Aluminum Market Daily  ·  Metals Desk铝早报  ·  金属研究

Weekend: August CPI sends FedWatch hike odds to 90% and LME cash falls $69. Alumina breaks below all six averages.周末版:美国 8 月 CPI 令 FedWatch 加息概率升至 90%,LME 现货下跌 69 美元;氧化铝跌破全部六条均线。

Saturday, September 12, 2026
Morning Meeting晨会参考
Written the morning of Saturday September 12 (Asia), with BOTH markets closed for the weekend. ⚑⚑ WHAT IS NEW THIS EDITION: the complete SHFE September 11 trading day for all twelve aluminium and all twelve alumina contracts (open, high, low, close, settlement, volume, open interest), retrieved from a 224-row daily series whose September 10 values reconcile to the tick against last edition's print; ⚑⚑ the COMPLETE September 11 NIGHT SESSION for six aluminium and five alumina contracts, whose eleven prior-settlement fields reconcile EXACTLY against the September 11 day settlements; ⚑⚑ ALL SIX SHFE averages exact for a SECOND consecutive edition, and — for the FIRST time on this ledger — ALL SIX ALUMINA averages exact and LIVE on the AO2701 main-contract series, ending the four-observation-stale September 4 carry; ⚑⚑ the Westmetall September 11 row, so all six LME averages are exact for a FOURTH consecutive edition; ⚑⚑ the US August CPI RESULT, the number this brief published a band for in advance, and the post-CPI FedWatch move; ⚑⚑ the CNAL five-caliber regional spot set for September 11, which RESUMES after a single-edition break; ⚑ a third-party SMM mirror for September 11 whose arithmetic RETROACTIVELY PROVES the September 10 mirror rows were real quotations and not carried rows; the September 11 DXY and Brent settlement; and China's August trade and aluminium export data. ⊘ NOT RETRIEVABLE: SMM's own primary September 11 spot quotation set (login-walled for a TWELFTH consecutive edition); SMM's three-region daily inventory series, a SIXTH consecutive edition; the SHFE Friday weekly warehouse and warrant print in its primary caliber (a third-party −3% w/w note is listed and flagged); any AD (cast aluminium alloy) contract data; SMM's regional transaction premia, shipment and purchasing sentiment indices and Chalco's five regional alumina listings; LME volume and open interest; a dated Chinese alumina spot index for September 11; and the October fed funds strip, absent for a FIFTEENTH consecutive edition. ⚑ NOTE ON CALIBER: the SHFE trading day dated September 11 BEGINS with the 21:00 September 10 night session, so the September 11 open (24,310), high (24,315) and low (24,075) are the night ticks this brief already published one edition ago. The daytime portion is reported separately and explicitly throughout.
Steven Yuan
⚑⊘ SEVEN TESTS, SEVEN RESOLUTIONS — THE FULLEST GRADING ON THIS LEDGER Every one of the seven watchlist items written down for September 11 resolved inside a single session: FIVE bearish (open interest, alumina, LME MA5, the CPI reaction, the alumina OI direction), ONE split (the 24,075 level held to the tick while the MA10 close confirmation failed by 44.50), and ONE — the CPI band — where the number landed INSIDE the band and the market reacted as though it had not. Nothing was carried unresolved
⊘⊘ US AUGUST CPI: THE NUMBER PASSED THE BAND AND THE REACTION FUNCTION FAILED IT Headline +0.4% m/m and +3.4% y/y — the pre-written band was '+3.3% to +3.4% in-band, above +3.5% bearish, below +3.2% bullish'. It landed at the top of the in-band range. But CORE printed +0.3% m/m against July's +0.2% (core y/y eased to +2.4% from +2.5%), gasoline rose +3.9% and accounted for over a third of the monthly gain, and FedWatch went from ~70% to ~90% for a September 16 HIKE in one afternoon
⊘ SO THE TEST IS RECORDED AS A METHOD DEFECT, NOT A PASS This brief keyed the test to the headline y/y. The market keyed it to the core m/m and to the composition. The in-band headline produced the out-of-band reaction, which means the mapping — not the threshold — was wrong. Recorded, and the CPI/PPI tests are re-specified from the next edition to key on core m/m and on the FedWatch move itself rather than on the headline (derivation)
⊘⊘ FEDWATCH: ~69.3% PRE-CPI → ~90% POST-CPI FOR A SEPTEMBER 16 HIKE Against the 56% / 59% / 60.6% three-way caliber spread this brief carried all week. EY-Parthenon switched its call from HOLD to a 25 bp HIKE, which would take the target range to 3.75–4.00%. This is the single largest one-day repricing of the Fed path on this brief's ledger and it is unambiguously bearish for aluminium
⚑ AND THE DOLLAR'S FOUR-SESSION DECLINE ENDED — BUT ONLY JUST DXY 99.095 at September 11, +0.047 (+0.05%), holding above 99 for the first time in five sessions. Month −0.92%. ⚑ HONEST COUNTERWEIGHT: a +0.05% close on a day the Fed path repriced 20 points is a REMARKABLY small dollar move, and it is the clearest evidence on this page that the yen/BoJ channel this brief has named for five editions is still the dominant driver (interpretation)
⊘⊘ LONDON TOOK THE ENTIRE MACRO HIT — THE LARGEST CASH FALL ON THIS LEDGER LME cash settled $3,274.00, −$69.00 (−2.06%), the biggest one-day cash decline this brief has recorded. The 3-month settled $3,250.00, −$68.00. Cash LOST MA5 ($3,320.80) by $46.80 and MA10 ($3,293.80) by $19.80, and is holding MA20 ($3,256.05) by $17.95 and MA30 ($3,263.48) by $10.52
⚑⊘ THE BACKWARDATION NARROWED FOR THE FIRST TIME IN FIVE — BY ONE DOLLAR +$24.00, against the progression −$0.50 → +$2.00 → +$6.00 → +$14.00 → +$25.00 → +$24.00. The prompt tightness did NOT unwind with the price: a $69 fall in cash cost the structure exactly one dollar. That is the strongest single piece of evidence on this page that the physical prompt market is not what sold off (interpretation)
⚑ ALL SIX LME AVERAGES EXACT FOR A FOURTH CONSECUTIVE EDITION MA5 $3,320.80, MA10 $3,293.80, MA20 $3,256.05, MA30 $3,263.48, MA40 $3,244.04, MA60 $3,217.94. ⊘ The non-monotone stack WIDENED a third step: MA30 is now $7.43 above MA20, against $5.75 and $2.99. ⚑ The September 10 recomputation returns $3,324.50 / $3,287.65 / $3,256.30 / $3,262.05 / $3,241.44 / $3,220.07 — matching last edition to the cent on five of six and to one hundredth on the sixth (derivation)
⚑⚑ SHFE HELD THE NAMED LEVEL EXACTLY — 24,075, NOT ONE POINT LOWER The pre-written test was: 'hold 24,075 and close above MA10 (24,279.50) and the night was oversold; break 24,037.50 and close below it and the sixty-session structure is broken.' The September 11 trading-day low is 24,075 — the SAME TICK as the September 10 night low and the SAME TICK as the August 31 close. The support half of the test passed perfectly
⊘ BUT THE CONFIRMATION HALF FAILED BY 44.50 POINTS AL2610 closed 24,235. The test required a close above 24,279.50. It missed by 44.50 — and against the LIVE MA10 of 24,311.00 it is 76.00 below. So the honest ruling is SPLIT: the level is intact, the recovery is not confirmed, and neither the 'oversold' nor the 'structure broken' branch can be declared (derivation)
⊘ THE SEPTEMBER 11 TRADING DAY IN FULL AL2610: open 24,310, high 24,315, low 24,075, close 24,235, settlement 24,190, −385 (−1.57%) against the 24,575 prior settlement; close-to-close −255 (−1.04%). Volume 213,873 (+69,493, +48.1%), open interest 211,390 (−25,055, −10.60%). ⚑ CALIBER NOTE: the open, high and low are the September 10 night ticks — the SHFE trading day begins at 21:00 the previous evening
⚑ THE DAYTIME PORTION WAS A RECOVERY, AND IT IS RECORDED SEPARATELY The September 10 night session ended at 24,125. The September 11 close is 24,235. So the daytime hours bought back +110 of the night's −450, roughly 24.4% of it, and the session finished 160 above its low and only 80 below its high. A 1.04% down day whose entire loss was made in the night before the day session opened (derivation)
⊘ TWELVE OF TWELVE ALUMINIUM SETTLEMENTS FELL, BY 315 TO 400 AL2609 −350 / AL2610 −385 / AL2611 −385 / AL2612 −380 / AL2701 −385 / AL2702 −400 / AL2703 −395 / AL2704 −395 / AL2705 −370 / AL2706 −365 / AL2707 −355 / AL2708 −315. A tight 85-point band across twelve months is a parallel shift, not a front-end event. ⚑ But only THREE of twelve closed below their opens' mirror — AL2609, AL2704 and AL2708 closed ABOVE their opens, against ZERO the session before (derivation)
⚑⚑ ALL SIX SHFE AVERAGES EXACT FOR A SECOND EDITION — AND THE SERIES SELF-VERIFIES MA5 24,433.00, MA10 24,311.00, MA20 24,055.00, MA30 24,029.83, MA40 23,847.25, MA60 23,601.25. Recomputed at September 10 the same 224-row series returns 24,444.00 / 24,279.50 / 24,037.50 / 24,009.33 / 23,821.50 / 23,599.08 — EXACTLY the six numbers published last edition. The series is verified, not asserted (derivation)
⊘ THE CLOSE LOST MA5 AND MA10 AND IS HOLDING MA20 BY 180.00 24,235 sits 198.00 below MA5, 76.00 below MA10, and 180.00 above MA20 (24,055.00), 205.17 above MA30, 387.75 above MA40, 633.75 above MA60. Two of six averages are now lost on the CLOSE, against zero on September 10 (derivation)
⊘⊘ AND THE SEPTEMBER 11 NIGHT SESSION TOOK MA20 AND MA30 AL2610 night: open 24,100, high 24,100, low 23,920, last 24,000, −190 (−0.79%) against the 24,190 settlement. 24,000 sits 55.00 BELOW MA20 (24,055.00) and 29.83 BELOW MA30 (24,029.83), and is holding MA40 (23,847.25) by 152.75. FOUR of six averages are now lost on the night caliber
⊘ THE NIGHT LOW OF 23,920 IS THE LOWEST TICK SINCE SEPTEMBER 1 23,920 is below every close from September 1 onward (24,110 / 24,100 / 24,370 / 24,290 / 24,420 / 24,460 / 24,560 / 24,490 / 24,235) and is EXACTLY the August 28 close. The board has given back the whole of the September advance on an intraday basis, while still holding it on a closing basis (derivation)
⚑ BUT THE NIGHT WAS A ROLL, WHICH IS THE OPPOSITE OF THE DAY Across six contracts, day 560,984 → night 559,372, only −1,612. AL2610 fell 12,438 while AL2611 rose 5,421, AL2612 3,601, AL2701 1,851 and AL2702 258. Position MIGRATED down the curve rather than leaving. This is the exact inverse of the September 10 night's outright −20,436 (derivation)
⚑⚑ AND IN THAT NIGHT AL2611 OVERTOOK AL2610 ON OPEN INTEREST Night OI: AL2611 213,268 against AL2610 198,952 — AL2611 is now 14,316 LARGER. On the day session AL2610 still led by 3,543 (211,390 vs 207,847), down from 24,734 on September 10. The main-contract handover to AL2611 is underway and should complete within days; this brief will switch its main-contract caliber when the DAY session confirms it (derivation)
⊘⊘ THE ROLL RULING IS DEAD — AT TWICE THE THRESHOLD THIS BRIEF WROTE DOWN The pre-written test: '±5,000 complex OI is still a roll; −15,000 or more voids the roll explanation and it becomes EXIT.' Complex open interest fell 29,349 to 593,718, −4.71%. That is 1.96x the void threshold. The roll explanation carried for two editions is formally WITHDRAWN
⊘⊘ AND THE THREE-SESSION ARITHMETIC INVERTS COMPLETELY September 8 → September 11: main 258,395 → 211,390 = −47,005; complex 628,099 → 593,718 = −34,381. So ~12,624 rolled and ~34,381 LEFT. Two editions ago the split was ~16,900 roll / ~5,000 exit. Exit now outweighs roll 2.7 to 1, having been outweighed 3.4 to 1. This is the single largest reversal of an interpretation on this brief's ledger (derivation)
⊘ VOLUME CONFIRMS IT WAS A REAL EVENT, NOT A DRIFT Main volume 213,873, +69,493 (+48.1%); complex volume 414,471, +147,820 (+55.4%), the heaviest on this brief's ledger. Heavy volume plus heavy open-interest reduction plus a falling price is distribution, and it is recorded as such (interpretation)
⊘ THE CLOSE FELL TO THE 70TH AND 85TH PERCENTILES 24,235 sits at the 70th percentile of the twenty-session close distribution and the 85th of the sixty-session, against 95th and 98.3rd on September 10 and 100th/100th on September 9. Within the twenty-session range (23,595–24,560) the close sits at 66.3%; within the sixty-session range (22,445–24,560) at 84.6% (derivation)
⊘⊘ ALUMINA BROKE, AND IT BROKE ON RISING OPEN INTEREST AO2701: open 2,762, high 2,792, low 2,691, close 2,697, settlement 2,731, −39 (−1.41%) on settlement and −61 (−2.21%) close-to-close. Volume 275,018 (+112,279, +69.0%). Open interest 220,066, +24,675 (+12.63%). ⊘⊘ Complex OI 462,497, +20,900 (+4.73%), against −1,830 the session before. A falling price on rising position is NEW SHORTS, not long liquidation, and that is a different and more durable signature (interpretation)
⚑⚑ AND FOR THE FIRST TIME ON THIS LEDGER ALL SIX ALUMINA AVERAGES ARE LIVE Computed on a 160-row AO2701 series: MA5 2,757.00, MA10 2,753.10, MA20 2,737.15, MA30 2,738.67, MA40 2,749.22, MA60 2,781.45. The four-observation-stale September 4 AO2610 carry that this brief flagged for five editions is RETIRED
⊘⊘ AND AO2701 CLOSED BELOW ALL SIX OF THEM AT ONCE 2,697 is 60.00 / 56.10 / 40.15 / 41.67 / 52.22 / 84.45 below MA5 through MA60. A close below the entire average stack in a single session, from a position above MA5 the day before, is the cleanest trend break on this page — and the night took it a further 24 points lower to 2,673 (derivation)
⊘⊘ COST CALIBER ONE FAILED A SECOND TIME — AND BY SEVENTEEN TIMES THE MARGIN AO2610 closed 2,667, which is 51 BELOW the 2,718 line. The clearance chain reads +39 (Sep 4) → +19 (Sep 8) → +2 (Sep 9) → −3 (Sep 10) → −51 (Sep 11). The pre-written condition was 'a second consecutive close below confirms the downgrade.' It is CONFIRMED, and the settlement of 2,695 is 23 below the line as well, so both calibers now fail
⊘⊘ COST CALIBER TWO IS NOW UNPRICED ON ALL TWELVE CONTRACTS The first close above 2,826.21 was AO2704 on September 9 and AO2705 on September 10. On September 11 there is NO SUCH CONTRACT: the highest close on the whole curve is AO2708 at 2,815, which is 11.21 BELOW the line. Forward cost inflation is no longer priced anywhere on the alumina board out to August 2027 (derivation)
⊘ TWELVE OF TWELVE ALUMINA CONTRACTS CLOSED BELOW THEIR OPENS Against eleven of twelve on September 10. The complete September 11 close curve: AO2609 2,680 / AO2610 2,667 / AO2611 2,684 / AO2612 2,694 / AO2701 2,697 / AO2702 2,710 / AO2703 2,730 / AO2704 2,757 / AO2705 2,769 / AO2706 2,796 / AO2707 2,806 / AO2708 2,815. One inversion, at the expiring AO2609/AO2610 front pair; monotone from AO2610 outward (derivation)
⊘ AND THE CONTANGO KEPT NARROWING Front-to-tail (AO2610 → AO2708) +148, or 5.55% of the front, against +175 and 6.51% on September 10 and +184 and 6.83% on September 9. Three consecutive narrowings. A flattening forward curve on a falling front is the market pricing the surplus forward rather than the cost floor (interpretation)
⊘ THE ALUMINA NIGHT CONFIRMED IT AND KEPT ADDING POSITION AO2701 2,673 (−58, −2.12%) on 92,893 lots with open interest 235,026, a FURTHER +14,960 on the day. AO2610 2,651 (−44, −1.63%), AO2611 2,659 (−45, −1.66%), AO2612 2,672 (−39, −1.44%), AO2702 2,690 (−55, −2.00%). Five of five down between 1.44% and 2.12%, and position still building
⚑⊘ THE MARGIN PROXY: COMPRESSION CONTINUES, BUT THE CURVE COMPRESSION COLLAPSED Matched pair AL2610/AO2610 at the stated 1.93 t/t assumption: 24,235 − 1.93 × 2,667 = 19,087.69, against 19,250.05 — a further −162.36. ⚑ But the day-curve compression (front AL2609/AO2609 19,067.60 minus tail AL2708/AO2708 18,827.05) fell to 240.55 from 402.75, a −162.20 narrowing and the largest on this ledger. The smelter margin is falling in level and flattening in shape at the same time (derivation)
⚑⚑ THE SPOT TEST RESOLVED BULLISH ON ITS MIDDLE BRANCH — AND CLEARED THIS BRIEF'S OWN FLAG The pre-written bands: above 24,400 = spot did not follow (bullish); 24,150–24,300 = spot lagged one day; below 24,100 = spot led lower (bearish). SMM A00 printed 24,240 — squarely in the middle band. ⚑⚑ AND 24,560 − 320 = 24,240 EXACTLY, which retroactively PROVES the September 10 mirror rows were real quotations. The 'three zero-change rows look like a carry' flag is WITHDRAWN and the +70 basis published for September 10 STANDS
⚑ THE BASIS CONVERGED TO +5 — ALMOST EXACTLY THE NAMED BAND 24,240 − 24,235 = +5 against the close; +50 against the 24,190 settlement; +240 against the 24,000 night last. The pre-written normal band was '0 to −50 against the close.' +5 is five points above its top edge. Chain: +70 (Sep 4) → +55 (Sep 7) → −20 (Sep 8) → −20/0 (Sep 9) → +70 (Sep 10) → +5 (Sep 11) (derivation)
⚑ AND THE CNAL FIVE-CALIBER SET RESUMED AFTER A SINGLE-EDITION BREAK September 11: Shanghai 24,210–24,250 avg 24,230 (−310); Changjiang 24,220–24,260 avg 24,240 (−320); Zhongyuan 24,240 (−320); Nanhai/Foshan 24,660–24,760 avg 24,710 (−340); Guangdong Nanchu 24,410–24,470 avg 24,440 (−360). Both spot sources are live again in the same edition
⊘ BUT EVERY REGIONAL PREMIUM NARROWED, AND SOUTH CHINA LED THE FALL Nanhai − Shanghai 480, against 530 on September 9; Nanchu − Shanghai 210, against 260; Changjiang − Shanghai +10, unchanged. South China fell 340–360 against East China's 310–320, so the region that had been widening for four sessions gave the most back. The regional-tightness leg is downgraded (interpretation)
⊘ LME STOCKS: A SECOND NON-ZERO ROW, AND THE NEW THRESHOLD WORKED FIRST TIME 244,100 mt at September 11, −250. Last edition wrote the substance threshold explicitly — 'a single draw of −500 mt or more is required' — precisely to avoid repeating the −175 debacle. −250 is a three-figure draw and is BELOW the threshold, so it is correctly classified as noise without any argument. The method fix is validated on its first use (derivation)
⊘ AND THE LONDON FLOW IS STILL EFFECTIVELY FLAT SIXTEEN-INTERVAL FLOW: August 18 246,925 → September 11 244,100 = −2,825 mt (−1.14%), with TEN of seventeen intervals exactly zero. SIXTY-SESSION: June 17 316,525 → 244,100 = −72,425 mt (−22.88%), but ~166 mt per interval since August 18 against ~1,582 before. The stock row is not what is moving this market (derivation)
⚑⚑ THE RATIO REVERSED HARD — SHANGHAI OUTPERFORMED BY 1.02 POINTS 24,235 ÷ $3,274.00 = 7.4023, against 7.3257, 7.3270 and 7.3564. SHFE −1.04% close-to-close against LME cash −2.06%: Shanghai out-performed by 1.02 percentage points and ended three consecutive sessions of London outperformance. The macro shock was priced in London, not in Shanghai (interpretation)
⊘⚑ BRENT SETTLED $104.61, −2.8%, BUT THE WEEK WAS +8.7% The caliber conflict this brief carried — $101.25 against $105.71 for September 10 — resolves in favour of the HIGHER row: a −2.8% settlement at $104.61 implies roughly $107.6 the session before. ⚑ The $101.25 caliber is retired as a different contract month. Brent has now held above $100 for three sessions; the week's +8.7% came on Middle East risk and Friday's fall on diplomatic headlines (derivation)
⊘ INVENTORY: NO NEW NATIONAL PRINT, SO 796,000 MT IS CARRIED SMM publishes Monday and Thursday; the next national ingot print is Monday September 14. The carried chain is Aug 24 860,000 → Aug 27 852,000 → Aug 31 837,000 → Sept 3 815,000 → Sept 7 802,000 → Sept 10 796,000, with Mysteel at 768,000 on a constant 28,000 mt definitional offset. The deceleration ruling — the destock halved rather than stopped — stands unchanged into Monday
⊘ AND THE THREE-REGION BREAKDOWN IS MISSING FOR A SIXTH CONSECUTIVE EDITION So this brief still cannot say WHERE the deceleration happened, which is the question that would distinguish an East China arrival cluster from a genuine demand stall. The SHFE Friday weekly warrant print was again not retrievable in its primary caliber; a third-party note has SHFE aluminium stocks −3% w/w and is listed WITHOUT weight
⚑⊘ CHINA'S AUGUST TRADE WAS STRONG AND THE ALUMINIUM LINE WAS STRONGER Exports +25% y/y to $401.44bn, imports +28.2% to $282.36bn, surplus $119.09bn. Unwrought aluminium and product exports 626,000 mt in August, +17.2% y/y; January–August 4.67 million mt, +16.7% y/y. ⊘ The counterweight: an export line growing 16.7% into a world adding tariff barriers is a concentration risk, not only a demand signal (interpretation)

