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

The top test resolves bearish: SHFE makes a new high at 24,675 and closes back at 24,490. The night session breaks 450 lower to 24,125.顶部检验偏空落地:沪铝创新高 24,675 后收回至 24,490;夜盘下挫 450 元至 24,125。

Friday, September 11, 2026
Morning Meeting晨会参考
Written the morning of Friday September 11 (Asia), BEFORE the SHFE day session opens. ⚑ WHAT IS NEW THIS EDITION: the complete SHFE September 10 day session for all twelve aluminium and all twelve alumina contracts (open, high, low, close, settlement, volume, open interest); ⚑⚑ the COMPLETE September 10 NIGHT SESSION for six aluminium and six alumina contracts, whose ten prior-settlement fields reconcile EXACTLY against the September 10 day settlements; ⚑⚑ a 223-row retrieved SHFE close series, so ALL SIX SHFE averages are exact for the FIRST time on this ledger and the September 4 basis is verified to the cent; ⚑⚑ SMM's Thursday September 10 national ingot social-inventory print, the number this brief published a band for in advance; ⚑ the Mysteel national caliber for September 3, 7 and 10, which resolves the long-running SMM-Mysteel spread; the Westmetall September 10 row and sixty retrieved settlement rows, so ALL SIX LME averages are exact for a THIRD consecutive edition; US August PPI; and the US August CPI consensus. ⊘ NOT RETRIEVABLE: SMM's own September 10 spot quotation set (login-walled for an ELEVENTH consecutive edition — a third-party mirror caliber is listed in parallel and flagged); the CNAL five-caliber regional spot set for September 10, a FIRST break in that series; SMM's three-region daily inventory series, a FIFTH consecutive edition; the SHFE Friday weekly warehouse and warrant print; 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; and the October fed funds strip, absent for a FOURTEENTH consecutive edition.
Steven Yuan
⚑⚑⊘ THE TOP TEST RESOLVED — ON THE EXACT BEARISH BRANCH THIS BRIEF NAMED Yesterday's text: 'close above 24,665 and the September 9 giveback is consolidation; close below 24,560 and 24,665 becomes the short-term top.' AL2610 first made a NEW high of 24,675, 10 above 24,665 — then closed 24,490, which is 70 BELOW 24,560. The second branch fired. This is the cleanest test resolution on this brief's recent ledger and it went the bearish way
⊘ THE SEPTEMBER 10 DAY SESSION IN FULL AL2610: open 24,590, high 24,675, low 24,420, close 24,490, settlement 24,575, −70 (−0.28%) against the 24,560 prior settlement, volume 144,380 (−28,468, −16.5%), open interest 236,445 (−10,623, −4.30%). Close-to-close −70. The giveback from the high WIDENED to 185 from 105
⚑⊘ ONE HONEST COUNTERWEIGHT: THE SETTLEMENT SITS 85 ABOVE THE CLOSE Settlement 24,575 is +15 against the prior settlement even though the close is −70. A settlement that far above the close means the volume-weighted session traded HIGHER and the selling was concentrated at the end. On a settlement caliber September 10 was a small UP day; on a close caliber it was a down day. Both are recorded (derivation)
⊘ TWELVE OF TWELVE ALUMINIUM CONTRACTS CLOSED BELOW THEIR OPENS A complete inversion of September 9's twelve-for-twelve ABOVE, and a repeat of September 4's sweep. Twelve of twelve ALSO closed below their highs, by 165 / 185 / 195 / 190 / 180 / 185 / 170 / 180 / 190 / 125 / 140 / 150 — a band of 125–195 against 80–140 on September 9. The failure widened on both measures (derivation)
⊘⊘ THEN THE NIGHT SESSION BROKE — THE LARGEST NIGHT MOVE ON THIS LEDGER AL2610 night: open 24,310 (a 180-point gap BELOW the close), high 24,315, low 24,075, last 24,125 — −450 (−1.83%) against the 24,575 settlement and −365 against the close. Volume 122,943, open interest 218,334, a FURTHER −18,111 on the day session
⊘ AND IT WAS A PARALLEL SHIFT, NOT A FRONT-END EVENT Six of six retrieved contracts fell between 1.75% and 1.83%: AL2609 24,120 (−435), AL2610 24,125 (−450), AL2611 24,135 (−430), AL2612 24,105 (−445), AL2701 24,090 (−440), AL2702 24,100 (−440). A uniform move down the whole curve is repricing, not a squeeze release (interpretation)
⚑ THE NIGHT BOARD RECONCILES TO THE CENT — TEN OF TEN Every retrieved night quote carries a prior-settlement field, and all ten match the September 10 day settlements exactly: AL 24,555 / 24,575 / 24,565 / 24,550 / 24,530 / 24,540 ✓ and AO 2,721 / 2,737 / 2,748 / 2,783 ✓. The night data is verified, not inferred (derivation)
⚑⚑ ALL SIX SHFE AVERAGES ARE EXACT FOR THE FIRST TIME ON THIS LEDGER A 223-row close series was retrieved. MA5 24,444.00, MA10 24,279.50, MA20 24,037.50, MA30 24,009.33, MA40 23,821.50, MA60 23,599.08. The September 10 close clears all six by 46.00 / 210.50 / 452.50 / 480.67 / 668.50 / 890.92. MA20 through MA60 have been CARRIED on a September 4 basis for six editions; they are now live
⚑ AND THE SAME SERIES VERIFIES THE CARRIED BASIS — THEN CORRECTS ONE NUMBER Recomputed at September 4 the series returns 23,981.75 / 23,859.50 / 23,693.38 / 23,576.67 — EXACTLY the four values this brief has carried. That isolates the single error: the September 7 CLOSE is 24,420, not 24,345. 24,345 was the SETTLEMENT. Restated: MA5 at September 9 was 24,420.00, not 24,405.00; MA10 was 24,212.50, not 24,204.50 (derivation)
⊘ THE NIGHT SESSION LOST MA5 AND MA10 AND IS HOLDING MA20 24,125 sits 319.00 BELOW MA5 (24,444.00) and 154.50 BELOW MA10 (24,279.50), but 87.50 ABOVE MA20 (24,037.50). The night low of 24,075 came within 37.50 of MA20 — and is EXACTLY the August 31 close. The whole bullish structure below MA20 is still intact; everything above it is gone (derivation)
⊘ THE CLOSE FELL OUT OF THE 100TH PERCENTILE FOR THE FIRST TIME IN THREE SESSIONS 24,490 sits at the 95th percentile of the twenty-session close distribution and the 98.3rd of the sixty-session, against 100th and 100th on September 9. The window high remains September 9's 24,560; the window low remains July 1's 22,445 (derivation)
⊘⚑ OPEN INTEREST: THE SECOND CONSECUTIVE FALL, AND NOW ON A FALLING PRICE Main OI 247,068 → 236,445 (−10,623) on a DOWN day, which is the long-liquidation signature — the mirror image of September 9's short-covering read. Over two sessions the main contract shed 21,950 (258,395 → 236,445, −8.5%) while the complex shed only 5,032 (628,099 → 623,067). So ~16,900 lots rolled and ~5,000 left. The roll ruling survives the second session (derivation)
⊘ BUT THE NIGHT OPEN INTEREST IS NOT A ROLL Across the six retrieved contracts, day 591,852 → night 571,416, −20,436 — and AL2610 alone is −18,111 of it while AL2611 fell 1,828, AL2612 238, with AL2701 and AL2702 UP 139 and 272. Position was reduced on a 1.83% decline and it was not absorbed by the back months (derivation)
⚑⚑ SMM'S THURSDAY PRINT ARRIVED: 796,000 MT — AND IT LANDS IN THIS BRIEF'S OWN BAND GAP The pre-written bands were 775,000–785,000 (confirm), 786,000–795,000 (deceleration) and above 800,000 (stall). 796,000 falls in the 5,000-wide GAP between the second and third. The ruling has to be written honestly: the 36,000 mt/week pace FAILED confirmation; the stall threshold was NOT reached; the print sits one thousand above the deceleration band
