利发展 (Poly Developments, 600048): whether it can close up on volume — the cleanest window for observing "whether the policy is genuinely being priced" (has not moved, clean ex-non-recurring profit, central SOE).
• 华峰测控 (Huafeng Test & Control, 688200): enters MSCI after today's close. If it rallies hard on "earnings + MSCI," that precisely verifies that the market did not read the ex-non-recurring line — this is the reverse verification point for whether this piece's judgment holds.

8.4 Risk Signals

• Index: the Shanghai Composite is only 0.96% below the 8/18 closing high of 3,990.30 and 1.21% below 4,000. If it gaps straight into that zone and cannot hold, a long upper shadow on the daily candle easily forms — the pattern most worth guarding against this week.
• PMI is released at 09:30 (the same minute as the open). July was 49.2%, −1.1pct MoM, and it has been below the boom-bust line continuously. If August falls further, that is a headwind for pro-cyclicals (chemicals, steel, non-ferrous) — and Friday those were precisely the sectors leading. This is today's single biggest unknown variable.
• Tail risk of earnings blowups: the weekend batch's median net-profit YoY is −5.6%, with only 48% posting growth. Confirmed non-pre-announced blowups: 泰格医药 (Tigermed, 300347) attributable −RMB 413 million (Q2 alone a loss of RMB 462 million; Q1 was still profitable), 禾迈股份 (Hoymiles, 688032) −1,107.86%, 中控技术 (Supcon Technology, 688777) −56.20%, 中信博 (Arctech Solar, 688408) −229.18%, 孚能科技 (Farasis Energy, 688567) −145.07%. None had earnings pre-announcements; today is pricing Day 1.
• *ST 萃华 (Cuihua Jewelry, 002731) failed to disclose its periodic report on time, is suspended and has flagged delisting risk — avoid the ST segment as a whole today.
• Collective cooling of high-flying names: 10 of the 24 consecutive-limit-up names issued risk announcements over the weekend, so yesterday's limit-up tier will most likely shrink today.
• Suspensions: 有研硅 (GRINM Semiconductor, 688432) and 雪天盐业 (Xuetian Salt, 600929) are suspended from today.

---

9. Final Recommendations

9.1 The 5 Names Most Worth Watching Today

⚠️ A note covering all 5: all are tagged watch closely, and none is given "priority deep-dive." Reasons: ① 4 of the 5 face the same headwind sector (semis saw net outflows of RMB 16.470 billion on Friday and SMH was −3.47%); ② August PMI is released at 09:30 today, simultaneously with the open, and July was already 49.2% with −1.1pct MoM — on a day when macro data lands at the same moment as the open, giving any name a "priority deep-dive" would underestimate the uncertainty.

⚠️ Why there is no developer in the recommendation slots: the real-estate system reform is today's second-biggest event, but its overall direction for developers is negative (pre-sale threshold raised to topping-out + mortgages pushed to completion filing), and the three leaders' half-year ROEs are only 0.5%–1.0%; the only cleanly benefiting party is second-hand agencies, and 我爱我家 (5i5j Holding, 000560) is already +18.4% and right at the top of its range while core brokerage gross profit is still −20.9%. Under this piece's unified scoring caliber, no property-chain name reaches a recommendation slot — 招商蛇口 (China Merchants Shekou, 001979)'s ex-non-recurring profit is only 12% of its headline figure, and including it would apply a double standard that contradicts this piece's own judgments on 华峰测控 (Huafeng Test & Control) and 沐曦股份 (MetaX). The property chain is handled as a "branch to watch closely," with no single-stock recommendation slot.