一 · 行情表现

前置声明:写稿时点、本期性质与口径说明

  • 本期写于 9月12日(周六)上午,沪伦两市均已休市。
  • ⚑⚑ **本期新增:9月11日沪市全部十二个铝合约与十二个氧化铝合约的开、高、低、收、结算、成交量与持仓量(取自 224 行日度序列,其 9月10日数值与上期公布值逐点吻合);⚑⚑ 9月11日完整夜盘(六个铝合约、五个氧化铝合约),其十一个前结算价字段与 9月11日结算价逐一精确对上;⚑⚑ 六条沪铝均线连续第二期精确,且本台账首次实现六条氧化铝均线全部精确并转为实时(基于 AO2701 主力序列,终结沿用四期的 9月4日 AO2610 陈旧基准);⚑⚑ Westmetall 9月11日行,LME 六条均线连续第四期精确;⚑⚑ 美国 8月 CPI 实际值——即本简报上一期事先公布判别带的那个数字——以及 CPI 后的 FedWatch 变动;⚑⚑ CNAL 五口径区域现货 9月11日数据,中断一期后恢复;⚑ 第三方 SMM 镜像口径 9月11日数据,其算术回溯性地证明 9月10日的镜像行是真实报价而非沿用行;9月11日 DXY 与 Brent 结算价;中国 8月进出口与铝材出口数据。**
  • ⊘ **无法取得:SMM 自身 9月11日一手现货报价(连续第十二期登录墙);SMM 三地日度库存,连续第六期;上期所周五周度仓单与库存的一手口径(第三方「周环比 −3%」的说法列示并加注);9月11日任何铸造铝合金(AD)合约数据;SMM 区域成交升贴水、出货与采购意愿指数及中铝五地挂牌价;LME 成交量与持仓量;有明确日期的 9月11日中国氧化铝现货指数;10月联邦基金利率条,连续第十五期缺失。**
  • ⚑ **口径说明(重要):上期所 9月11日交易日自 9月10日 21:00 夜盘开始,故 9月11日的开盘 24,310、最高 24,315、最低 24,075 即为本简报上一期已公布的夜盘价位。本期全文将日间时段单独列出并明确标注,不与交易日口径混用。**

⚑⚑ 点名的 24,075 守得分毫不差——而确认条件差 44.50 点未达成

  • ⚑⚑ **事先写下的检验原文:「日盘守住 24,075 并收在 24,279.50(MA10) 之上,则夜盘属超跌;下破 24,037.50 并收于其下,则六十日上升结构被实质破坏」。**
  • ⚑⚑ **9月11日交易日最低价为 24,075——与 9月10日夜盘低点同一个价位,一点未破,且恰为 8月31日收盘价。支撑一半完美通过。**
  • ⊘ **确认一半失败,差 44.50 点:AL2610 收 24,235,低于检验要求的 24,279.50;对照实时 MA10(24,311.00)则低 76.00。**
  • ⊘ **故本期判定为「各半」:位置完好,反弹未获确认,两个分支均不可宣告(derivation)。**

⊘ 9月11日交易日全貌,以及必须单独记录的日间回升

  • ⊘ **AL2610 9月11日(交易日口径):开 24,310,高 24,315,低 24,075,收 24,235,结算 24,190,较 24,575 前结算 −385(−1.57%);收盘对收盘 −255(−1.04%)。成交量 213,873(+69,493,+48.1%),持仓量 211,390(−25,055,−10.60%)。**
  • ⚑ **日间时段是回升,本期与失败同等记录:9月10日夜盘收在 24,125,9月11日收在 24,235,即日间时段买回了夜盘 450 点跌幅中的 110 点(约 24.4%);全日收盘高于最低点 160、仅低于最高点 80。这 1.04% 的跌幅全部发生在日盘开盘之前(derivation)。**
  • ⊘ **十二个铝合约结算价全部下跌,幅度 −315 至 −400:AL2609 −350 / AL2610 −385 / AL2611 −385 / AL2612 −380 / AL2701 −385 / AL2702 −400 / AL2703 −395 / AL2704 −395 / AL2705 −370 / AL2706 −365 / AL2707 −355 / AL2708 −315。85 点的带宽横跨十二个月份,是平行下移而非前端事件。**
  • ⚑ **但仅三个合约收在自身开盘价之下的反面成立:AL2609、AL2704、AL2708 收在开盘价之上,而 9月10日为零。十二个合约全部收在自身最低价之上(derivation)。**

⚑⚑ 六条沪铝均线连续第二期精确,且序列自我验证

  • ⚑⚑ **MA5 24,433.00 / MA10 24,311.00 / MA20 24,055.00 / MA30 24,029.83 / MA40 23,847.25 / MA60 23,601.25。**
  • ⚑⚑ **同一 224 行序列在 9月10日重算得 24,444.00 / 24,279.50 / 24,037.50 / 24,009.33 / 23,821.50 / 23,599.08——与上期公布的六个数字完全一致。序列经验证而非断言(derivation)。**
  • ⊘ **收盘价 24,235 失守 MA5 与 MA10:低于 MA5 198.00、低于 MA10 76.00;高于 MA20 180.00、高于 MA30 205.17、高于 MA40 387.75、高于 MA60 633.75。收盘口径下已有两条均线失守(9月10日为零条)。**
  • ⊘ **收盘分位继续下移:24,235 位于二十日收盘分布的 70 百分位、六十日的 85 百分位(9月10日为 95 与 98.3,9月9日为 100 与 100)。在二十日区间(23,595–24,560)中位于 66.3%,在六十日区间(22,445–24,560)中位于 84.6%(derivation)。**