⊘ AND THE ARITHMETIC IS UNAMBIGUOUS EVEN IF THE BAND WAS NOT September 3 815,000 → September 10 796,000 = −19,000 Thursday-to-Thursday, against a two-observation pace of 36,000. That is 52.8% of the prior pace and 79.4% of August's own ~23,935. Monday-to-Thursday is only −6,000 against the prior Thursday-to-Monday −13,000. The destock did not stop; it halved (derivation)
⚑ THE GOOD NEWS IN THE INVENTORY COLUMN IS A CALIBER, NOT A LEVEL Mysteel prints 768,000 mt the same day, with its own chain Aug 31 809,000 → Sept 3 789,000 → Sept 7 774,000 → Sept 10 768,000. The SMM−Mysteel spread across four dates is 28,000 / 26,000 / 28,000 / 28,000. That is a DEFINITIONAL OFFSET, not a data conflict, and the long-standing 'unreconciled spread' flag is retired
⚑ AND BOTH CALIBERS AGREE ON THE DIRECTION OF THE DECELERATION Mysteel Thursday-to-Thursday is −21,000 against SMM's −19,000, and Mysteel's own step chain reads −20,000, −15,000, −6,000. Two independent national calibers now say the same thing about the same week. The cumulative draw from the early-July 1.13 million mt phase high is −334,000 mt, −29.6% (derivation)
⊘ ALUMINA'S COST CALIBER ONE FAILED — AFTER HOLDING BY TWO AO2610 closed 2,715, which is THREE BELOW the 2,718 line. The clearance chain is now +39 (Sep 4) → +19 (Sep 8) → +2 (Sep 9) → −3 (Sep 10). The settlement of 2,721 still holds it by 3, and AO2609 closed 2,688, 30 below. The 'alumina cost supports the aluminium floor' leg is DOWNGRADED, exactly as the pre-written condition required
⚑⊘ AO2701 IN ITS FIRST EDITION AS MAIN CONTRACT Open 2,778, high 2,787, low 2,753, close 2,758, settlement 2,770, −21 (−0.75%) against the 2,791 prior settlement, volume 162,739 (−21.1%), open interest 195,391 (+7,733). It is the only aluminium-or-alumina main contract on this page whose open interest ROSE
⚑ THE ALUMINA CURVE WENT FULLY MONOTONE FOR THE FIRST TIME ON THIS LEDGER 2,688 / 2,715 / 2,730 / 2,742 / 2,758 / 2,771 / 2,787 / 2,815 / 2,833 / 2,850 / 2,857 / 2,863 — ZERO inversions, against one on September 9 and two on September 8. ⊘ But the front-to-tail contango narrowed to +175 (6.51% of the front) from +184 (6.83%), and eleven of twelve contracts closed below their opens (derivation)
⊘ COST CALIBER TWO MOVED BACK OUT The first close above 2,826.21 is AO2705 (2,833), having been AO2704 on September 9. Forward cost inflation is priced eight months out again rather than seven — the September 9 improvement lasted one session. AO2704 closed 2,815, now 11.21 below the line (derivation)
⚑⚑ THE LARGEST DIVERGENCE ON THE PAGE IS IN THE NIGHT SESSION Aluminium fell 1.75% to 1.83% across six contracts. Alumina fell 0.07% to 0.43% across six: AO2610 2,719 (−2), AO2611 2,737 (0, FLAT), AO2612 2,744 (−4), AO2701 2,761 (−9), AO2702 2,771 (−12), AO2609 2,688 (−8). The metal sold off and its cost input did not follow (derivation)
⊘⊘ WHICH MEANS THE MARGIN PROXY COMPRESSED 372.72 IN A SINGLE NIGHT On a matched AL2610/AO2610 pair at the stated 1.93 t/t assumption: day 24,490 − 1.93 × 2,715 = 19,250.05; night 24,125 − 1.93 × 2,719 = 18,877.33. A −372.72 move in one session is the largest on this brief's ledger. ⚑ Day-session curve compression, on the unchanged front/tail caliber, actually NARROWED to 402.75 from 420.12 (derivation)
⚑⊘ LONDON SPLIT AGAIN, AND THE SPLIT IS WIDENING Cash settled $3,343.00, −$9.00 (−0.27%) — still the second-highest settlement in twenty-one sessions. But the 3-month settled $3,318.00, −$20.00, so the backwardation WIDENED a FOURTH consecutive step: −$0.50 → +$2.00 → +$6.00 → +$14.00 → +$25.00. Cash fell and prompt tightness increased on the same row
⊘⚑ AND THE STOCK ROW MOVED FOR THE FIRST TIME IN FOUR 244,350 mt at September 10, −175. The pre-written reversal condition was 'a new three-figure draw means the event ruling is re-examined.' −175 IS a three-figure draw, so the condition is met on its LETTER — and 175 mt is 0.07% of stock and one eighth of September 4's −1,450, so it is not met on its SUBSTANCE. The 'event, not trend' ruling STANDS and the band is recorded as badly drawn
⊘ THE LONDON FLOW ARITHMETIC BARELY MOVES SIXTEEN-INTERVAL FLOW: August 18 246,925 → September 10 244,350 = −2,575 mt (−1.04%), ten of sixteen intervals exactly zero, September 4's −1,450 still 56.3% of the whole. SIXTY-SESSION: June 17 316,525 → 244,350 = −72,175 mt (−22.80%), but ~161 mt per interval since August 18 against ~1,582 before — 10.2% of the prior pace (derivation)
⚑ ALL SIX LME AVERAGES ARE EXACT FOR A THIRD CONSECUTIVE EDITION MA5 $3,324.50, MA10 $3,287.65, MA20 $3,256.30, MA30 $3,262.05, MA40 $3,241.44, MA60 $3,220.08. Cash clears all six by 18.50 / 55.35 / 86.70 / 80.95 / 101.56 / 122.92. ⊘ The non-monotone stack WIDENED: MA30 is now $5.75 above MA20, against $2.99 on September 9 (derivation)
⚑ THE RATIO FELL A THIRD SESSION — BUT BARELY 24,490 ÷ $3,343.00 = 7.3257 at September 10, against 7.3270 and 7.3564. SHFE −0.28% close-to-close against LME cash −0.27%: London out-performed for a third consecutive session by the smallest margin of the three (derivation)
⊘ THE SPOT SET IS THE WEAKEST DATA ON THIS PAGE AND IS FLAGGED AS SUCH SMM was login-walled for an ELEVENTH edition and the CNAL five-caliber set did not publish for September 10 — the first break in that series on this ledger. A third-party mirror gives SMM A00 24,560, Foshan 24,780 and ADC12 24,450 with a change of ZERO on all three, which is exactly what a CARRIED row looks like. Listed, and NOT weighted
⚑⊘ IF THE MIRROR IS TAKEN AT FACE VALUE, THE BASIS FLIPPED TO A PREMIUM 24,560 − 24,490 = +70. The chain would be +70 (Sep 4) → +55 (Sep 7) → −20 (Sep 8) → −20 / 0 (Sep 9) → +70 (Sep 10). Yesterday's band was '0 to −20 is caught up'. +70 is ABOVE the band, which reads as the board falling and spot NOT following — but only if the mirror is a real quote and not a carry (interpretation)
⚑ THE DOLLAR FELL A FOURTH CONSECUTIVE SESSION DXY ~98.70 in Asian hours on September 10, a fourth straight decline and a fourth session below 99, in a four-month low area; a parallel caliber has it holding ~98.8. Month −1.03% carried. ⊘ The channel is still the yen and the Bank of Japan's September 18 meeting rather than the Fed (derivation)
⊘ US AUGUST CPI LANDS TONIGHT, AND THE FED IS STILL A THREE-WAY SPREAD CPI at 08:30 ET September 11 = 20:30 Beijing, the last inflation print before September 16. Consensus: headline +0.4% m/m and +3.4% y/y (one caliber says it EASES to +3.3%); core +0.4% m/m / +2.4% y/y on one caliber and +0.22% m/m on another — LISTED AND NOT RECONCILED. FedWatch calibers: ~56%, 59%, 60.6%. ⊘ The October strip is absent for a FOURTEENTH edition
⚑⊘ US AUGUST PPI BEAT ON THE HEADLINE AND MISSED ON THE CORE +0.4% m/m and +5.4% y/y against a +5.3% consensus and a +4.7% prior; core +0.2% against +0.3%. Final demand goods +1.1% against services +0.1%, final demand energy +4.2%, diesel +24.1% m/m. For aluminium this is an energy-cost pass-through that has NOT yet reached core — it raises the smelting cost floor with a lag (interpretation)
⊘ AND CRUDE HELD ABOVE $100 A SECOND SESSION Brent $101.25 at September 10, +0.04%. ⊘ A second caliber carries $105.71 for the same date, a 4.4% gap, listed and NOT reconciled and given no weight. Above $100 the read is unchanged and two-sided: a higher energy floor under smelting, and a harder September 16 decision (interpretation)