9.2 The 3 Strongest Branches Today

9.3 Directions Not Worth Chasing Today

1. Domestic GPU / compute concepts (especially 沐曦股份 MetaX 688802) — the "net profit +429%" headline is the opposite of the substance; ex-non-recurring attributable is −RMB 48.8595 million, and the profit comes from an unrealized mark-up the company itself calls "not sustainable"; on Friday the U.S. compute end (NVDA −4.57%, AMAT −4.29%) was clearly weaker than the memory end; and the 9/2 燧原科技 (Enflame Technology, 688801) subscription will divert money from this direction.
2. Yesterday's high-flying consecutive-limit-up names as a group — 10 of the 24 issued risk announcements over the weekend, and 深中华 A (China Bicycle A, 000017) disclosed on itself "no AI business revenue" + a 7.59% gross margin + a PE of 190.89x, while 千金药业 (Qianjin Pharmaceutical, 600479) disclosed that 22.05% of its profit growth came from consolidation.
3. Semiconductor equipment (华峰测控 Huafeng Test & Control 688200 as the representative) — headline +112.60% but ex-non-recurring only +39.06%, and ex-non-recurring ROE fell rather than rose; today's MSCI inclusion will amplify attention, making the risk of buying it by mistake the highest.
4. Mislabeled "property policy beneficiaries" — 香江控股 (Xiangjiang Holdings, 600162)'s 2 consecutive limit-ups are actually driven by an RMB 970 million compute order that is a related-party transaction, not by property policy; and its PB of 3.07 is 9.6x Poly's. 万科 A (China Vanke A, 000002) meanwhile faces passive selling pressure into the close from the MSCI deletion.
5. Grid equipment / the "six networks" concept — the catalyst is a fourth-month second fermentation of the April 28 Politburo meeting, not incremental; leader 风范股份 (Fengfan Power, 601700) already has 10 boards in 20 days and a seal-order-to-turnover ratio of only 0.065.
6. Strong-earnings names already fully priced by July pre-announcements (澜起科技 Montage Technology 688008, 赣锋锂业 Ganfeng Lithium 002460, 源杰科技 Yuanjie Semiconductor 688498) — the numbers are pretty but the expectation gap has gone to zero.
7. The ST segment — *ST 萃华 (Cuihua Jewelry, 002731) is suspended today with a delisting-risk warning, and deadline effects concentrate risk across similar names.

9.4 Final One-Sentence Judgment

Today is a day for "re-ranking the board using ex-non-recurring profit and cash flow": among the 944 interim reports that landed over the weekend, the two loudest headlines (MetaX +429%, Huafeng Test & Control +112.60%) have ex-non-recurring figures that are respectively negative and only +39%, while the names whose ex-non-recurring growth beats headline growth and whose cash flow covers profit (ChangXin, Nationalchip, SG Micro, Wintest, Wanchen) are the ones genuinely not yet priced; the real-estate system reform simultaneously raises the pre-sale threshold to topping-out and pushes mortgages to completion filing, which means ROE compression for developers rather than a positive, and of the two property names that already hit the limit one was earnings-driven and the other is fundamentally a compute-power related-party transaction — so today's correct posture is "look at ex-non-recurring profit, look at cash flow, look at who has not moved yet, look at whether the label is wrong," not "look at the gainers list and the headlines."

---

10. Pre-Publication QA Log

This section records QA and self-corrections, because five of these items directly changed conclusions.

First-draft conclusions that were overturned:

1. 华峰测控 (Huafeng Test & Control, 688200) was demoted from a recommendation slot to Pass. The first draft, relying on an aggregation database's "net profit +112.60%" and "Q2 +241.8% QoQ," listed it as one of the strongest candidates. After checking the original text: ex-non-recurring is only +39.06%, and of RMB 416 million attributable, RMB 173 million (41.6%) is non-recurring; ex-non-recurring weighted ROE fell from 4.81% to 4.24%. Because the aggregation database provides only a "net profit" field, the previously computed Q2 +241.8% QoQ was built on a contaminated caliber and has been voided.
2. 沐曦股份 (MetaX, 688802) was demoted from candidate to Pass. The anomaly was first found by arithmetic (Q2 net profit of RMB 711 million ÷ Q2 revenue of RMB 762 million = a 93.3% net margin, above its gross margin, which is operationally impossible), then confirmed from the original text as ex-non-recurring attributable of −RMB 48.8595 million; tracing further to the source, it is a Level 1 listed equity costing about RMB 107 million that has marked up to RMB 994 million — not the government grants or IPO-proceeds interest the first draft had guessed.
3. ⚠️ The doubt about ChangXin Memory's gross margin was overturned (and the correction is upward). The first draft, on the grounds that "84.7% exceeds any period in DRAM industry history (Samsung's peak 60–70%)," marked the number as "unverified, not to be used as a basis for conclusions." After verification: self-computing from the income statement gives exactly 84.74%, and SK Hynix's H1 was 81.63%, Micron's last two quarters 80.85%, with the single-quarter comparison of ChangXin Q1 79.16% vs Hynix Q1 79.27% (a gap of 0.11pct). The historical rule itself has been voided by this cycle. The three alternative explanations — government grants, inventory reversals, and depreciation — have each been ruled out (depreciation explains at most 5.4pct). This is the only record in this piece of "the doubt does not hold and the correction goes upward."
4. ChangXin Memory's PB was corrected from 19.79 to 29.5 (self-computed). The quote vendor's 19.79 reconciles with neither the parent caliber (29.5x) nor the total-equity caliber (14.7x). And two facts entirely missed in the first draft were added: minority interest is 27.8% of consolidated net profit; minority equity of RMB 136.027 billion already exceeds parent equity.
5. The attribution for both limit-up property names was entirely wrong and has been rewritten. The first draft called both "lossmaking names that rose ahead of the policy landing." In fact: 我爱我家 (5i5j Holding, 000560) had already returned to profit in H1 2026 (ex-non-recurring RMB 88.0 million above the headline RMB 80.8 million), and its limit-up was driven by the interim report released after the close on 8/27; 香江控股 (Xiangjiang Holdings, 600162) was driven by the RMB 970 million compute-power related-party order announced after the close on 8/28, unrelated to property, and its PB of 3.07 is a sector high, not a low.
6. The first draft missed an entire policy document. The MOHURD et al. "Circular on Improving the Commodity Housing Sales System" (Jian Fang Gui [2026] No. 3) raises the pre-sale condition itself to "the individual building shall have completed structural topping-out," which matters more than the credit-side adjustment to mortgage timing. The offsetting clause the first draft also omitted was added (the first principal repayment on development loans is likewise deferred to after completion filing), on which basis the conclusion was corrected from "a cash-flow negative" to "a chronic ROE-compression effect, not an acute rupture."
7. Grandfathering was corrected from "no reliable data available" to "obtained." Article 36 of the "Opinions" (anchored to the "loan contract") and section 1(3) of the sales circular (anchored to the "construction works planning permit"). And on that basis it was discovered that ifeng.com's phrasing "already obtained a pre-sale permit" is an erroneous rewrite of the original text, which has been copied and spread across multiple search summaries.

Unresolved uncertainties (listed honestly):

• The full original text of the "Individual Housing Loan Administrative Measures (Trial)" (Jin Gui [2026] No. 5) and whether its supplementary provisions contain project-level transition clauses — the NFRA site was rate-limited this time and the text was not obtained. If a separate project-level transition arrangement exists, §2.2's judgment on developers needs to be rewritten. This is the largest single-point uncertainty in this piece.
• The actual duration distribution from "structural topping-out → completion filing," and the share of mortgage receipts within developers' sales collections — neither was obtained from an authoritative source, so §2.2 makes only a directional judgment and does not quantify. Industry experience suggests a range of roughly 9–15 months, but there is no citable first-hand statistic, so no specific number is written.
• The original PDF of the earnings-projection paragraph in ChangXin's 7/23–7/24 listing announcement — at present the RMB 52.0–58.0 billion ex-non-recurring range is confirmed only second-hand, and the various paraphrases of the attributable range contradict each other (RMB 50.0–57.0 billion vs RMB 66.0–75.0 billion, the latter suspected to be a total-profit caliber). The phrase "35.8% above the ceiling" depends entirely on that document.
• The name of MetaX's RMB 994 million Level 1 equity investment and its change in market value after 6/30 — the interim report discloses neither the underlying name nor subsequent events, yet it alone constitutes 100% of MetaX's profit. Without pinning it down, its income statement cannot be modeled.
• The specific catalyst for Friday's U.S. semiconductor selloff — the search result "Nvidia's $105 billion of support for OpenAI's Ohio data center" was verified to be an 8/16–8/18 event. This piece therefore assigns no specific attribution, describing only the verifiable internal structure.
• Sector fund flows: the THS-caliber pre-close snapshot differs from Eastmoney's "main-force net inflow" caliber and cannot be subtracted across pieces; Eastmoney's clist has been 502 for a fourth consecutive trading day, so there is no second independent caliber for cross-verification.
• Q2 single-quarter data: all derived as "H1 statements − Q1 statements," and computed on attributable net profit (the aggregation-database caliber) rather than ex-non-recurring — so they are unusable for companies with volatile non-recurring items (Huafeng Test & Control, MetaX), and this piece stops using them at the relevant points.

---

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