⊘⊘ 9月11日夜盘拿走 MA20 与 MA30

  • ⊘⊘ **AL2610 夜盘:开 24,100,高 24,100,低 23,920,最新 24,000,较 24,190 结算价 −190(−0.79%)。成交 81,993,持仓 198,952(−12,438)。**
  • ⊘⊘ **24,000 低于 MA20(24,055.00)55.00、低于 MA30(24,029.83)29.83,高于 MA40(23,847.25)152.75。夜盘口径下六条均线已失守四条;仅 MA40 与 MA60 完好。**
  • ⊘ **夜盘低点 23,920 为 9月1日以来最低价位,低于 9月1日以来全部收盘价(24,110 / 24,100 / 24,370 / 24,290 / 24,420 / 24,460 / 24,560 / 24,490 / 24,235),且恰为 8月28日收盘价。盘中口径下 9月的全部涨幅已被抹去,收盘口径下尚未(derivation)。**
  • ⊘ 夜盘六合约:AL2609 **23,970(−235,−0.97%)/ AL2610 24,000(−190,−0.79%)/ AL2611 24,000(−180,−0.74%)/ AL2612 23,990(−180,−0.74%)/ AL2701 23,955(−190,−0.79%)/ AL2702 23,950(−190,−0.79%)。**
  • ⚑ **但夜盘是移仓,与日盘性质相反:六合约日盘合计 560,984 → 夜盘 559,372,仅 −1,612;其中 AL2610 −12,438,而 AL2611 +5,421、AL2612 +3,601、AL2701 +1,851、AL2702 +258。持仓沿曲线迁移而非离场,与 9月10日夜盘 −20,436 的净减仓完全相反(derivation)。**
  • ⚑⚑ **且在同一夜盘中 AL2611 持仓反超 AL2610:213,268 对 198,952,AL2611 已多出 14,316;日盘口径下 AL2610 仍领先 3,543(211,390 对 207,847),而 9月10日领先 24,734。主力换月正在进行,本简报将在日盘确认后切换主力口径(derivation)。**

⊘⊘ 「移仓」的定性作废——且是在本简报自设阈值的两倍处作废

  • ⊘⊘ **事先写下的判别带:「全系持仓变动在 ±5,000 以内仍视为移仓;全系 −15,000 以上则移仓解释作废,应改记为离场」。**
  • ⊘⊘ **全系持仓 593,718,−29,349(−4.71%),为作废阈值的 1.96 倍。沿用两期的「移仓」解释正式撤销。**
  • ⊘⊘ **三期算术完全反转:9月8日 → 9月11日,主力 258,395 → 211,390 = −47,005;全系 628,099 → 593,718 = −34,381。即约 12,624 手移仓、约 34,381 手离场。两期之前的同一计算给出的是 16,900 移仓对 5,000 离场。离场以 2.7 比 1 压倒移仓,而此前是被 3.4 比 1 压倒——为本台账上最大的一次定性反转(derivation)。**
  • ⊘ **成交量证明这是事件而非漂移:主力 213,873(+48.1%),全系 414,471(+147,820,+55.4%),为本台账最重。巨量 + 大幅减仓 + 价格下跌 = 派发,如实记录(interpretation)。**

⊘⊘ 伦敦:本台账最大单日现货跌幅,但近端结构只让了一美元

  • ⊘⊘ 现货结算 **3,274.00 美元(−69.00,−2.06%),为本简报记录以来最大单日现货跌幅;三个月结算 3,250.00 美元(−68.00,−2.05%)。**
  • ⊘ **现货失守 MA5 与 MA10:低于 MA5(3,320.80)46.80、低于 MA10(3,293.80)19.80;高于 MA20(3,256.05)17.95、高于 MA30(3,263.48)10.52、高于 MA40(3,244.04)29.96、高于 MA60(3,217.94)56.06。MA20 的 17.95 是本轮以来最窄的支撑余量。**
  • ⚑⊘ **backwardation 五期以来首次收窄,且只收窄 1.00 美元:−0.50 → +2.00 → +6.00 → +14.00 → +25.00 → +24.00。现货跌 69 美元,近端结构只让了 1 美元——这是本页最有力的一条证据:周五抛售铝的不是实货近端市场(interpretation)。**
  • ⚑ **六条 LME 均线连续第四期精确;9月10日重算得 3,324.50 / 3,287.65 / 3,256.30 / 3,262.05 / 3,241.44 / 3,220.07,六条中五条与上期逐分吻合、第六条相差 0.01(derivation)。**
  • ⊘ **均线倒挂第三次扩大:MA30 已高于 MA20 7.43 美元(前值 5.75、2.99)。**
  • ⚑⚑ **比值大幅反转,沪市跑赢 1.02 个百分点:24,235 ÷ 3,274.00 = 7.4023(前值 7.3257、7.3270、7.3564)。沪铝 −1.04% 对伦铝现货 −2.06%,终结伦敦连续三期的跑赢。宏观冲击在伦敦定价,在上海基本被拒绝(interpretation)。**

二 · 铝价及相关市场

⚑⚑ 现货检验落在中间分支——且顺带撤销了本简报自己的一个错误怀疑

  • ⚑⚑ **事先写下的三条带:A00 在 24,400 以上 = 盘面跌而现货不跟随(偏多);24,150–24,300 = 现货滞后一日跟随;24,100 以下 = 现货领跌(偏空)。**
  • ⚑⚑ **SMM A00 9月11日报 24,240(−320),正落在中间带的中央。现货滞后盘面一日跟随,既未领跌也未拒绝跟随。**
  • ⚑⚑ **算术回溯性地证明上期数据为真:24,560 − 320 = 24,240;佛山 24,780 − 350 = 24,430;ADC12 24,450 − 150 = 24,300——三项全部精确吻合。9月10日的镜像行是携带真实零变动的真实报价,不是沿用行。上期「三个零变动行疑似沿用」的标记予以撤销,9月10日 +70 的基差读数维持,而该项怀疑如实记录为「错误」。**
  • ⚑ **基差收敛至 +5:24,240 − 24,235 = +5(对收盘);对结算价 24,190 为 +50;对夜盘最新 24,000 为 +240。事先写下的常态带为「对收盘 0 至 −50」,+5 仅高出其上沿 5 点。链条:+70(9/4)→ +55(9/7)→ −20(9/8)→ −20 / 0(9/9)→ +70(9/10)→ +5(9/11)(derivation)。**

⚑⊘ CNAL 五口径中断一期后恢复——但每一个区域升水都收窄了

  • ⚑⚑ **9月11日 CNAL 全套:上海 24,210–24,250,均 24,230(−310);长江 24,220–24,260,均 24,240(−320);中原 24,240(−320);南海/佛山 24,660–24,760,均 24,710(−340);广东南储 24,410–24,470,均 24,440(−360)。两个现货来源十二期以来首次在同一期同时可用。**
  • ⊘ **区域价差全线收窄:南海 − 上海 480(9月9日为 530);南储 − 上海 210(9月9日为 260);长江 − 上海 +10,不变。**
  • ⊘ **华南领跌:跌幅 340–360,对华东的 310–320。此前连续四日走阔的那个区域,这一次回吐最多,区域偏紧一项降级(interpretation)。**

⊘⊘ 氧化铝:一场同时击穿六条均线的破位,而持仓在上升

  • ⊘⊘ **AO2701(主力)9月11日:开 2,762,高 2,792,低 2,691,收 2,697,结算 2,731,较 2,770 前结算 −39(−1.41%);收盘对收盘 −61(−2.21%)。成交 275,018(+112,279,+69.0%),持仓 220,066(+24,675,+12.63%)。**
  • ⚑⚑ **本台账首次:六条氧化铝均线全部精确并转为实时(基于 160 行 AO2701 序列):MA5 2,757.00 / MA10 2,753.10 / MA20 2,737.15 / MA30 2,738.67 / MA40 2,749.22 / MA60 2,781.45。沿用五期、且已陈旧四个观测值的 9月4日 AO2610 基准予以退役。**
  • ⊘⊘ **而 9月11日收盘价 2,697 一次性低于全部六条,依次低 60.00 / 56.10 / 40.15 / 41.67 / 52.22 / 84.45。前一日尚在 MA5 之上,一场之内跌穿全部均线,是本页最干净的趋势破位;夜盘再低 24 点至 2,673(derivation)。**
  • ⊘⊘ **成本口径一第二次失守,幅度由 3 点扩大至 51 点:AO2610 收 2,667,低于 2,718 51 点。余量链条:+39(9/4)→ +19(9/8)→ +2(9/9)→ −3(9/10)→ −51(9/11)。事先写下的条件是「连续第二日收于 2,718 之下则降级坐实」——已坐实;且结算价 2,695 亦低于该线 23 点,两个口径同时失守。**
  • ⊘⊘ **成本口径二 2,826.21 已「无处定价」:9月9日首次超越位置为 AO2704,9月10日外移至 AO2705,9月11日全曲线无任何合约收于其上——最高者为 AO2708 的 2,815,仍低 11.21。远期成本通胀在至 2027年8月的整条氧化铝曲线上均未被定价(derivation)。**
  • ⊘ **十二个合约全部收在自身开盘价之下(9月10日为十一个)。**
  • ⊘ **9月11日完整收盘曲线:AO2609 2,680 / AO2610 2,667 / AO2611 2,684 / AO2612 2,694 / AO2701 2,697 / AO2702 2,710 / AO2703 2,730 / AO2704 2,757 / AO2705 2,769 / AO2706 2,796 / AO2707 2,806 / AO2708 2,815。仅一处倒挂,位于临近交割的 AO2609/AO2610 前端对;自 AO2610 向外完全单调。**
  • ⊘ **首尾 contango 第三次收窄:AO2610 → AO2708 为 +148,占前端 5.55%(前值 +175 / 6.51%,再前 +184 / 6.83%)。前端下跌同时远期曲线走平,是市场把过剩定价到远期,而不是把成本定价到前端(interpretation)。**
  • ⊘⊘ **持仓上升是本栏最重要的一句:全系持仓 462,497(+20,900,+4.73%),而 9月10日为 −1,830;全系成交 397,039(+153,247,+62.9%)。价格下跌而持仓上升,是新空入场而非多头减仓,其性质比减仓式下跌更具持续性(interpretation)。**
  • ⊘ **夜盘继续确认并继续加仓:AO2701 2,673(−58,−2.12%),成交 92,893,持仓 235,026(再 +14,960);AO2610 2,651(−44,−1.63%)、AO2611 2,659(−45,−1.66%)、AO2612 2,672(−39,−1.44%)、AO2702 2,690(−55,−2.00%)。五个合约全部下跌 1.44%–2.12%,而持仓仍在增加。**
  • ⊘ **盘面已贴水于现货参考:沿用且无日期的中国氧化铝现货口径为 2,707,已高于 AO2701 收盘价 10 点。该口径无日期,列示且不赋权。**

⚑⊘ 冶炼利润代理:水平继续压缩,但曲线压缩坍缩

  • ⊘ **配对口径(AL2610/AO2610,1.93 吨/吨,假设先行声明):日盘 24,235 − 1.93 × 2,667 = 19,087.69(前值 19,250.05,−162.36);夜盘 24,000 − 1.93 × 2,651 = 18,883.57(再 −204.12)。**
  • ⊘ **主力配对口径(AL2610/AO2701):24,235 − 1.93 × 2,697 = 19,029.79(前值 19,166.86,−137.07)(derivation)。**
  • ⚑ **但日盘曲线压缩大幅收窄:前端(AL2609/AO2609)19,067.60(−239.56),尾端(AL2708/AO2708)18,827.05(−77.36),压缩 240.55(前值 402.75),收窄 162.20,为本台账最大(derivation)。**
  • ⊘ **水平在下降、形状在走平,同时发生:即冶炼利润的绝对值在收缩,而市场对其期限差异的定价在消失(interpretation)。**

⊘ 铝合金与其他参考

  • ⊘ **SMM ADC12 24,300(−150),跌幅明显小于原铝的 −320,再生与原生价差被动收窄。**
  • ⊘ **ADC12 口径分歧仍为六个且不予调和:镜像 24,300;南海/大沥贸易报价 24,650–24,850(无日期);SMM 23,950;长江现货铝合金 24,000;长江华东综合 23,615;佛山主流 23,500–23,800。A356.2 沿用 24,400;AD2610 沿用 23,545,9月11日无任何 AD 合约数据。**

三 · 库存

⊘ 本期无新的全国数据,79.6 万吨沿用至下周一

  • ⊘ **SMM 全国铝锭社会库存为周一、周四发布,故 9月11日无数据,下一个数字为 9月14日(周一)。沿用链条:8/24 860,000 → 8/27 852,000 → 8/31 837,000 → 9/3 815,000 → 9/7 802,000 → 9/10 796,000。**
  • ⊘ **沿用判定(不变):周四对周四 −19,000,为此前 36,000 两观测速度的 52.8%、8月自身约 23,935 的 79.4%;周一至周四仅 −6,000。去库没有停止,它减半了,并且已低于 8月。自 7月初 113 万吨阶段高点累计 −334,000 吨(−29.6%)。**
  • ⚑ **Mysteel 全国口径沿用:8/31 809,000 → 9/3 789,000 → 9/7 774,000 → 9/10 768,000;SMM 减 Mysteel 在四个日期上为 28,000 / 26,000 / 28,000 / 28,000,恒定定义性偏移的结论维持。**

⚑ 伦敦:第二个非零行,而上期新写的实质门槛首次生效即正确

  • ⚑ **244,100 吨(9月11日,−250),为连续第二个非零行。**
  • ⚑⚑ **上期为修补 −175 的设定缺陷而明确写下的门槛是「单笔 −500 吨以上才构成实质」。−250 是三位数抽库但低于该门槛,故被正确归类为噪音,且无需任何争辩。方法修正首次使用即生效(derivation)。**
  • ⊘ **十六区间流量:8月18日 246,925 → 9月11日 244,100 = −2,825 吨(−1.14%);十七个区间中十个恰为零;9月4日的 −1,450 仍占全部的 51.3%。**
  • ⊘ **六十日流量:6月17日 316,525 → 244,100 = −72,425 吨(−22.88%);8月18日之前约 1,582 吨/区间,之后约 166 吨/区间,为此前速度的 10.5%。**

⊘ 缺失项:第六期没有区域分解

  • ⊘ **SMM 三地日度库存连续第六期不可得,沿用至 9月3日:641,000 吨(广东 143,000、无锡 300,000、巩义 198,000),去库 −10,500。**
  • ⊘ **因此本简报仍然无法回答「79.6 万吨的减速发生在哪一地」——而这正是区分「华东到货集中」与「真实需求停滞」的那个问题。**
  • ⊘ **上期所周五周度仓单与库存的一手口径再次不可得;第三方称上期所铝库存周环比约 −3%,列示且不赋权。**