一 · 行情表现

前置声明:写稿时点与本期性质

  • 本期写于 9月11日(周五)上午,早于沪市日盘开盘。
  • ⚑ **本期新增:9月10日沪市日盘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;⚑⚑ 9月10日完整夜盘(六个铝合约与六个氧化铝合约),其十个前结算价字段与 9月10日日盘结算价逐一精确对上;⚑⚑ 取回 223 行沪铝收盘价序列,本简报台账上首次实现六条沪铝均线全部为精确值,且 9月4日基准被逐分验证;⚑⚑ SMM 周四(9月10日)全国铝锭社会库存——即本简报上一期事先公布判别区间的那个数字;⚑ Mysteel 全国口径 9月3日、7日、10日三个数据,由此解决了长期存在的 SMM–Mysteel 价差问题;Westmetall 9月10日行及六十行结算价,LME 六条均线连续第三期精确;美国 8月 PPI;以及美国 8月 CPI 市场预期。**
  • ⊘ **无法取得:SMM 自身 9月10日现货报价(连续第十一期登录墙——以第三方镜像口径平行列示并加注);CNAL 五口径区域现货 9月10日数据缺失,为该序列在本台账上的首次中断;SMM 三地日度库存,连续第五期;上期所周五周度仓单与库存;9月10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;10月联邦基金利率条,连续第十四期缺失。**

⚑⚑⊘ 顶部检验落地——且恰好落在本简报点名的那个偏空分支上

  • ⚑⚑ **AL2610 9月10日:开 24,590,高 24,675,低 24,420,收 24,490,结算 24,575,较 24,560 前结算 −70(−0.28%),成交量 144,380(−28,468,−16.5%),持仓量 236,445(−10,623,−4.30%)。收盘对收盘 −70。**
  • ⚑⚑ **盘中高点 24,675 为六十日窗口新高,高于 9月9日的 24,665 达 10 点,高于 9月8日高点 125 点,高于 6月5日高点 155 点。**
  • ⊘⊘ **随后收于 24,490,低于昨日检验点名的 24,560 这条线 70 点。昨日原文:「收在 24,665 之上,则 9月9日的回吐属整理;收在 24,560 之下,则 24,665 成为本轮短期顶」。偏空分支按其自身条款触发,本期无需再记「两半」。**
  • ⊘ **自高点的回吐由 105 扩大至 185。**
  • ⊘ **十二个铝合约全部收在自身开盘价之下——与 9月9日的十二比零完全反转;十二个合约同时收在自身最高价之下,回吐幅度依次为 165 / 185 / 195 / 190 / 180 / 185 / 170 / 180 / 190 / 125 / 140 / 150,区间 125–195,较 9月9日的 80–140 扩大(derivation)。**
  • ⚑⊘ **一条必须同等记录的反证:结算价 24,575 高于收盘价 85 点,且较前结算 +15。结算价高出收盘价如此之多,意味着成交量加权的盘中价格高于收盘,卖压集中在尾盘。以结算价口径计,9月10日是小幅上涨日;以收盘价口径计,是下跌日。两者并列(derivation)。**