解读

  • ⊘ **库存栏本期是本页唯一「没有发生任何事」的栏目,而在一个盘面破位、宏观转向、氧化铝崩塌的一周里,没有新数据本身就是一种风险:下周一的数字将同时承担两周的信息量(interpretation)。**
  • ⚑ **伦敦仓单几乎不动而 backwardation 仅让了 1 美元,两者互相印证:实货近端并未参与这次抛售(interpretation)。**
  • ⚑ **一项方法上的进步值得单独记录:上期因判别带不能覆盖数轴而误判,本期补写的 −500 吨门槛在第一次使用时就给出了正确且无争议的分类。规则写在数字之前,才具有约束力(interpretation)。**

四 · 宏观因素

⊘⊘ 美国 8月 CPI:数字通过了判别带,反应函数否决了它

  • ⊘⊘ **实际值(9月11日发布):整体 +0.4% 环比、+3.4% 同比;核心 +0.3% 环比(7月为 +0.2%)、同比 +2.4%(前值 2.5%)。汽油 +3.9%,贡献了当月整体涨幅的三分之一以上。**
  • ⊘⊘ **事先写下的判别带:「整体同比落在 +3.3% 至 +3.4% 之内且核心不超预期,则 FedWatch 的 56%–60.6% 区间大致维持;超过 +3.5% 则加息定价上修、美元反弹,对铝偏空;低于 +3.2% 则相反」。**
  • ⊘⊘ **整体同比 +3.4%,落在带内的上沿——而市场给出的是「超过 +3.5%」那一档的反应。**
  • ⊘⊘ **判定:本项记为方法缺陷,而非通过。本简报把检验挂在整体同比上,市场挂在核心环比与结构上。阈值没有错,映射错了。自下一期起,通胀类检验改以核心环比与 FedWatch 变动本身为判别对象(derivation)。**

⊘⊘ FedWatch:一个下午之内由约 70% 跳至约 90%

  • ⊘⊘ **9月16日加息 25 个基点的定价:CPI 前约 69.3%,CPI 后接近 90%;而本简报整周沿用的三口径为 56% / 59% / 60.6%。**
  • ⊘ **EY-Parthenon 将其公开判断由「按兵不动」改为「加息 25 个基点」,对应目标区间升至 3.75%–4.00%。**
  • ⊘⊘ **这是本台账上单日最大的一次联储路径重定价,对铝明确偏空(interpretation)。
  • ⊘ **10月联邦基金利率条连续第十五期缺失,故仍无法获得联储路径的期限结构。**

⚑ 而美元几乎没有反应——这本身是本页最重要的一条反证

  • ⚑ **DXY 99.095(9月11日,+0.047,+0.05%),四连跌就此终结,并为五期以来首次收在 99 上方。月度 −0.92%。**
  • ⚑⚑ **在联储路径重定价约 20 个百分点的当天,美元只动了 5 个基点。这是本简报连续五期所称「日元与 9月18日日本央行会议仍是主导通道、而非联储」这一判断迄今最强的证据(interpretation)。**

⊘⚑ 能源:Brent 单日 −2.8% 而周度 +8.7%,且口径冲突解决

  • ⊘ **Brent 结算 104.61 美元(9月11日,−2.8%),因外交进展回落;周四盘中曾触及约 108 美元;当周 +8.7%,连续第三个交易日站上 100 美元。**
  • ⚑ **本简报沿用的口径冲突(9月10日 101.25 对 105.71)就此解决,且解决在更高的那一行:−2.8% 结算于 104.61,隐含前一交易日约 107.6。101.25 口径作为不同合约月份予以退役(derivation)。**
  • ⊘ **对铝为双向:100 美元之上的能源抬高冶炼成本底部(偏多),同时使 9月16日的决定更难(偏空)(interpretation)。**

⚑⊘ 中国:8月出口强劲,铝材出口更强

  • ⚑ **出口同比 +25% 至 4,014.4 亿美元;进口 +28.2% 至 2,823.6 亿美元;贸易顺差 1,190.9 亿美元。**
  • ⚑ **8月未锻轧铝及铝材出口 626,000 吨,同比 +17.2%;1–8月累计 467 万吨,同比 +16.7%。**
  • ⊘ **反证必须同等记录:在一个不断增设关税壁垒的外部环境中,一条以 16.7% 增速扩张的出口线同时是集中度风险而不只是需求信号(interpretation)。**

解读

  • ⊘⊘ **本期宏观是单边偏空的:联储路径单日重定价约 20 个百分点,是本页所有价格变动的共同原因(interpretation)。**
  • ⚑ **但传导渠道并不完整:美元几乎没动,而铝价跌了 2%,说明这次下跌更多是风险资产的一致性重定价,而非美元渠道(interpretation)。**
  • ⚑⊘ **沪伦的分歧正好落在这条逻辑上:宏观冲击在伦敦被完整定价(−2.06%),在上海只被部分接受(−1.04%),比值单日上升 0.0766,为本轮最大(derivation)。**

五 · 基本面

⊘ 氧化铝供应过剩是本期基本面栏的主线

  • ⊘ **CERA 预计 2026 年全球氧化铝市场过剩 179 万吨。**
  • ⊘ **中国 1–6月氧化铝进口 227.7 万吨,同比 +749.1%。**
  • ⊘ **一个口径称国内 2026 年计划新增氧化铝产能约 860 万吨,全年过剩量预计超过 1,000 万吨;另一口径给出 2,700–2,900 元/吨的区间预测。两者均无精确日期,列示且不赋权。**
  • ⊘ **与本页盘面一致:AO2701 一次性跌穿六条均线、全系持仓上升 20,900、成本口径二全曲线无定价——三者共同指向的是供应端而非需求端的定价(interpretation)。**

⚑ 海外供应:一条被广泛转述却常常缺少日期的消息,本期补上日期

  • ⊘ **俄铝几内亚 Friguia 事件的正确日期是 8月24日:列车在 Dubreka 省 Kagbelen 附近脱轨,切断通往该 约 45 万吨/年氧化铝厂的铁路通道,出口一度中断、产量降速,燃料与烧碱的到厂运输同样受阻。本期以其正确日期列示,因为该消息本周在无日期的情况下再度流传。**
  • ⚑ 沿用:Alunorte 在 8月减产至 50% 后已恢复满产;力拓 Yarwun 自 10月起 −120 万吨/年;美铝 2026 年指引下调 20–30 万吨;Qatalum 约 60% 名义产能;Alba 1–3 号线停产;EGA Al Taweelah 已重启 18% 电解槽;几内亚约 1.5 亿吨/年出口上限目标。

⊘ 贸易政策:本期出现首个「同时针对中美两个来源」的措施

  • ⊘ **墨西哥对来自中国与美国的铝型材征收最终反倾销税(9月10日)。这是本台账上第一条同时打击两个来源的贸易措施,其含义与单边措施不同:它压缩的是整条北美型材供应链的可替代性(interpretation)。**
  • ⊘ 沿用:欧盟拟议对非 OECD 国家的废铝出口禁令;加拿大 25%→50% 已生效且点名铝杆;泰国 21.94% 临时复审;新西兰 ALENZ 保障措施申请;阿根廷、印度等地既有措施。

⊘ 下游:旺季尚未在数据上兑现

  • ⊘ **铝型材产能利用率处于 37.05% 的周期底部;龙头下游加工企业开工率 59.9%(8月13日);再生铝合金开工率 49.1%(沿用)。**
  • ⚑ **Mysteel 8月 A00 月均价:华东 23,869 元/吨(环比 +3.34%)、华南 24,010 元/吨(+3.79%)、华中 23,762.4 元/吨(+3.42%)。8月价格在涨,而开工率没有跟上——这正是「高价抑制消费」的典型形态(interpretation)。**
  • ⚑ 印度电子行业用铝需求预计以 **10.18% 复合增速增长(沿用,长期项)。**

⊘ 现货成交情绪:本期仍无法发布任何意愿指数

  • ⊘ **SMM 区域成交升贴水、出货意愿指数与采购意愿指数连续第十二期不可得。本简报因此不发布情绪指数,也不以任何代理指标替代。**
  • ⚑ **可以间接推断而不可量化的是:A00 单日 −320 而基差仍收敛至 +5,说明现货跟随而未领跌;而五个区域升水全线收窄,说明持货方出货意愿上升、下游接货意愿未同步上升(interpretation,明确标注为推断)。**

六 · 电解铝现货市场

⚑ 五地报价(9月11日,CNAL 口径)

  • ⚑ **上海 24,210–24,250,均 24,230(−310);长江 24,220–24,260,均 24,240(−320);中原 24,240(−320);南海/佛山 24,660–24,760,均 24,710(−340);广东南储 24,410–24,470,均 24,440(−360)。**
  • ⚑ **SMM 镜像口径:A00 24,240(−320);A00 佛山 24,430(−350);ADC12 24,300(−150)。**
  • ⊘ **SMM 与 CNAL 的上海口径相差 10 元(24,240 对 24,230),为两者可同时取得的少数交易日之一,差值极小,互相印证(derivation)。**

⊘ 区域结构:四日走阔之后的一次全线回吐

  • ⊘ **南海 − 上海 480(9月9日 530,−50);南储 − 上海 210(9月9日 260,−50);长江 − 上海 +10(不变)。**
  • ⊘ **华南跌幅(340–360)大于华东(310–320),即此前最强的区域本期最弱。**
  • ⊘ **本期无法给出成交量化情绪,故对「出货意愿/采购意愿」只作定性推断并明确标注(见上)。**

⚑ 基差与交割月结构

  • ⚑ **基差 +5(对 AL2610 收盘)、+50(对结算)、+240(对夜盘最新)。**
  • ⊘ **AL2609 已进入尾声:9月11日收 24,240、结算 24,205、持仓 11,780(−3,240),夜盘 23,970。**
  • ⚑⚑ **主力换月正在发生:夜盘 AL2611 持仓 213,268 已超过 AL2610 的 198,952;日盘 AL2610 仍以 3,543 领先。本简报将在日盘确认后切换口径,并在切换当期明确标注(derivation)。**

七 · 关键价格驱动因素

⚑ 偏多因素(17 项)

  • ⚑⚑ **点名的 24,075 守得分毫不差:9月11日交易日最低价与 9月10日夜盘低点同一价位,且恰为 8月31日收盘价。**
  • ⚑⚑ **日间时段买回夜盘跌幅的 110 点(约 24.4%),收盘高于最低点 160、仅低于最高点 80。**
  • ⚑⚑ **伦敦现货跌 69 美元,backwardation 只让了 1 美元(+25.00 → +24.00),实货近端未参与抛售。**
  • ⚑⚑ **在联储路径单日重定价约 20 个百分点的当天,DXY 只涨 0.05%,美元渠道传导明显不足。**
  • ⚑⚑ **沪市跑赢伦敦 1.02 个百分点,比值由 7.3257 升至 7.4023,终结伦敦三期跑赢。**
  • ⚑⚑ **现货检验落在中间带并撤销了本简报自己的错误怀疑:24,560 − 320 = 24,240 精确吻合,镜像口径为真实报价。**
  • ⚑⚑ **基差收敛至 +5,现货跟随而未领跌。**
  • ⚑⚑ **9月11日夜盘是移仓而非离场:六合约合计仅 −1,612,AL2610 的 −12,438 被 AL2611 的 +5,421 等吸收。**
  • ⚑ **MA40(23,847.25)与 MA60(23,601.25)完好,夜盘最新价距 MA40 仍有 152.75。**
  • ⚑ **三个铝合约收在自身开盘价之上(9月10日为零),十二个全部收在最低价之上。**
  • ⚑ **LME 库存 244,100 吨,十七个区间中十个为零,近端实货未见松动。**
  • ⚑ **两个现货来源十二期以来首次同时可用,CNAL 五口径中断一期后恢复。**
  • ⚑ **Brent 连续第三日站上 100 美元(104.61),当周 +8.7%,抬高冶炼成本底部。**
  • ⚑ **中国 8月未锻轧铝及铝材出口 626,000 吨(+17.2%),1–8月 +16.7%。**
  • ⚑ **中国 8月出口 +25%、顺差 1,190.9 亿美元,外需侧未见转弱。**
  • ⚑ 沿用:LME 库存处 36 年低位附近;Wood Mackenzie 2026 年全球短缺约 90 万吨(8月4日);ING 关于实货偏紧的判断。
  • ⚑ 沿用海外供应扰动:力拓 Yarwun、美铝指引下调、Qatalum、Alba、EGA、几内亚出口上限、俄铝 Friguia(8月24日)。

⊘ 偏空因素(33 项)

  • ⊘⊘ **FedWatch 9月16日加息定价由约 69.3% 跳至接近 90%,单日约 20 个百分点。**
  • ⊘⊘ **美国 8月核心 CPI +0.3% 环比(前值 +0.2%),整体 +0.4% 环比、+3.4% 同比。**
  • ⊘⊘ **EY-Parthenon 由「按兵不动」改判「加息 25 个基点」,目标区间将升至 3.75%–4.00%。**
  • ⊘⊘ **LME 现货 −69.00 美元(−2.06%),为本台账最大单日跌幅;三个月 −68.00。**
  • ⊘⊘ **LME 现货失守 MA5 与 MA10,MA20 仅余 17.95 美元、MA30 仅余 10.52 美元。**
  • ⊘⊘ **「移仓」定性在自设阈值的 1.96 倍处作废:全系持仓 −29,349。**
  • ⊘⊘ **三期口径下离场以 34,381 对 12,624 压倒移仓,与此前 16,900 对 5,000 完全反转。**
  • ⊘⊘ **成交量创本台账新高(全系 414,471,+55.4%)且伴随大幅减仓与下跌,为派发形态。**
  • ⊘⊘ **9月11日夜盘失守 MA20(−55.00)与 MA30(−29.83),夜盘口径下六条均线已失守四条。**
  • ⊘⊘ **夜盘低点 23,920 为 9月1日以来最低,恰为 8月28日收盘价,盘中口径下 9月涨幅已被抹去。**
  • ⊘⊘ **AO2701 一次性跌穿全部六条均线(低 40.15 至 84.45)。**
  • ⊘⊘ **氧化铝全系持仓 +20,900、AO2701 +24,675,夜盘再 +14,960——新空入场。**
  • ⊘⊘ **成本口径一第二次失守且由 3 点扩大至 51 点,结算价亦低于该线 23 点。**
  • ⊘⊘ **成本口径二 2,826.21 在十二个合约上全部未被定价,最高收盘 2,815 仍低 11.21。**
  • ⊘⊘ **十二个氧化铝合约全部收在自身开盘价之下(前值十一个)。**
  • ⊘ **首尾 contango 第三次收窄至 +148(5.55%)。**
  • ⊘ **配对利润代理日盘再 −162.36、夜盘再 −204.12。**
  • ⊘ **十二个铝合约结算价全部下跌(−315 至 −400)。**
  • ⊘ **AL2610 收盘失守 MA5(−198.00)与 MA10(−76.00)。**
  • ⊘ **收盘分位由 95/98.3 降至 70/85。**
  • ⊘ **主力持仓 −25,055(−10.60%),两期合计 −35,678。**
  • ⊘ **五个区域现货升水全线收窄,华南领跌(−340 至 −360)。**
  • ⊘ **南海 − 上海 480(前 530)、南储 − 上海 210(前 260)。**
  • ⊘ **LME 均线倒挂第三次扩大:MA30 高于 MA20 7.43 美元。**
  • ⊘ **本期无新的全国库存数据,去库减半的判定沿用而未获更新。**
  • ⊘ **SMM 三地日度库存连续第六期缺失,无法定位减速发生地。**
  • ⊘ **上期所周度仓单一手口径再次缺失。**
  • ⊘ **CERA 预计 2026 年全球氧化铝过剩 179 万吨;中国 1–6月氧化铝进口 +749.1%。**
  • ⊘ **一个口径称 2026 年国内新增氧化铝产能约 860 万吨、全年过剩超 1,000 万吨(无日期,不赋权)。**
  • ⊘ **铝型材产能利用率 37.05% 周期底部,龙头下游开工率 59.9%。**
  • ⊘ **墨西哥对中美两地铝型材征最终反倾销税,为本台账首个双来源措施。**
  • ⊘ **10月联邦基金利率条连续第十五期缺失,联储路径期限结构不可得。**
  • ⊘ 沿用:高盛 2,350 美元/吨(无日期,不赋权),现较 9月11日结算价低约 **29.1%。**

八 · 一句话观点

事先写下的七项检验在一场交易中全部落地,创本台账纪录——五项偏空、一项各半、一项数字通过而反应函数否决:美国 8月 CPI 整体 +0.4% 环比、+3.4% 同比恰落在事先写下的 +3.3%–3.4% 带内,但核心 +0.3% 环比(前值 +0.2%)、汽油 +3.9% 贡献了整体涨幅三分之一以上,FedWatch 对 9月16日加息的定价一个下午之内由约 69.3% 跳至接近 90%(本周沿用的三口径为 56%/59%/60.6%),EY-Parthenon 由「按兵不动」改判加息 25 个基点——阈值没有错,是本简报把检验挂在了整体同比而非核心环比上,该缺陷如实记录并自下期起改以核心环比与 FedWatch 变动本身为判别对象;冲击几乎全部由伦敦承担,现货 **−69.00 美元(−2.06%)为本台账最大单日跌幅、失守 MA5 与 MA10 而 MA20 仅余 17.95、MA30 仅余 10.52,但backwardation 五期以来首次收窄且只让了 1.00 美元(+25.00 → +24.00)、库存仅 −250 吨(低于上期新写的 −500 实质门槛,该门槛首次使用即给出正确分类),说明抛售的不是实货近端;沪市方面,点名的 24,075 守得分毫不差(与夜盘低点同一价位、恰为 8月31日收盘价),但确认条件差 44.50 点未达成(收 24,235,低于 24,279.50),故判为各半,而日间时段实际买回了夜盘 450 点中的 110 点、全日收盘高于最低点 160;真正的坏消息在持仓与氧化铝:全系持仓 −29,349,为自设作废阈值的 1.96 倍,「移仓」定性正式撤销,三期口径下离场以 34,381 对 12,624 压倒移仓(此前为 16,900 对 5,000),且发生在本台账最重的成交量上(全系 414,471,+55.4%),是派发而非换手;氧化铝 AO2701 收 2,697(−61,−2.21%)一次性跌穿其自身全部六条均线(本期首次实现六条氧化铝均线精确并实时),成本口径一第二次失守且由 3 点扩大至 51 点、成本口径二在整条曲线上无任何合约定价,而全系持仓 +20,900、AO2701 +24,675、夜盘再 +14,960——价格下跌伴随持仓上升是新空入场而非多头减仓,其持续性强于减仓式下跌;9月11日夜盘随后拿走 MA20(−55.00)与 MA30(−29.83),低点 23,920 为 9月1日以来最低且恰为 8月28日收盘价,但夜盘持仓仅 −1,612 且 AL2611 反超 AL2610(213,268 对 198,952),即夜盘是移仓、日盘是离场,两者性质必须分开记录;唯一干净的偏多落地在现货:SMM A00 24,240(−320),24,560 − 320 = 24,240 精确吻合,回溯证明上期镜像行是真实报价,本简报自己的「疑似沿用」怀疑被证伪并予撤销,基差收敛至 +5、现货跟随而未领跌,CNAL 五口径亦中断一期后恢复,只是五个区域升水全线收窄且华南领跌;而两条最重要的反证是——在联储路径重定价约 20 个百分点的当天 DXY 只动了 0.05%,以及沪市跑赢伦敦 1.02 个百分点、比值单日升 0.0766 为本轮最大,说明宏观冲击在伦敦被完整定价、在上海只被部分接受;综合而言这是本简报迄今最弱的一页,而下周一的问题窄而可算:日盘能否收复 24,055.00,还是 23,920 成为新的参照点(interpretation)。**

九 · 下周一(9月14日)日盘与夜盘关注清单

  • ⚑⚑ 第一,MA20(24,055.00)是被收复还是被确认失守。 夜盘最新 24,000 已在其下 55.00,MA30(24,029.83)亦在其下 29.83。判别写在数字之前:9月14日日盘收在 24,055.00 之上,则夜盘破位为超跌、周五的「各半」倒向多头一侧;收在 23,920 之下(9月11日夜盘低点 = 8月28日收盘价),则本轮回调升级为趋势转折,下一参照为 MA40 23,847.25;收在 23,920–24,055 之间记为无结论区,不作任何定性。参考位:24,311.00(MA10)/ 24,235(9月11日收盘)/ 24,075(已守住的关键位)/ 24,055.00(MA20)/ 24,029.83(MA30)/ 24,000(夜盘最新)/ 23,920(夜盘低点)/ 23,847.25(MA40)。
  • ⚑⚑ 第二,离场是否继续,以及主力换月何时在日盘确认。 本期全系持仓 −29,349 已使「移仓」作废。判别带(收紧):9月14日全系持仓再降 10,000 以上,则离场为趋势而非单日事件,应下调对本轮全部反弹的权重;回升 5,000 以上,则周五属一次性平仓潮;±5,000 以内记为待定。同时观察 AL2611 日盘持仓是否超过 AL2610(周五夜盘已超 14,316,日盘仍差 3,543)——一旦日盘确认,本简报将于当期切换主力口径并明确标注。
  • ⚑⚑ 第三,氧化铝的新空是继续加仓还是获利了结。 这是本页性质最明确的一项:价格跌而持仓升。判别带:AO2701 持仓继续上升且价格继续下跌,则空头趋势成立,成本支撑一项应完全移出多头清单;持仓下降而价格反弹,则周五为空头挤压前的最后一段;AO2610 若重新收于 2,718 之上,则成本口径一的失守可回撤,但需连续两日才予采纳(与本期采纳降级时所用的对称条件一致)。另观察最高收盘价能否重新越过 2,826.21(成本口径二),目前全曲线无一合约在其上。
  • ⚑⚑ 第四,伦敦的 backwardation 与库存,哪一个先让步——本期已给出方向。 现货跌 69 美元而结构只让 1 美元,是本页最强的实货证据。判别:现货能否守住 MA20 3,256.05(余量仅 17.95)与 MA30 3,263.48(余量 10.52);backwardation 是继续收窄、翻回 contango(则实货论据作废)还是重新走阔(则论据加强);库存是否出现单笔 −500 吨以上的抽库(本期新门槛,−250 已被正确归类为噪音)。
  • ⚑ 第五,9月16日 FOMC 之前的定价是否继续上修,以及美元是否终于跟上。 FedWatch 已至约 90%。判别:若定价升至 95% 以上且 DXY站上 99.5,则美元渠道开始传导,对铝进一步偏空;若定价维持约 90% 而 DXY 仍在 99 附近横盘,则本简报关于「日元/日本央行 9月18日会议才是主导通道」的判断继续成立,宏观压力被部分隔离。10月联邦基金利率条若终于可得,应立即取回以还原期限结构。
  • ⚑ 第六,周一的 SMM 全国铝锭库存数字——本期是它唯一没有更新的栏目,而它将承担两周的信息量。 沿用链条为 9/7 802,000 → 9/10 796,000,周四对周四 −19,000(此前速度的 52.8%)。判别带(本期重写,且完整覆盖数轴):≤ 785,000 = 去库重新加速;785,001–792,000 = 维持当前减速节奏;792,001–799,000 = 进一步减速;≥ 799,001 = 停滞或累库。同时观察 Mysteel 同日数字与 SMM 的差值是否仍在 26,000–28,000 区间(若偏离,则「恒定定义性偏移」的结论需重新审视)。
  • ⚑ 第七,现货能否在盘面继续下跌时守住「跟随而不领跌」的姿态,以及区域升水是否继续收窄。 周五 A00 −320、基差 +5、五地升水全线收窄且华南跌幅最大。判别带:若盘面周一再跌而 A00 跌幅小于盘面、基差转为正升水扩大,则现货偏紧的论据加强;若 A00 跌幅大于盘面且南海 − 上海价差跌破 450,则区域偏紧一项应整体移出多头清单。另需持续跟踪 SMM 三地日度库存能否在连续第六期缺失后补发,以及上期所周度仓单的一手口径。

免责声明

  • 本简报仅供内部晨会参考。数据来源包括上海期货交易所公开日度与夜盘数据及其第三方镜像、Westmetall、TradingEconomics、CME FedWatch、美国劳工统计局、路透、CNBC、CNAL 世铝网、AL Circle、Mysteel、SMM(上海有色网)、中国海关总署(经转述)、Mining.com / Mining Weekly、CERA、Wood Mackenzie、ING 以及公开报道的高盛与 EY-Parthenon 观点。标注「⚑ / ⊘」者为事实数据;标注「(interpretation)」或「(derivation)」者为本简报的判断或计算,二者在全文中明确区分。来源冲突时,所有口径平行列示且不予调和,除非一手来源证伪其中之一。依赖假设系数的推导(如 1.93 吨氧化铝/吨铝)先声明假设再计算。本期作出三项自我更正并予以明确记录:其一,上一期对 9月10日镜像现货「三个零变动行疑似沿用」的怀疑被证伪(24,560 − 320 = 24,240 精确吻合),该标记撤销,9月10日 +70 的基差读数维持;其二,本简报对美国 CPI 的判别带挂错了序列(挂在整体同比而非核心环比),导致数字在带内而市场反应在带外,该缺陷记录并自下期起修正;其三,Brent 的双口径冲突解决在更高的一行,101.25 口径作为不同合约月份退役。此外,上一期新写的 LME 库存「−500 吨实质门槛」在本期首次使用即给出正确分类,予以记录;本期第六项关注清单的库存判别带已重写为完整覆盖数轴,以避免 9月10日 796,000 吨落入空隙的同类缺陷。本简报不构成投资建议,亦不对据此作出的任何决策承担责任。

1 · Market Performance

Statement of writing time, the nature of this edition, and one caliber note

  • Written the morning of Saturday, September 12, with BOTH markets closed for the weekend.
  • ⚑⚑ NEW THIS EDITION: the complete SHFE September 11 trading day for all twelve aluminium and all twelve alumina contracts, retrieved from a 224-row daily series whose September 10 values reconcile to the tick against last edition’s print; ⚑⚑ the COMPLETE September 11 NIGHT SESSION for six aluminium and five alumina contracts, whose eleven prior-settlement fields reconcile EXACTLY against the September 11 settlements; ⚑⚑ ALL SIX SHFE averages exact for a SECOND consecutive edition, and — for the FIRST time on this ledger — ALL SIX ALUMINA averages exact and LIVE on the AO2701 main-contract series; ⚑⚑ the Westmetall September 11 row, so all six LME averages are exact for a FOURTH consecutive edition; ⚑⚑ the US August CPI RESULT and the post-CPI FedWatch move; ⚑⚑ the CNAL five-caliber regional spot set for September 11, RESUMED after a single-edition break; ⚑ a third-party SMM mirror whose arithmetic RETROACTIVELY PROVES the September 10 mirror rows were real quotations; the September 11 DXY and Brent settlement; and China’s August trade and aluminium export data.
  • ⊘ NOT RETRIEVABLE: SMM’s own primary September 11 quotation set (login-walled for a TWELFTH consecutive edition); SMM’s three-region daily inventory series, a SIXTH consecutive edition; the SHFE Friday weekly warehouse and warrant print in its primary caliber (a third-party −3% w/w note is listed and flagged); any AD contract data; SMM’s regional transaction premia and sentiment indices and Chalco’s five listings; LME volume and open interest; a dated Chinese alumina spot index for September 11; and the October fed funds strip, absent for a FIFTEENTH consecutive edition.
  • ⚑ CALIBER NOTE (important): the SHFE trading day dated September 11 BEGINS with the 21:00 September 10 night session, so the September 11 open (24,310), high (24,315) and low (24,075) are the night ticks this brief published one edition ago. The daytime portion is reported SEPARATELY and explicitly throughout.