⊘⊘ 夜盘破位:−450 点,本台账最大单次夜盘跌幅

  • ⊘⊘ **AL2610 夜盘:开 24,310(较日盘收盘跳空低开 180 点),高 24,315,低 24,075,最新 24,125——较 24,575 结算价 −450(−1.83%),较收盘 −365。成交量 122,943,持仓量 218,334,较日盘再减 18,111。**
  • ⊘ **且是平行下移,不是前端事件:取回的六个合约全部下跌 1.75% 至 1.83%——AL2609 24,120(−435)/ AL2610 24,125(−450)/ AL2611 24,135(−430)/ AL2612 24,105(−445)/ AL2701 24,090(−440)/ AL2702 24,100(−440)。整条曲线同幅下移是重新定价,不是逼仓解除(interpretation)。**
  • ⚑ **夜盘数据经过验证而非推断:十个前结算价字段与 9月10日日盘结算价全部精确对上——铝 24,555 / 24,575 / 24,565 / 24,550 / 24,530 / 24,540 ✓,氧化铝 2,721 / 2,737 / 2,748 / 2,783 ✓(derivation)。**
  • ⊘ **夜盘持仓不是移仓:六个合约日盘合计 591,852 → 夜盘 571,416,−20,436;其中 AL2610 一个合约就占 −18,111,AL2611 −1,828、AL2612 −238,而 AL2701 +139、AL2702 +272。在 1.83% 的跌幅中减仓,且未被远月吸收(derivation)。**

⚑⚑ 六条沪铝均线本台账首次全部为精确值,并附一处更正

  • ⚑⚑ 取回 **223 行收盘价序列:MA5 24,444.00 / MA10 24,279.50 / MA20 24,037.50 / MA30 24,009.33 / MA40 23,821.50 / MA60 23,599.08。9月10日收盘价 24,490 分别高出 46.00 / 210.50 / 452.50 / 480.67 / 668.50 / 890.92。MA20 至 MA60 已被沿用六期,本期起转为实时。**
  • ⚑ **同一序列验证了沿用基准,并因此定位出唯一错误:在 9月4日重算得 23,981.75 / 23,859.50 / 23,693.38 / 23,576.67,与本简报一直沿用的四个数字完全一致。故错误仅有一处:9月7日收盘价为 24,420,而非 24,345——24,345 是当日结算价。更正后:9月9日 MA5 应为 24,420.00(原 24,405.00),MA10 应为 24,212.50(原 24,204.50)(derivation)。**
  • ⊘ **日盘低点 24,420 已下破 MA5(24,444.00)24.00 点;夜盘最新价 24,125 低于 MA5 319.00、低于 MA10 154.50,但高于 MA20(24,037.50)87.50,而夜盘低点 24,075 距 MA20 仅 37.50,且恰为 8月31日收盘价(derivation)。**
  • ⊘ **收盘分位三期以来首次跌出百分位顶端:24,490 位于二十日收盘分布的 95 百分位、六十日的 98.3 百分位(9月9日均为 100)。窗口高点仍为 9月9日的 24,560,低点仍为 7月1日的 22,445(derivation)。**

⊘⚑ 持仓:连续第二期下降,且这一次是在价格下跌中

  • ⊘ 主力持仓 247,068 → **236,445(−10,623),在下跌日中减少——这是多头减仓的特征,与 9月9日上涨中减仓的空头回补读法正好相反。**
  • ⚑ 两期合计:主力 258,395 → 236,445,−21,950(−8.5%);而全系 628,099 → 623,067,仅 **−5,032。即约 16,900 手移仓、约 5,000 手离场,移仓的定性在第二期继续成立(derivation)。**
  • ⊘ 全系成交量 **266,651 手(−96,464,−26.6%);主力成交 144,380(−16.5%)。创新高的那一天成交最重,失守的这一天成交明显收缩。**

⚑⊘ 伦敦:现货回落,但 backwardation 第四次走阔

  • ⚑⊘ 现货结算 **3,343.00 美元(−9.00,−0.27%)——仍为二十一个交易日中的第二高结算价。**
  • ⚑⚑ 三个月结算 **3,318.00 美元(−20.00),故 backwardation 第四次连续走阔:−0.50 → +2.00 → +6.00 → +14.00 → +25.00。现货下跌与近端趋紧同日发生(derivation)。**
  • ⚑ **六条 LME 均线连续第三期精确:MA5 3,324.50、MA10 3,287.65、MA20 3,256.30、MA30 3,262.05、MA40 3,241.44、MA60 3,220.08,现货分别高出 18.50 / 55.35 / 86.70 / 80.95 / 101.56 / 122.92。**
  • ⊘ **均线倒挂扩大:MA30 已高于 MA20 5.75 美元(9月9日为 2.99)(derivation)。**
  • ⚑ **比值连续第三期下降,但幅度最小:24,490 ÷ 3,343.00 = 7.3257(前值 7.3270、7.3564)。沪铝 −0.28% 对伦铝 −0.27%(derivation)。**
  • ⊘ 平行口径,列示且不并入任何均线:9月11日实时三个月 **3,287 美元(−6.50,−0.20%);另一口径 3,296 美元(−0.65%),区间 3,278–3,321,开 3,316。两者互不调和。**