⚑⚑ The named 24,075 held to the tick — and the confirmation missed by 44.50

  • ⚑⚑ The pre-written test, verbatim: ‘hold **24,075 and close above 24,279.50 (MA10) and the night was oversold; break 24,037.50 and close below it and the sixty-session structure is broken.’**
  • ⚑⚑ The September 11 trading-day low is **24,075 — the SAME TICK as the September 10 night low, not one point lower, and EXACTLY the August 31 close. The support half passed perfectly.**
  • ⊘ **The confirmation half failed by 44.50 points: AL2610 closed 24,235 against the 24,279.50 the test required; against the LIVE MA10 of 24,311.00 it is 76.00 below.**
  • ⊘ **So the ruling is SPLIT: the level is intact, the recovery is not confirmed, and neither branch can be declared (derivation).**

⊘ The September 11 trading day in full — and the daytime recovery, recorded separately

  • ⊘ **AL2610, September 11 (trading-day caliber): open 24,310, high 24,315, low 24,075, close 24,235, settlement 24,190, −385 (−1.57%) against the 24,575 prior settlement; close-to-close −255 (−1.04%). Volume 213,873 (+69,493, +48.1%), open interest 211,390 (−25,055, −10.60%).**
  • ⚑ **The daytime hours were a recovery and are recorded with equal weight: the September 10 night ended at 24,125; September 11 closed at 24,235; so the day bought back +110 of the night’s −450, roughly 24.4% of it, and finished 160 above its low and only 80 below its high. The whole of the 1.04% loss was made before the day session opened (derivation).**
  • ⊘ **Twelve of twelve aluminium settlements fell, by −315 to −400: AL2609 −350 / AL2610 −385 / AL2611 −385 / AL2612 −380 / AL2701 −385 / AL2702 −400 / AL2703 −395 / AL2704 −395 / AL2705 −370 / AL2706 −365 / AL2707 −355 / AL2708 −315. An 85-point band across twelve months is a parallel shift, not a front-end event.**
  • ⚑ **But THREE of twelve closed ABOVE their opens — AL2609, AL2704 and AL2708 — against ZERO on September 10, and twelve of twelve closed above their lows (derivation).**

⚑⚑ All six SHFE averages exact for a second edition, and the series self-verifies

  • ⚑⚑ **MA5 24,433.00 / MA10 24,311.00 / MA20 24,055.00 / MA30 24,029.83 / MA40 23,847.25 / MA60 23,601.25.**
  • ⚑⚑ Recomputed at September 10 the same 224-row series returns **24,444.00 / 24,279.50 / 24,037.50 / 24,009.33 / 23,821.50 / 23,599.08 — exactly the six numbers published last edition. The series is verified, not asserted (derivation).**
  • ⊘ **The close lost MA5 and MA10: 24,235 is 198.00 below MA5 and 76.00 below MA10, and 180.00 above MA20, 205.17 above MA30, 387.75 above MA40, 633.75 above MA60. Two of six averages are lost on the CLOSE, against zero on September 10.**
  • ⊘ **The close fell to the 70th and 85th percentiles of the twenty- and sixty-session close distributions, against 95th/98.3rd and 100th/100th. Within the twenty-session range (23,595–24,560) it sits at 66.3%; within the sixty-session range (22,445–24,560) at 84.6% (derivation).**

⊘⊘ And the September 11 night session took MA20 and MA30

  • ⊘⊘ **AL2610 night: open 24,100, high 24,100, low 23,920, last 24,000, −190 (−0.79%) against the 24,190 settlement. Volume 81,993, open interest 198,952 (−12,438).**
  • ⊘⊘ **24,000 sits 55.00 BELOW MA20 (24,055.00) and 29.83 BELOW MA30 (24,029.83), and 152.75 above MA40 (23,847.25). Four of six averages are now lost on the night caliber; only MA40 and MA60 are intact.**
  • ⊘ **The night low of 23,920 is the lowest tick since September 1 — below every close from September 1 onward — and is EXACTLY the August 28 close. On an intraday caliber the whole September advance is gone; on a closing caliber it is not (derivation).**
  • ⊘ Six night contracts: AL2609 **23,970 (−235, −0.97%) / AL2610 24,000 (−190, −0.79%) / AL2611 24,000 (−180, −0.74%) / AL2612 23,990 (−180, −0.74%) / AL2701 23,955 (−190, −0.79%) / AL2702 23,950 (−190, −0.79%).**
  • ⚑ **But the night was a ROLL, the opposite of the day: six contracts, day 560,984 → night 559,372, only −1,612. AL2610 −12,438 while AL2611 +5,421, AL2612 +3,601, AL2701 +1,851, AL2702 +258. Position migrated down the curve rather than leaving — the exact inverse of the September 10 night’s outright −20,436 (derivation).**
  • ⚑⚑ **And in that same night AL2611’s open interest OVERTOOK AL2610’s: 213,268 against 198,952, a 14,316 lead. On the day session AL2610 still leads by 3,543 (211,390 vs 207,847), down from 24,734. The main-contract handover is underway; this brief will switch calibers when a DAY session confirms it (derivation).**

⊘⊘ The roll ruling is dead — at 1.96x this brief’s own threshold

  • ⊘⊘ The pre-written band: ‘±5,000 complex OI is still a roll; −15,000 or more voids the roll explanation and it becomes EXIT.’
  • ⊘⊘ **Complex open interest 593,718, −29,349 (−4.71%) — 1.96 times the void threshold. The roll explanation carried for two editions is formally WITHDRAWN.**
  • ⊘⊘ **And the three-session arithmetic inverts completely: September 8 → September 11, main 258,395 → 211,390 = −47,005; complex 628,099 → 593,718 = −34,381. So ~12,624 rolled and ~34,381 LEFT. Two editions ago the same calculation gave 16,900 roll against 5,000 exit. Exit now outweighs roll 2.7 to 1 having been outweighed 3.4 to 1 — the largest reversal of an interpretation on this ledger (derivation).**
  • ⊘ **Volume proves it was an event, not a drift: main 213,873 (+48.1%), complex 414,471 (+147,820, +55.4%), the heaviest on this ledger. Heavy volume + heavy position reduction + a falling price is distribution, and it is recorded as such (interpretation).**

⊘⊘ London: the largest cash fall on this ledger — and the structure gave up one dollar

  • ⊘⊘ Cash settled **$3,274.00, −$69.00 (−2.06%) — the largest single-day cash decline this brief has recorded; the 3-month settled $3,250.00, −$68.00 (−2.05%).**
  • ⊘ **Cash LOST MA5 and MA10: $46.80 below MA5 ($3,320.80) and $19.80 below MA10 ($3,293.80); $17.95 above MA20 ($3,256.05), $10.52 above MA30 ($3,263.48), $29.96 above MA40, $56.06 above MA60. The $17.95 is the narrowest support clearance of this run.**
  • ⚑⊘ **The backwardation narrowed for the FIRST time in five — and only by $1.00: −$0.50 → +$2.00 → +$6.00 → +$14.00 → +$25.00 → +$24.00. A $69 fall in cash cost the prompt structure exactly one dollar. This is the strongest evidence on the page that whatever sold aluminium on Friday, it was not the physical prompt market (interpretation).**
  • ⚑ **All six LME averages exact for a FOURTH consecutive edition; recomputed at September 10 the series returns $3,324.50 / $3,287.65 / $3,256.30 / $3,262.05 / $3,241.44 / $3,220.07 — five of six match last edition to the cent and the sixth to one hundredth (derivation).**
  • ⊘ **The non-monotone stack widened a third step: MA30 is $7.43 above MA20, against $5.75 and $2.99.**
  • ⚑⚑ **The ratio reversed hard and Shanghai out-performed by 1.02 points: 24,235 ÷ $3,274.00 = 7.4023, against 7.3257, 7.3270 and 7.3564. SHFE −1.04% against LME cash −2.06%, ending three consecutive sessions of London outperformance. The macro shock was priced in London and largely refused in Shanghai (interpretation).**

⚑⚑ The spot test resolved on its middle branch — and cleared this brief’s own false suspicion

  • ⚑⚑ The pre-written bands: above **24,400 = spot did not follow (bullish); 24,150–24,300 = spot lagged one day; below 24,100 = spot led lower (bearish).**
  • ⚑⚑ **SMM A00 printed 24,240 (−320) — squarely in the middle of the middle band. Spot followed the board by a day; it neither led lower nor refused to follow.**
  • ⚑⚑ **And the arithmetic vindicates last edition’s data: 24,560 − 320 = 24,240; Foshan 24,780 − 350 = 24,430; ADC12 24,450 − 150 = 24,300 — all three EXACT. The September 10 mirror rows were real quotations carrying a genuine zero change, not carried rows. The ‘three zero-change rows look like a carry’ flag is WITHDRAWN, the +70 basis published for September 10 STANDS, and the suspicion is recorded as having been WRONG.**
  • ⚑ **The basis converged to +5: 24,240 − 24,235 = +5 against the close; +50 against the 24,190 settlement; +240 against the 24,000 night last. The pre-written normal band was ‘0 to −50 against the close’; +5 is five points above its top edge. Chain: +70 (Sep 4) → +55 (Sep 7) → −20 (Sep 8) → −20/0 (Sep 9) → +70 (Sep 10) → +5 (Sep 11) (derivation).**

⚑⊘ CNAL resumed after a single-edition break — but every regional premium narrowed

  • ⚑⚑ **CNAL, September 11: Shanghai 24,210–24,250 avg 24,230 (−310); Changjiang 24,220–24,260 avg 24,240 (−320); Zhongyuan 24,240 (−320); Nanhai/Foshan 24,660–24,760 avg 24,710 (−340); Guangdong Nanchu 24,410–24,470 avg 24,440 (−360). Both spot sources are live in the same edition for the first time in twelve.**
  • ⊘ **Every regional spread narrowed: Nanhai − Shanghai 480 (530 at Sept 9); Nanchu − Shanghai 210 (260); Changjiang − Shanghai +10, unchanged.**
  • ⊘ **South China led the decline, 340–360 against East China’s 310–320. The region that had been widening for four sessions gave back the most; the regional-tightness leg is downgraded (interpretation).**
  • ⊘ SMM Shanghai 24,240 against CNAL Shanghai 24,230 — **a 10-yuan gap on one of the few days both are available; the two calibers corroborate each other (derivation).**

⊘⊘ Alumina: a break through all six averages at once, on RISING open interest

  • ⊘⊘ **AO2701, September 11: open 2,762, high 2,792, low 2,691, close 2,697, settlement 2,731, −39 (−1.41%) on settlement and −61 (−2.21%) close-to-close. Volume 275,018 (+112,279, +69.0%), open interest 220,066 (+24,675, +12.63%).**
  • ⚑⚑ **FIRST ON THIS LEDGER: all six alumina averages exact and LIVE on a 160-row AO2701 series: MA5 2,757.00 / MA10 2,753.10 / MA20 2,737.15 / MA30 2,738.67 / MA40 2,749.22 / MA60 2,781.45. The four-observation-stale September 4 AO2610 carry flagged for five editions is RETIRED.**
  • ⊘⊘ **And the close of 2,697 is below ALL SIX at once, by 60.00 / 56.10 / 40.15 / 41.67 / 52.22 / 84.45. From above MA5 the day before to below MA60 in one session is the cleanest trend break on this page; the night took it 24 lower to 2,673 (derivation).**
  • ⊘⊘ **Cost caliber one failed a SECOND time, and the margin went from 3 to 51: AO2610 closed 2,667, 51 below 2,718. Chain: +39 (Sep 4) → +19 (Sep 8) → +2 (Sep 9) → −3 (Sep 10) → −51 (Sep 11). The pre-written condition — ‘a second consecutive close below confirms the downgrade’ — is CONFIRMED, and the settlement of 2,695 is 23 below the line as well, so both calibers now fail.**
  • ⊘⊘ **Cost caliber two is worse than failed — it is UNPRICED: the first close above 2,826.21 was AO2704 on September 9 and AO2705 on September 10; on September 11 there is NO SUCH CONTRACT anywhere out to August 2027, because the highest close on the board is AO2708 at 2,815, still 11.21 below (derivation).**
  • ⊘ **Twelve of twelve alumina contracts closed below their opens (eleven of twelve the session before). The complete close curve: AO2609 2,680 / AO2610 2,667 / AO2611 2,684 / AO2612 2,694 / AO2701 2,697 / AO2702 2,710 / AO2703 2,730 / AO2704 2,757 / AO2705 2,769 / AO2706 2,796 / AO2707 2,806 / AO2708 2,815. One inversion, at the expiring AO2609/AO2610 front pair; monotone from AO2610 outward.**
  • ⊘ **The contango narrowed a third time: front-to-tail +148, 5.55% of the front, against +175 / 6.51% and +184 / 6.83%. A flattening forward curve on a falling front is the market pricing the surplus forward rather than the cost floor (interpretation).**
  • ⊘⊘ **The most important sentence in this column is about position, not price: complex OI 462,497 (+20,900, +4.73%), against −1,830 the session before, on volume 397,039 (+153,247, +62.9%). A falling price on rising open interest is NEW SHORTS entering, not longs leaving — a more durable signature than liquidation (interpretation).**
  • ⊘ **The night confirmed it and kept adding: AO2701 2,673 (−58, −2.12%) on 92,893 lots with open interest 235,026 (a further +14,960); AO2610 2,651 (−44, −1.63%), AO2611 2,659 (−45, −1.66%), AO2612 2,672 (−39, −1.44%), AO2702 2,690 (−55, −2.00%). Five of five down 1.44%–2.12% with position still building.**
  • ⊘ **The board now trades at a discount to the listed physical reference: the carried and UNDATED Chinese alumina spot caliber of 2,707 is 10 above the AO2701 close. Undated, listed and NOT weighted.**

⚑⊘ The smelter margin proxy: the level keeps compressing, the curve compression collapsed

  • ⊘ **Matched pair AL2610/AO2610, 1.93 t/t (assumption stated before the calculation): day 24,235 − 1.93 × 2,667 = 19,087.69 against 19,250.05, −162.36; night 24,000 − 1.93 × 2,651 = 18,883.57, a further −204.12.**
  • ⊘ **Main-contract pair AL2610/AO2701: 24,235 − 1.93 × 2,697 = 19,029.79 against 19,166.86, −137.07 (derivation).**
  • ⚑ **But the day-curve compression collapsed: front (AL2609/AO2609) 19,067.60 (−239.56), tail (AL2708/AO2708) 18,827.05 (−77.36), compression 240.55 against 402.75 — a −162.20 narrowing, the largest on this ledger (derivation).**
  • ⊘ **The level is falling and the shape is flattening at the same time: the absolute margin is shrinking while the market’s pricing of its term differences disappears (interpretation).**

⊘ Alloy and other references

  • ⊘ **SMM ADC12 24,300 (−150) — a much smaller fall than primary’s −320, so the secondary-to-primary spread compressed passively.**
  • ⊘ **The ADC12 caliber spread remains SIX and is NOT reconciled: mirror 24,300; a Nanhai/Dali trade quotation 24,650–24,850 (undated); SMM 23,950; Changjiang spot alloy 24,000; Changjiang composite East China 23,615; Foshan mainstream 23,500–23,800. A356.2 carried at 24,400; AD2610 carried at 23,545, with no AD contract data of any kind for September 11.**