二 · 铝价及相关市场

⊘⊘ 现货:本台账最弱的一页,两条序列同期中断

  • ⊘⊘ **SMM 连续第十一期登录墙,且 CNAL 五口径区域现货 9月10日未发布——后者为该序列在本台账上的首次中断。本简报所依赖的两个现货来源在同一期同时失效,此处如实陈述,不作弥缝。**
  • ⊘ **第三方镜像口径(9月10日),列示但不赋权:SMM A00 24,560,涨跌为零;SMM A00 佛山 24,780,涨跌为零;ADC12 24,300–24,600、均价 24,450,涨跌为零。在盘面下跌 70 点的一天,同一来源出现三个「零变动」行,是「沿用行」而非「报价」的特征,故加注。**
  • ⚑⊘ **若按面值采信,则基差为 24,560 − 24,490 = +70,链条读作 +70(9/4)→ +55(9/7)→ −20(9/8)→ −20 / 0(9/9)→ +70(9/10)——高于昨日「0 至 −20 为跟上」的判别带,即盘面下跌而现货没有跟随。此项记为条件性结论,在真实报价确认之前不予赋权(interpretation)。**
  • ⊘ **Mysteel 区域口径(9月10日 22:30),平行列示:巩义 24,430;华南 24,800。对照镜像口径的上海 24,560,隐含华南−上海价差 240(9月9日 CNAL 口径为 260),但两者口径不同,不就价差收敛作任何结论。**
  • ⊘ 沿用 9月9日 CNAL 全套:上海 24,540 / 长江 24,550 / 中原 24,550 / 南海 25,070 / 广东南储 24,800;价差 南海−上海 530、南储−上海 260、长江−上海 +10。
  • ⊘ **SMM 区域成交升贴水、出货与采购意愿指数连续第十一期不可得,故本简报仍然无法发布意愿指数——此处明言,而非以代理指标替代。**

⊘ 氧化铝:成本口径一失守三点,AO2701 首次以主力身份入账

  • ⊘⊘ **AO2610 9月10日:开 2,711,高 2,733,低 2,711,收 2,715,结算 2,721,较 2,736 前结算 −15(−0.55%),成交 39,698,持仓 58,412(−9,934)。**
  • ⊘⊘ **成本口径一 2,718 失守:收盘价 低于该线 3 点。余量链条为 +39(9/4)→ +19(9/8)→ +2(9/9)→ −3(9/10)。结算价 2,721 仍高出 3 点;AO2609 收 2,688,低于该线 30 点。「氧化铝成本支撑铝价底部」一项按事先写下的条件降一档。**
  • ⚑⊘ **AO2701(新主力)9月10日:开 2,778,高 2,787,低 2,753,收 2,758,结算 2,770,较 2,791 前结算 −21(−0.75%),成交 162,739(−21.1%),持仓 195,391(+7,733)——是本页唯一持仓上升的主力合约。**
  • ⚑ **氧化铝收盘曲线本台账首次完全单调:2,688 / 2,715 / 2,730 / 2,742 / 2,758 / 2,771 / 2,787 / 2,815 / 2,833 / 2,850 / 2,857 / 2,863,倒挂为零(9月9日一处、9月8日两处)。**
  • ⊘ 但首尾 contango 由 +184(占前端 6.83%)收窄至 **+175(6.51%);十一个合约收在自身开盘价之下,仅 AO2610 收在其上;全系持仓 441,597(−1,830),全系成交 243,792(−116,187,−32.3%)。**
  • ⊘ **成本口径二 2,826.21 首次被超越的位置由 AO2704 外移回 AO2705(收 2,833),即远期成本通胀重新定价在八个月之外而非七个月。9月9日的改善只持续了一期(derivation)。**

⚑⚑ 本页最大的背离:同一夜盘,铝跌 1.8%,氧化铝几乎不动

  • ⚑⚑ 氧化铝夜盘六个合约:AO2609 2,688(−8)/ AO2610 **2,719(−2,−0.07%)/ AO2611 2,737(0,持平)/ AO2612 2,744(−4)/ AO2701 2,761(−9,−0.32%)/ AO2702 2,771(−12,−0.43%)。AO2701 夜盘持仓 203,668(+8,277),成交 90,883。**
  • ⊘⊘ **配对口径夜盘利润代理(AL2610/AO2610,1.93 吨/吨,假设先行声明):日盘 24,490 − 1.93 × 2,715 = 19,250.05;夜盘 24,125 − 1.93 × 2,719 = 18,877.33,单场 −372.72,为本台账最大(derivation)。**
  • ⚑ **日盘曲线口径的压缩反而收窄:前端 19,307.16(−61.49)、尾端 18,904.41(−44.12),压缩 402.75(前值 420.12),收窄 17.37(derivation)。**
  • ⊘ **金属下跌而其成本投入不跟随,意味着这一跌幅由冶炼环节全额承担(interpretation)。**

三 · 库存

⚑⚑ SMM 周四库存落地 79.6 万吨——落在本简报自己的判别带空隙里

  • ⚑⚑ **9月10日:796,000 吨,较 9月7日的 802,000 −6,000,较 9月3日的 815,000 −19,000。完整链条:8/24 860,000 → 8/27 852,000 → 8/31 837,000 → 9/3 815,000 → 9/7 802,000 → 9/10 796,000。**
  • ⊘⊘ **判别结果(判别带在数字公布前一期即已写下):775,000–785,000 为确认;786,000–795,000 为减速;800,000 以上为停滞。796,000 落在第二带与第三带之间那个 5,000 吨宽的空隙里。判定:确认失败,停滞未至,该数字高出减速带上沿 1,000 吨。一组不能完全覆盖数轴的判别带本身就是方法缺陷,本期予以记录,并自下一期起修正。**
  • ⊘ **速度(derivation):周四对周四 −19,000,为 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,而此前周四至周一为 −13,000。去库没有停止,它减半了,并且已低于 8月。**
  • ⚑ 自 7月初 113 万吨的阶段高点累计 **−334,000 吨(−29.6%)。**

⚑⚑ 本栏真正的好消息是口径,不是水平

  • ⚑⚑ **Mysteel 全国口径同日报 768,000 吨,其自身链条为 8/31 809,000 → 9/3 789,000(−20,000)→ 9/7 774,000(−15,000)→ 9/10 768,000(−6,000)。**
  • ⚑⚑ **SMM 减 Mysteel,四个日期依次为 28,000 / 26,000 / 28,000 / 28,000——这是恒定的定义性偏移量,而非数据冲突。本简报自 8月31日起沿用的「未调和价差」标记予以撤销。**
  • ⚑ **两个独立全国口径对同一周的方向判断一致:Mysteel 周四对周四 −21,000,SMM −19,000(derivation)。**

⊘ 伦敦:四期以来首次变动,但只有 −175 吨

  • ⊘ **244,350 吨(9月10日,−175)。事先写下的反转条件是「出现新的三位数抽库,则本期定性需重新审视」。−175 满足该条件的字面,不满足其实质(占库存 0.07%,仅为 9月4日 −1,450 的八分之一)。「事件而非趋势」的判定维持,而该条件本身记为设定不当(与上文的库存判别带同一性质的方法问题)。**
  • ⊘ **十六区间流量(derivation):8月18日 246,925 → 9月10日 244,350 = −2,575 吨(−1.04%);十六个区间中十个恰为零;9月4日的 −1,450 仍占全部的 56.3%。**
  • ⊘ **六十日流量(derivation):6月17日 316,525 → 244,350 = −72,175 吨(−22.80%);8月18日之前约 1,582 吨/区间,之后约 161 吨/区间,为此前速度的 10.2%。**

解读

  • ⊘ **本页最硬的多头证据本期降级了一档,但没有作废:去库仍在进行,只是速度减半并已低于 8月,且两个独立口径同时确认了这一点(interpretation)。**
  • ⚑⊘ **伦敦仍是本页最清晰的矛盾:仓单四期只动了 175 吨,而现货升水连续第四次走阔至 25.00 美元。当仓单几乎不动而近端升水持续走阔时,更可能是仓单在衡量错误的东西(interpretation)。**
  • ⚑ **一项方法上的进步应与一项方法上的缺陷同时记录:本期解决了 SMM–Mysteel 口径问题(进步),同时暴露了判别带不能覆盖数轴的问题(缺陷)。两者都写在页面上(interpretation)。**

四 · 宏观因素

⚑ 美元连续第四日下跌,但通道仍不在联储

  • ⚑ **DXY 约 98.70(9月10日亚洲时段),连续第四个交易日下跌、连续第四日位于 99 下方,处四个月低位区;另一口径称其在 98.8 附近企稳。月度 −1.03%(沿用)。**
  • ⊘ **named driver 仍是日元与 9月18日日本央行会议,而非联储。对铝偏多,但其原因与美国利率无关(derivation)。**

⊘ 美国 8月 CPI 今晚落地,而联储定价仍是三个口径

  • ⊘ **CPI:9月11日 08:30 美东 = 北京时间 20:30,为 9月16日前最后一条通胀数据。**
  • ⊘ **市场预期,口径冲突且不予调和:整体 +0.4% 环比、+3.4% 同比(另一口径称同比回落至 +3.3%);核心 +0.4% 环比 / +2.4% 同比(一个口径)与 +0.22% 环比(另一个口径)。两组核心口径互相矛盾,平行列示。**
  • ⊘ **CME FedWatch 三个口径:约 56%、59%、60.6%(9月8日)定价 25 个基点加息。缄默期在跑。10月联邦基金利率条连续第十四期缺失。**
  • ⚑⊘ **美国 8月 PPI(9月10日发布):+0.4% 环比、+5.4% 同比,高于 +5.3% 预期与 +4.7% 前值;核心 +0.2%,低于 +0.3% 预期。最终需求商品 +1.1% 对服务 +0.1%;最终需求能源 +4.2%;柴油单月 +24.1%。**
  • ⊘ **Brent 101.25 美元(9月10日,+0.04%),连续第二个交易日站上 100 美元。⊘ 另一口径同日报 105.71 美元,相差 4.4%,列示且不调和、不赋权。**

解读

  • ⚑ 美元四连跌对铝偏多,但其驱动来自日元而非联储,因此不应被读作联储转鸽(interpretation)。
  • ⊘ **PPI 的结构比其标题更重要:最终需求能源 +4.2%、柴油 +24.1% 是尚未传导至核心的能源成本冲击。对铝而言,它抬升电解成本底,同时让 9月16日的决议更难转鸽——两面性,并按两面性赋权(interpretation)。**
  • ⊘ **今晚的 CPI 是本页唯一能在 24 小时内改变宏观栏的事件,而夜盘已经在它之前先行下跌了 1.83%(interpretation)。**

五 · 基本面

⚑⊘ 供应、成本与需求

  • ⚑ **成本端本期转向偏空:氧化铝成本口径一失守(AO2610 收 2,715,低于 2,718),成本口径二首次被超越的位置外移回 AO2705,电解环节的成本支撑叙事降一档。**
  • ⊘ 但夜盘的配对利润代理单场压缩 372.72,说明下跌由冶炼利润吸收,而非由成本下移让出(derivation)。
  • ⚑ **Alunorte 8月减产至产能 50%,其后已在与燃气供应商达成临时码头准入协议后恢复满产——本期确认。**
  • ⊘ 一条未标注日期的市场说明将伦敦近期强势归因于美伊供应不确定性与 **LME 库存接近 36 年低位;另一条同样未标日期的说明则称伦铝因美伊可能达成协议而自四年高位回落。两者是同一事实的两种读法,均未标日期,并列记录、均不赋权。**
  • ⊘ 一条沿用的 Rusal 条目(因氧化铝成本减产 6% / 约 25 万吨/年)未标注日期,与另行取得的 **210 万吨氧化铝封存提议并列列示,不予赋权。**
  • ⚑ 沿用供应台账:几内亚约 1.5 亿吨/年出口上限目标(−25%);力拓 Yarwun 自 10月起 −120 万吨/年;美铝 2026 年指引下调 20–30 万吨;美铝收购南方32 铝资产(2026年6月30日公告,总额约 41 亿美元,含至多 7.5 亿美元或有价值权,预计 2027 上半年交割);Qatalum 约 60% 名义产能;Alba 1–3 线停产(19%);EGA Al Taweelah 至 8月中旬已重启 18% 电解槽,其氧化铝精炼厂 25% 与 50% 的冲突仍未调和。
  • ⊘ 需求端本期无新增可核实数据:8月未锻轧铝及铝材出口 626,000 吨(1–8月 +16.7%)为沿用;中国 8月社融与新增贷款仍未公布。
  • ⊘ **现货成交情绪本期无法给出:SMM 意愿指数连续第十一期登录墙,CNAL 区域数据本期缺失。本简报不以代理指标替代,直接留白。**

六 · 电解铝现货市场

⊘⊘ 9月10日:两个现货来源同时失效,本期只能给出条件性读数

  • ⊘⊘ **这是本台账上现货栏最弱的一期,原因已在第二节说明:SMM 登录墙(第十一期)+ CNAL 五口径未发布(首次中断)。**
  • ⊘ 可得内容仅有:第三方镜像口径 A00 **24,560(零变动)、佛山 24,780(零变动)、ADC12 均价 24,450(零变动);Mysteel 口径巩义 24,430、华南 24,800(9月10日 22:30)。**
  • ⚑⊘ 由此只能给出条件性基差 **+70(24,560 − 24,490),并注明其成立与否完全取决于镜像行是报价还是沿用(interpretation)。**
  • ⊘ 华东、华中、华南的成交活跃度、出货意愿与采购意愿本期无任何可核实数据,不作描述。
  • ⊘ ADC12 口径差扩大至六个:24,450 / 24,750(无日期)/ 23,950 / 24,000 / 23,615 / 23,500–23,800,不予赋权;9月10日无任何 AD 合约数据。