3 · Inventory

⊘ No new national print, so 796,000 mt is carried into Monday

  • ⊘ **SMM publishes Monday and Thursday, so there is no September 11 print; the next number is Monday September 14. Carried chain: 8/24 860,000 → 8/27 852,000 → 8/31 837,000 → 9/3 815,000 → 9/7 802,000 → 9/10 796,000.**
  • ⊘ **Carried ruling (unchanged): Thursday-to-Thursday −19,000 is 52.8% of the prior 36,000 two-observation pace and 79.4% of August’s own ~23,935; Monday-to-Thursday only −6,000. The destock HALVED rather than stopped, and it is now running below August. Cumulative from the early-July 1.13 million mt phase high −334,000 mt, −29.6%.**
  • ⚑ **Mysteel national caliber, carried: 8/31 809,000 → 9/3 789,000 → 9/7 774,000 → 9/10 768,000; SMM − Mysteel 28,000 / 26,000 / 28,000 / 28,000 across four dates. The constant definitional offset finding stands.**

⚑ London: a second non-zero row, and last edition’s new threshold worked first time of asking

  • ⚑ **244,100 mt at September 11, −250 — the second consecutive non-zero row.**
  • ⚑⚑ Last edition wrote the substance threshold explicitly — ‘a single draw of −500 mt or more is required’ — precisely to stop a repeat of the −175 debacle. **−250 is a three-figure draw and is BELOW the threshold, so it is classified as noise without any argument. The method fix is validated on its first use (derivation).**
  • ⊘ **Sixteen-interval flow: August 18 246,925 → September 11 244,100 = −2,825 mt (−1.14%); ten of seventeen intervals exactly zero; September 4’s −1,450 still 51.3% of the whole.**
  • ⊘ **Sixty-session flow: June 17 316,525 → 244,100 = −72,425 mt (−22.88%); ~1,582 mt per interval to August 18 and ~166 mt per interval since, 10.5% of the prior pace.**

⊘ The missing column: a sixth consecutive edition with no regional breakdown

  • ⊘ **SMM’s three-region daily series was not retrievable for a SIXTH consecutive edition and is carried at September 3: 641,000 mt (Guangdong 143,000, Wuxi 300,000, Gongyi 198,000), draw −10,500.**
  • ⊘ **So this brief STILL cannot say WHERE the deceleration happened — which is precisely the question that would separate an East China arrival cluster from a genuine demand stall.**
  • ⊘ **The SHFE Friday weekly warrant print was again not retrievable in its primary caliber; a third-party note has SHFE aluminium stocks down ~3% w/w and is listed WITHOUT weight.**

Interpretation

  • ⊘ **Inventory is the one column where nothing happened this week — and in a week that broke the board, turned the macro and collapsed alumina, no new data is itself a risk: Monday’s number will carry two weeks of information (interpretation).**
  • ⚑ **The near-motionless LME stock row and the one-dollar cost of a $69 cash fall corroborate each other: the physical prompt market did not participate in this sell-off (interpretation).**
  • ⚑ **One methodological gain deserves its own line: last edition’s bands failed because they did not partition the line; the −500 mt threshold written down in response produced a correct, uncontested classification on its very first use. A rule written before the number is a rule that binds (interpretation).**

4 · Macro Factors

⊘⊘ US August CPI: the number passed the band and the reaction function overruled it

  • ⊘⊘ **Result (released September 11): headline +0.4% m/m, +3.4% y/y; core +0.3% m/m (July +0.2%), core +2.4% y/y (from 2.5%). Gasoline +3.9%, accounting for over a third of the monthly gain.**
  • ⊘⊘ The pre-written band: ‘headline y/y within +3.3% to +3.4% with core not above expectations** → FedWatch’s 56%–60.6% range roughly maintained; above +3.5% → hike pricing up, dollar rebounds, bearish; below +3.2% → the reverse.’**
  • ⊘⊘ **The headline landed at the top of the in-band range — and the market delivered the ‘above +3.5%’ reaction.**
  • ⊘⊘ **Ruling: this is recorded as a METHOD DEFECT, not a pass. This brief keyed the test to the headline y/y; the market keyed it to the core m/m and the composition. The threshold was not wrong — the mapping was. From the next edition, inflation tests key on core m/m and on the FedWatch move itself (derivation).**

⊘⊘ FedWatch: ~70% to ~90% in one afternoon

  • ⊘⊘ **September 16 25 bp HIKE pricing: ~69.3% pre-CPI, close to 90% post-CPI; against the 56% / 59% / 60.6% three-caliber spread this brief carried all week.**
  • ⊘ **EY-Parthenon switched its published call from HOLD to a 25 bp HIKE, taking the target range to 3.75%–4.00%.**
  • ⊘⊘ **This is the largest single-day repricing of the Fed path on this ledger and it is unambiguously bearish for aluminium (interpretation).**
  • ⊘ **The October fed funds strip is absent for a FIFTEENTH consecutive edition, so the term structure of the Fed path remains unavailable.**

⚑ And the dollar barely reacted — which is the most important counter-evidence on this page

  • ⚑ **DXY 99.095 at September 11, +0.047 (+0.05%) — ending a four-session decline and closing above 99 for the first time in five. Month −0.92%.**
  • ⚑⚑ **On the day the Fed path repriced roughly twenty points, the dollar moved five basis points. This is the strongest evidence yet for the judgement this brief has carried for five editions: the yen and the Bank of Japan’s September 18 meeting are still the dominant channel, not the Fed (interpretation).**

⊘⚑ Energy: Brent −2.8% on the day and +8.7% on the week, and the caliber conflict is resolved

  • ⊘ **Brent settled $104.61 (September 11, −2.8%) on diplomatic headlines, after touching roughly $108 on Thursday; +8.7% on the week, and a third consecutive session above $100.**
  • ⚑ **The caliber conflict this brief carried ($101.25 vs $105.71 for September 10) resolves in favour of the HIGHER row: a −2.8% settlement at 104.61 implies roughly $107.6 the session before. The $101.25 caliber is retired as a different contract month (derivation).**
  • ⊘ **Two-sided for aluminium: above $100 it raises the smelting cost floor (bullish) and makes the September 16 decision harder (bearish) (interpretation).**

⚑⊘ China: August exports strong, the aluminium line stronger

  • ⚑ **Exports +25% y/y to $401.44bn; imports +28.2% to $282.36bn; trade surplus $119.09bn.**
  • ⚑ **August unwrought aluminium and product exports 626,000 mt, +17.2% y/y; January–August 4.67 million mt, +16.7% y/y.**
  • ⊘ **The counterweight is recorded with equal weight: an export line growing 16.7% into a world adding tariff barriers is a concentration risk as well as a demand signal (interpretation).**

Interpretation

  • ⊘⊘ **The macro column is one-sidedly bearish this edition: a ~20-point single-day repricing of the Fed path is the common cause of every price move on this page (interpretation).**
  • ⚑ **But the transmission is incomplete: the dollar barely moved while aluminium fell 2%, which reads as a broad risk-asset repricing rather than a dollar-channel move (interpretation).**
  • ⚑⊘ **And the Shanghai–London divergence sits exactly on that logic: the shock was fully priced in London (−2.06%) and only partly accepted in Shanghai (−1.04%), with the ratio up 0.0766 in a single session, the largest of this run (derivation).**

5 · Fundamentals

⊘ Alumina oversupply is the through-line of this edition’s fundamentals

  • ⊘ **CERA puts the 2026 global alumina market in a 1.79 million mt SURPLUS.**
  • ⊘ **China’s alumina imports ran 2.277 million mt in January–June, +749.1% y/y.**
  • ⊘ One caliber has **~8.6 million mt of new domestic alumina capacity planned for 2026 against a forecast full-year oversupply of more than 10 million mt; another gives a 2,700–2,900 yuan/mt range forecast. Neither is precisely dated; both are listed and NOT weighted.**
  • ⊘ **The board agrees: AO2701 through all six averages in one session, complex open interest +20,900, and cost caliber two unpriced across the whole curve. All three point at the supply side, not the demand side (interpretation).**

⚑ Overseas supply: one widely re-circulated item, given its correct date this edition

  • ⊘ **The Rusal Friguia item in Guinea is dated AUGUST 24: a train derailed near Kagbelen in Dubreka prefecture, cutting rail access to the ~450,000 mt/yr refinery, halting exports and slowing production, and disrupting fuel and caustic soda deliveries. It is listed here with its correct date because it has circulated this week without one.**
  • ⚑ Carried: Alunorte restored to full production after the August 50% cut; Rio Tinto Yarwun −1.2 million mt/yr from October; Alcoa’s 2026 guidance cut of 200,000–300,000 mt; Qatalum ~60% of nameplate; Alba lines 1–3 down; EGA Al Taweelah 18% of reduction cells restarted; Guinea’s ~150 million mt/yr export cap target.

⊘ Trade policy: the first measure on this ledger that hits two origins at once

  • ⊘ **Mexico imposed DEFINITIVE anti-dumping duties on aluminium extrusions from China AND the United States (September 10). This is the first item on this ledger targeting both origins simultaneously, and its implication differs from a one-sided measure: it compresses the substitutability of the entire North American extrusion supply chain (interpretation).**
  • ⊘ Carried: the EU’s planned scrap export ban to non-OECD countries; Canada’s 25%→50% step-up with aluminium rod named; Thailand’s 21.94% provisional review; New Zealand’s ALENZ safeguard petition; existing Argentine and Indian measures.

⊘ Downstream: the peak season has not yet shown up in the data

  • ⊘ **Aluminium extrusion capacity utilisation is at a 37.05% cyclical floor; leading downstream processors’ operating rate 59.9% (August 13); secondary alloy operating rate 49.1% (carried).**
  • ⚑ **Mysteel’s August A00 monthly averages: East China ¥23,869/mt (+3.34% m/m), South China ¥24,010/mt (+3.79%), Central China ¥23,762.4/mt (+3.42%). Prices rose in August and operating rates did not follow — the classic signature of high prices suppressing consumption (interpretation).**
  • ⚑ India’s aluminium demand in electronics is projected at a **10.18% CAGR (carried, long-horizon).**

⊘ Spot transaction sentiment: still no publishable index

  • ⊘ **SMM’s regional transaction premia, shipment and purchasing sentiment indices are unavailable for a TWELFTH consecutive edition. This brief therefore publishes no sentiment index and substitutes no proxy.**
  • ⚑ **What can be inferred but not quantified: A00 fell 320 in a session and the basis still converged to +5, so spot followed rather than led; while all five regional premiums narrowed, which reads as holders’ willingness to ship rising faster than downstream willingness to buy (interpretation — explicitly labelled as inference).**

6 · Primary Aluminum Spot Market

⚑ The five regional quotations (September 11, CNAL caliber)

  • ⚑ **Shanghai 24,210–24,250, avg 24,230 (−310); Changjiang 24,220–24,260, avg 24,240 (−320); Zhongyuan 24,240 (−320); Nanhai/Foshan 24,660–24,760, avg 24,710 (−340); Guangdong Nanchu 24,410–24,470, avg 24,440 (−360).**
  • ⚑ **SMM mirror caliber: A00 24,240 (−320); A00 Foshan 24,430 (−350); ADC12 24,300 (−150).**
  • ⊘ **SMM and CNAL differ by 10 yuan on Shanghai (24,240 vs 24,230) — one of the few sessions on which both are available, and the gap is small enough that the two calibers corroborate each other (derivation).**

⊘ Regional structure: a full giveback after four sessions of widening

  • ⊘ **Nanhai − Shanghai 480 (530 at September 9, −50); Nanchu − Shanghai 210 (260, −50); Changjiang − Shanghai +10, UNCHANGED.**
  • ⊘ **South China fell 340–360 against East China’s 310–320 — the strongest region became the weakest in one session.**
  • ⊘ **No quantified sentiment is available this edition, so shipment and purchasing willingness are given qualitatively and labelled as inference (above).**

⚑ Basis and the delivery-month structure

  • ⚑ **Basis +5 against the AL2610 close, +50 against the settlement, +240 against the night last.**
  • ⊘ **AL2609 is in its final stretch: September 11 close 24,240, settlement 24,205, open interest 11,780 (−3,240), night 23,970.**
  • ⚑⚑ **The main-contract handover is happening: in the night AL2611’s open interest (213,268) exceeded AL2610’s (198,952); on the day AL2610 still leads by 3,543. This brief will switch calibers once a DAY session confirms it, and will label the switch explicitly in the edition that makes it (derivation).**

7 · Key Price Drivers

⚑ Bullish factors (17)

  • ⚑⚑ **The named 24,075 held to the tick — the September 11 trading-day low is the same price as the September 10 night low and exactly the August 31 close.**
  • ⚑⚑ **The daytime hours bought back 110 of the night’s 450 (~24.4%), closing 160 above the low and only 80 below the high.**
  • ⚑⚑ **LME cash fell $69 and the backwardation gave up $1.00 (+$25.00 → +$24.00): the physical prompt market did not participate.**
  • ⚑⚑ **The dollar moved 0.05% on a day the Fed path repriced ~20 points — dollar-channel transmission is visibly incomplete.**
  • ⚑⚑ **Shanghai out-performed London by 1.02 percentage points; the ratio rose from 7.3257 to 7.4023, ending three sessions of London outperformance.**
  • ⚑⚑ **The spot test resolved on its middle branch and disproved this brief’s own suspicion: 24,560 − 320 = 24,240 exactly; the mirror rows were real quotations.**
  • ⚑⚑ **The basis converged to +5 — spot followed rather than led.**
  • ⚑⚑ **The September 11 night was a roll, not an exit: six contracts only −1,612, with AL2610’s −12,438 absorbed by AL2611’s +5,421 and others.**
  • ⚑ **MA40 (23,847.25) and MA60 (23,601.25) are intact; even the night last has 152.75 of clearance over MA40.**
  • ⚑ **Three aluminium contracts closed above their opens (zero on September 10) and twelve of twelve closed above their lows.**
  • ⚑ **LME stocks 244,100 mt with ten of seventeen intervals exactly zero — no sign of physical loosening.**
  • ⚑ **Both spot sources are live in the same edition for the first time in twelve; CNAL resumed after a one-edition break.**
  • ⚑ **Brent held above $100 for a third session ($104.61) and rose 8.7% on the week, raising the smelting cost floor.**
  • ⚑ **China’s August aluminium exports 626,000 mt (+17.2% y/y); January–August +16.7%.**
  • ⚑ **China’s August exports +25% with a $119.09bn surplus — no external demand deterioration in the data.**
  • ⚑ Carried: LME stocks near a 36-year low; Wood Mackenzie’s 2026 global deficit of ~900,000 mt (dated August 4); ING on physical tightness.
  • ⚑ Carried overseas supply disruption: Rio Tinto Yarwun, Alcoa’s guidance cut, Qatalum, Alba, EGA, Guinea’s export cap, Rusal Friguia (August 24).

⊘ Bearish factors (33)

  • ⊘⊘ **FedWatch September 16 hike pricing went from ~69.3% to close to 90% — roughly 20 points in one afternoon.**
  • ⊘⊘ **US August core CPI +0.3% m/m (from +0.2%), headline +0.4% m/m and +3.4% y/y.**
  • ⊘⊘ **EY-Parthenon switched from HOLD to a 25 bp HIKE, implying a 3.75%–4.00% target range.**
  • ⊘⊘ **LME cash −$69.00 (−2.06%), the largest one-day fall on this ledger; 3-month −$68.00.**
  • ⊘⊘ **LME cash lost MA5 and MA10, with only $17.95 over MA20 and $10.52 over MA30.**
  • ⊘⊘ **The roll ruling was voided at 1.96x this brief’s own threshold: complex OI −29,349.**
  • ⊘⊘ **On a three-session caliber exit outweighs roll 34,381 to 12,624, a complete inversion of 16,900 to 5,000.**
  • ⊘⊘ **Record complex volume (414,471, +55.4%) alongside heavy reduction and a falling price — distribution.**
  • ⊘⊘ **The September 11 night lost MA20 (−55.00) and MA30 (−29.83); four of six averages are gone on the night caliber.**
  • ⊘⊘ **The night low of 23,920 is the lowest since September 1 and exactly the August 28 close — the September advance is gone intraday.**
  • ⊘⊘ **AO2701 closed below all six of its own averages in one session (by 40.15 to 84.45).**
  • ⊘⊘ **Alumina complex OI +20,900, AO2701 +24,675, night +14,960 more — new shorts.**
  • ⊘⊘ **Cost caliber one failed a second time and the margin went from 3 to 51; the settlement is 23 below the line too.**
  • ⊘⊘ **Cost caliber two is unpriced on all twelve contracts; the highest close (2,815) is 11.21 below it.**
  • ⊘⊘ **Twelve of twelve alumina contracts closed below their opens (eleven the session before).**
  • ⊘ **The front-to-tail contango narrowed a third time to +148 (5.55%).**
  • ⊘ **The matched-pair margin proxy fell a further 162.36 on the day and 204.12 on the night.**
  • ⊘ **Twelve of twelve aluminium settlements fell (−315 to −400).**
  • ⊘ **AL2610’s close lost MA5 (−198.00) and MA10 (−76.00).**
  • ⊘ **The close fell from the 95th/98.3rd to the 70th/85th percentiles.**
  • ⊘ **Main open interest −25,055 (−10.60%), −35,678 over two sessions.**
  • ⊘ **All five regional spot premiums narrowed, with South China leading the fall (−340 to −360).**
  • ⊘ **Nanhai − Shanghai 480 (from 530) and Nanchu − Shanghai 210 (from 260).**
  • ⊘ **The LME average stack inversion widened a third step: MA30 $7.43 above MA20.**
  • ⊘ **No new national inventory print this edition, so the halved-destock ruling is carried rather than refreshed.**
  • ⊘ **SMM’s three-region series is missing for a sixth consecutive edition — the deceleration cannot be located.**
  • ⊘ **The SHFE weekly warrant print is again unavailable in its primary caliber.**
  • ⊘ **CERA puts the 2026 global alumina market in a 1.79 million mt surplus; China’s H1 alumina imports +749.1%.**
  • ⊘ **One caliber has ~8.6 million mt of new Chinese alumina capacity in 2026 and >10 million mt of full-year oversupply (undated, not weighted).**
  • ⊘ **Extrusion capacity utilisation at a 37.05% floor and leading downstream operating rates at 59.9%.**
  • ⊘ **Mexico’s definitive anti-dumping duties hit Chinese AND US extrusions — the first two-origin measure on this ledger.**
  • ⊘ **The October fed funds strip is absent for a fifteenth consecutive edition — no term structure for the Fed path.**
  • ⊘ Carried: Goldman Sachs’ reported $2,350/mt Q4 2026 projection (undated, not weighted), now **~29.1% below the September 11 settlement.**

8 · One-Sentence Market View

Seven pre-written tests resolved in a single session — a first on this ledger — and the grading is five bearish, one split and one where the number passed and the reaction function overruled it: US August CPI printed +0.4% m/m and +3.4% y/y, landing at the top of the pre-written +3.3%–3.4% in-band range, but core came in +0.3% m/m against July’s +0.2% with gasoline up 3.9% and worth over a third of the monthly gain, and FedWatch went from roughly 69.3% to close to 90% for a September 16 HIKE in one afternoon against the 56%/59%/60.6% spread carried all week, with EY-Parthenon switching from hold to a 25 bp hike — so the threshold was right and this brief’s mapping was wrong, a defect recorded as such and corrected from the next edition by keying inflation tests to core m/m and to the FedWatch move itself; London absorbed essentially all of it, cash settling $3,274.00 for −$69.00 and −2.06%, the largest one-day cash fall on this ledger, losing MA5 and MA10 and holding MA20 by just $17.95 and MA30 by $10.52 — yet the backwardation narrowed for the first time in five and only by a single dollar to +$24.00, and stocks fell just 250 mt, below the −500 mt substance threshold written down last edition, which produced a correct and uncontested classification on its very first use, so whatever sold aluminium on Friday it was not the physical prompt market; in Shanghai the named 24,075 held to the tick — the same price as the September 10 night low and exactly the August 31 close — while the confirmation half of the same test failed by 44.50 points at a 24,235 close against the 24,279.50 required, leaving a split ruling, and the daytime hours actually bought back 110 of the night’s 450 and finished 160 above the low; the genuinely bad news is in position and in alumina, because complex open interest fell 29,349, 1.96 times this brief’s own void threshold, formally withdrawing the roll ruling, with the three-session split inverting to 34,381 exit against 12,624 roll from 16,900 roll against 5,000 exit, on the heaviest complex volume on this ledger (414,471, +55.4%) — distribution rather than rotation — while AO2701 closed 2,697, down 61 and 2.21%, below all six of its own averages at once in the first edition in which those six are exact and live, cost caliber one failing a second time with the margin widening from 3 to 51 and cost caliber two now unpriced across the entire twelve-contract curve, and alumina open interest RISING 24,675 on the main and 20,900 on the complex with a further 14,960 in the night, which is new shorts entering rather than longs leaving and a more durable signature than liquidation; the September 11 night then took MA20 by 55.00 and MA30 by 29.83 with a 23,920 low that is the lowest tick since September 1 and exactly the August 28 close, but that same night’s position change was only −1,612 with AL2611 overtaking AL2610 at 213,268 to 198,952 — so the night was a roll and the day was an exit, and the two are recorded separately; the one clean bullish resolution is spot, where SMM A00 at 24,240 (−320) landed in the middle band and where 24,560 − 320 = 24,240 exactly, retroactively proving last edition’s mirror rows were real quotations, withdrawing this brief’s own ‘possible carry’ suspicion as wrong, keeping the +70 September 10 basis, and converging the basis to +5, with CNAL also resuming after a one-edition break though every regional premium narrowed and South China led the fall; and the two most important counterweights are that the dollar moved 0.05% on a day the Fed path repriced twenty points — the strongest evidence yet that the yen and the September 18 Bank of Japan meeting remain the dominant channel — and that Shanghai out-performed London by a full 1.02 percentage points with the ratio up 0.0766, the largest of this run, so the macro shock was priced in London and only partly accepted in Shanghai; taken together this is the weakest page this brief has produced, and Monday’s question is narrow and arithmetical: whether the day session reclaims 24,055.00, or whether 23,920 becomes the new reference (interpretation).

9 · Watchlist for Monday’s Day and Night Sessions (September 14)

  • ⚑⚑ First, whether MA20 (24,055.00) is reclaimed or confirmed lost. The night last of 24,000 is already 55.00 below it and 29.83 below MA30 (24,029.83). The discriminants are written before the numbers: a September 14 close above 24,055.00 makes the night break an overshoot and tips Friday’s split ruling to the bullish side; a close below 23,920 (the September 11 night low, and exactly the August 28 close) upgrades this from a pullback to a trend turn, with MA40 at 23,847.25 as the next reference; a close between 23,920 and 24,055 is recorded as a no-conclusion zone and no ruling is made. References: 24,311.00 (MA10) / 24,235 (Sept 11 close) / 24,075 (the level that held) / 24,055.00 (MA20) / 24,029.83 (MA30) / 24,000 (night last) / 23,920 (night low) / 23,847.25 (MA40).
  • ⚑⚑ Second, whether the exit continues, and when the main-contract handover confirms in a day session. Complex OI −29,349 has already voided the roll explanation. Tightened bands: a further −10,000 or more on the complex makes exit a trend rather than a one-day event and every rally on this ledger should be reweighted accordingly; +5,000 or more makes Friday a one-off liquidation wave; ±5,000 is undetermined. Watch also whether AL2611’s DAY-session open interest passes AL2610’s — it already leads by 14,316 in the night while trailing by 3,543 on the day — and this brief will switch its main-contract caliber, with an explicit label, in the edition where a day session confirms it.
  • ⚑⚑ Third, whether alumina’s new shorts keep adding or take profit. This is the most sharply defined item on the page: price falling while open interest rises. Bands: if AO2701’s open interest keeps rising while price keeps falling, the short trend is established and cost support should be removed from the bullish list entirely; if open interest falls while price rebounds, Friday was the last leg before a squeeze; if AO2610 closes back above 2,718, the cost caliber one failure can be reversed — but only after two consecutive closes, the symmetric condition to the one used to confirm the downgrade. Watch also whether any contract reclaims 2,826.21 (cost caliber two), where none currently sits.
  • ⚑⚑ Fourth, which gives way first in London — the backwardation or the stock row — with this edition already pointing one way. A $69 fall in cash costing the structure $1.00 is the strongest physical evidence on this page. Watch: whether cash holds MA20 $3,256.05 (clearance $17.95) and MA30 $3,263.48 (clearance $10.52); whether the backwardation keeps narrowing, flips to contango (which would void the physical argument) or re-widens (which would strengthen it); and whether any single draw of −500 mt or more appears (the new threshold, under which −250 was correctly classified as noise).
  • ⚑ Fifth, whether pricing keeps rising into the September 16 FOMC, and whether the dollar finally follows. FedWatch is already near 90%. Discriminants: pricing above 95% with DXY above 99.5 means the dollar channel has begun to transmit and is further bearish for aluminium; pricing holding near 90% with DXY still flat around 99 keeps this brief’s judgement intact that the yen and the September 18 Bank of Japan meeting are the dominant channel, and the macro pressure remains partly insulated. If the October fed funds strip becomes available it should be retrieved immediately to restore the term structure.
  • ⚑ Sixth, Monday’s SMM national ingot print — the one column that did not update this edition, and which will now carry two weeks of information. The carried chain is 9/7 802,000 → 9/10 796,000, with Thursday-to-Thursday −19,000 at 52.8% of the prior pace. Bands, rewritten this edition to PARTITION THE LINE COMPLETELY: ≤ 785,000 = the destock re-accelerates; 785,001–792,000 = the current decelerated pace is maintained; 792,001–799,000 = further deceleration; ≥ 799,001 = stall or build. Watch also whether the Mysteel print on the same date stays 26,000–28,000 below SMM — a departure would reopen the constant-definitional-offset finding.
  • ⚑ Seventh, whether spot keeps following rather than leading if the board falls again, and whether regional premiums keep narrowing. Friday gave A00 −320, a +5 basis, and all five premiums narrowing with South China worst. Bands: if the board falls again on Monday and A00 falls by less than the board with the basis moving to a widening premium, the physical-tightness argument strengthens; if A00 falls by more than the board and Nanhai − Shanghai breaks below 450, the regional-tightness leg should be removed from the bullish list altogether. Keep tracking whether SMM’s three-region daily series reappears after a sixth consecutive absence, and whether the SHFE weekly warrant print returns in its primary caliber.

Disclaimer

  • This brief is for internal morning-meeting reference only. Sources include the Shanghai Futures Exchange’s public day and night session data and third-party mirrors of it, Westmetall, TradingEconomics, CME FedWatch, the US Bureau of Labor Statistics, Reuters, CNBC, CNAL, AL Circle, Mysteel, SMM (Shanghai Metals Market), China’s General Administration of Customs (as reported), Mining.com / Mining Weekly, CERA, Wood Mackenzie, ING, and publicly reported Goldman Sachs and EY-Parthenon views. Items marked “⚑ / ⊘” are factual data; items marked “(interpretation)” or “(derivation)” are this brief’s judgement or arithmetic, and the two are distinguished throughout. Where sources conflict, all calibers are **listed in parallel and NOT reconciled unless a primary source disproves one. Derivations that depend on an assumed coefficient (such as 1.93 t alumina per tonne of aluminium) state the assumption before the calculation. This edition makes three self-corrections and records each explicitly: first, last edition’s suspicion that the September 10 mirror spot rows were carried rather than quoted is DISPROVED (24,560 − 320 = 24,240 exactly), the flag is withdrawn and the +70 September 10 basis stands; second, this brief’s US CPI band was keyed to the wrong series — the headline y/y rather than the core m/m — so the number landed inside the band while the market’s reaction landed outside it, a defect recorded and corrected from the next edition; third, the two-caliber Brent conflict is resolved in favour of the higher row, and the $101.25 caliber is retired as a different contract month. In addition, the −500 mt LME stock substance threshold written down last edition produced a correct classification on its first use, which is recorded; and the sixth watchlist item’s inventory bands have been rewritten to partition the line completely, so that the defect which let 796,000 mt fall into a gap on September 10 cannot recur. This brief does not constitute investment advice and accepts no responsibility for any decision made on the basis of it.**