七 · 关键价格驱动因素

⚑ 利多因素

  • ⚑⚑ **伦敦 backwardation 连续第四次走阔至 +25.00 美元(−0.50 → +2.00 → +6.00 → +14.00 → +25.00),且是在现货下跌的同一天发生的——近端实物趋紧的最强证据。**
  • ⚑⚑ **LME 现货 3,343.00 美元仍高于全部六条均线(高出 18.50 至 122.92),为二十一个交易日中第二高结算价。**
  • ⚑⚑ **MA20(24,037.50)本期未被测试:夜盘最低 24,075 距其仅 37.50 点而未下破,整个中期结构仍然完整。**
  • ⚑ **结算价口径下 9月10日是上涨日(24,575,较前结算 +15,高于收盘 85),卖压集中于尾盘而非贯穿全日。**
  • ⚑ **两期合计的持仓减少主要是移仓:主力 −21,950 对全系 −5,032,约 16,900 手留在系统内。**
  • ⚑ **去库仍在进行:SMM 79.6 万吨为本台账最低值,自 7月阶段高点累计 −33.4 万吨(−29.6%);Mysteel 76.8 万吨同向。**
  • ⚑ **两个独立全国库存口径首次被证明只相差一个恒定值(28,000),数据可信度上升。**
  • ⚑ **美元连续第四日下跌、连续第四日守在 99 下方,处四个月低位区。**
  • ⚑ **Brent 连续第二日站上 100 美元,抬升电解环节能源成本底;美国 PPI 中最终需求能源 +4.2%、柴油 +24.1% 指向尚未传导的成本冲击。**
  • ⚑ **氧化铝全系持仓仅减 1,830、AO2701 持仓逆势 +7,733,且收盘曲线本台账首次完全单调。**
  • ⚑ **氧化铝夜盘几乎不动(−0.07% 至 −0.43%,AO2611 持平),成本端没有确认铝的跌幅。**
  • ⚑ **LME 库存仍处 36 年低位区域,沿用的供应中断台账(几内亚上限、Yarwun、美铝指引、Qatalum、Alba、EGA)未见任何缓解。**
  • ⚑ Alunorte 已恢复满产属利空,但其 8月一度减至 50% 说明全球氧化铝供应链的单点脆弱性仍在。
  • ⚑ 中国 8月 PPI 有色金属冶炼及压延加工 **+0.8% 环比、+20.8% 同比(沿用),生产者价格仍在确认金属复价。**
  • ⚑ 本简报数据质量本期显著提升:六条沪铝均线首次全部精确、夜盘十项恒等式全对、SMM–Mysteel 价差解决,使后续判别更可靠。

⊘ 利空因素

  • ⊘⊘ **点名的顶部检验按偏空分支触发:先创新高 24,675,再收于 24,490,低于 24,560 这条线 70 点。**
  • ⊘⊘ **夜盘 −450(−1.83%),为本台账最大单次夜盘跌幅,且六个合约平行下跌 1.75%–1.83%,是重新定价而非技术性解除。**
  • ⊘⊘ **MA5 与 MA10 双双失守(夜盘低于二者 319.00 与 154.50),日盘低点亦已下破 MA5。**
  • ⊘ **十二个铝合约全部收在开盘价之下,且全部收在最高价之下,回吐区间由 80–140 扩大至 125–195。**
  • ⊘ **收盘分位由 100/100 降至 95/98.3。**
  • ⊘ **主力持仓在下跌日减少 10,623(多头减仓特征);夜盘六合约再减 20,436,且未被远月吸收(AL2610 独占 −18,111)。**
  • ⊘ **成交量在破位日收缩 16.5%(全系 −26.6%):上涨日放量、下跌日缩量本应偏多,但创新高当日的重量与次日的失守并列时,更像是高位换手未能成交于高位(interpretation)。**
  • ⊘⊘ **去库速度减半:周四对周四 −19,000,仅为 36,000 的 52.8%,且已低于 8月自身的约 23,935;周一至周四仅 −6,000。**
  • ⊘ **SMM 79.6 万吨落在本简报判别带的空隙中,确认失败——本页最硬的多头证据未能升级为三观测序列。**
  • ⊘⊘ **氧化铝成本口径一失守三点(2,715 对 2,718),余量链条 +39 → +19 → +2 → −3,成本支撑叙事按事先条件降一档。**
  • ⊘ **成本口径二首次被超越的位置由 AO2704 外移回 AO2705,远期成本通胀定价后退。**
  • ⊘ **氧化铝首尾 contango 由 +184 收窄至 +175,十一个合约收在开盘价之下,全系成交 −32.3%。**
  • ⊘⊘ **配对利润代理夜盘单场压缩 372.72——跌幅由冶炼利润全额吸收。**
  • ⊘ **伦敦库存 −175 吨,是四期以来唯一变动,十六区间中十个为零,去库速度仅为此前的 10.2%。**
  • ⊘ **两个现货来源同期失效,本页无法给出任何可核实的成交情绪、升贴水或区域价差,不确定性上升。**
  • ⊘ **镜像口径三个「零变动」行高度可疑,若其为沿用行,则 +70 的基差读数作废。**
  • ⊘ **美国 8月 CPI 今晚落地,在约 56%–60.6% 的加息定价与 9月16日决议之前,事件风险单边压在今晚。**
  • ⊘ **10月联邦基金利率条连续第十四期缺失,联储路径的期限结构仍不可得。**
  • ⊘ **CPI 核心预期本身存在互相矛盾的两个口径(+0.4% 与 +0.22% 环比),预期本身就不可靠。**
  • ⊘ **中国需求端本期零新增可核实数据;8月社融与新增贷款仍未公布。**
  • ⊘ 沿用:高盛 2,350 美元/吨(无日期,不赋权),现较 9月10日结算价低约 **29.7%。**

八 · 一句话观点

昨日写下的七项检验本期落地六项、其中四项偏空:点名的顶部检验按其偏空分支干净触发——AL2610 先以 24,675 创下六十日窗口新高(高于 9月9日 10 点、高于 6月5日高点 155 点),随后收于 24,490,低于「24,560 即成顶」这条线 70 点,自高点回吐由 105 扩大至 185,十二个铝合约全部收在开盘价之下(9月9日的完全反转)且全部收在最高价之下、回吐区间由 80–140 扩大至 125–195;随后夜盘破位,AL2610 跳空 180 点低开、最低 24,075(恰为 8月31日收盘价)、最新 24,125,较结算价 −450(−1.83%),为本台账最大单次夜盘跌幅,且六个合约平行下跌 1.75%–1.83%,MA5 与 MA10 双双失守而 MA20(24,037.50)仅以 87.50 守住、夜盘低点距其 37.50 未破;本期取回 223 行收盘序列使六条沪铝均线首次全部精确,并因 9月4日基准逐分吻合而定位出唯一一处错误——9月7日收盘价应为 24,420 而非 24,345(后者是结算价),更正后 9月9日 MA5 为 24,420.00;库存方面 SMM 周四落地 796,000 吨,恰好掉进本简报自己在「减速」(786–795)与「停滞」(800 以上)之间留下的 5,000 吨空隙,诚实的判定是确认失败、停滞未至,而算术并无歧义:周四对周四 −19,000 仅为 36,000 速度的 52.8%、且已低于 8月自身的约 23,935,周一至周四仅 −6,000,去库没有停止而是减半,但本栏真正的进展是口径而非水平——Mysteel 同日报 768,000,SMM 减 Mysteel 在四个日期上为 28,000 / 26,000 / 28,000 / 28,000,证明其为恒定定义性偏移而非数据冲突,沿用近两周的「未调和」标记予以撤销,且两个独立全国口径对同一周的减速方向一致;氧化铝按事先条件降级,AO2610 收 2,715 低于成本口径一 2,718 三点(余量链条 +39 → +19 → +2 → −3),成本口径二首次被超越的位置由 AO2704 外移回 AO2705,首尾 contango 由 +184 收窄至 +175,十一个合约收在开盘价之下;而本页最大的背离在于同一个夜盘里铝跌 1.75%–1.83% 而氧化铝只跌 0.07%–0.43%(AO2611 持平),配对利润代理单场压缩 372.72,为本台账最大,意味着这一跌幅由冶炼利润全额吸收;伦敦继续分裂,现货结算 3,343.00 美元(−9.00)仍为二十一个交易日次高且高于全部六条均线,但三个月 −20.00 使 backwardation 第四次连续走阔至 +25.00,库存四期以来首次变动却只有 −175 吨——满足「三位数抽库」的字面而远不满足其实质,故「事件而非趋势」的判定维持,该条件与上述库存判别带一并记为设定不当;宏观上美元连续第四日下跌并守在 99 下方(约 98.70,通道仍在日元与 9月18日日本央行),美国 8月 PPI 整体 +5.4% 超预期而核心 +0.2% 低于预期、其中最终需求能源 +4.2%、柴油单月 +24.1% 是尚未传导至核心的成本冲击,Brent 连续第二日站上 100 美元(101.25),而美国 8月 CPI 今晚 20:30(北京)落地,是 9月16日前最后一条通胀数据,FedWatch 仍是 56%/59%/60.6% 的三口径定价;因此整体是:盘面按点名条件失守、夜盘平行下移并抹去两条均线、去库减半、氧化铝成本线失守三点且金属跌而成本不跟随,而对面则是 MA20 未被测试、结算价口径下当日仍属上涨、两期持仓减少主要为移仓、伦敦近端升水第四次走阔、美元第四日下跌,以及本简报自身数据质量的一次显著跃升(六条均线首次全部精确、夜盘十项恒等式全对、SMM–Mysteel 价差解决)——这比昨日明显更弱的一页,而最可能改变它的是今日日盘是否去测试 MA20(interpretation)。

九 · 今日(9月11日)日盘与夜盘关注清单

  • ⚑⚑ 第一,MA20(24,037.50)会不会被测试,以及夜盘低点 24,075 是否被下破。 这是本页最重要的单一水平,因为夜盘最低仅距其 37.50 点且未破,而它同时恰为 8月31日收盘价。判别很简单: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,应改记为趋势转折而非回调。参考位:24,444.00(MA5)/ 24,279.50(MA10)/ 24,125(夜盘最新)/ 24,075(夜盘低点 = 8月31日收盘)/ 24,037.50(MA20)/ 24,009.33(MA30)。
  • ⚑⚑ 第二,现货能否在盘面下跌后给出真实报价,以及基差究竟是 +70 还是别的数字。 本期两个现货来源同时失效,镜像口径的三个「零变动」行高度可疑。判别条件事先写下:若 9月11日取得真实 SMM 或 CNAL 报价且 A00 在 24,400 以上,则「盘面下跌而现货不跟随」成立,基差转为实质升水,偏多;若 A00 落在 24,150–24,300,则现货只是滞后一日跟随,基差回到 0 至 −50 的常态带;若 24,100 以下,则现货领跌,偏空。
  • ⚑⚑ 第三,氧化铝会不会确认铝的跌幅,还是继续拒绝确认。 本期夜盘铝跌 1.83% 而 AO2701 只跌 0.32%、AO2611 持平,配对利润代理单场压缩 372.72。判别带:AO2610 若收于 2,718 之上,则成本口径一的失守属单日噪音,昨日降级可回撤;若连续第二日收于 2,718 之下,则降级坐实,并需同时观察 AO2701 是否跌破 2,753(9月10日低点)。另需观察氧化铝全系持仓在 −1,830 之后是否继续下降。
  • ⚑ 第四,主力持仓是继续减少还是回补,以及全系口径是否仍支持「移仓」的定性。 本期主力两期合计 −21,950 而全系仅 −5,032,夜盘六合约再减 20,436。判别带(沿用并收紧):9月11日全系持仓变动在 ±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;若主力在下跌中持仓回升,则为新空入场而非多头减仓,性质不同,需分别记录。
  • ⚑ 第五,LME 的 25.00 美元 backwardation 与几乎不动的库存,哪一个先让步。 本页最清晰的矛盾已连续第五期存在。观察:现货能否守住 3,324.50(MA5),该线本期仅余 18.50 点;backwardation 是走阔第五步、收窄还是翻回 contango;以及库存是回到连续零(则本期 −175 的字面触发彻底作废)还是出现第二笔三位数抽库(则事件定性需要真正重新审视——本次把门槛写明:单笔 −500 吨以上才构成实质,以免重复本期的设定缺陷)。
  • ⚑ 第六,美国 8月 CPI(今晚 20:30 北京)及其对 9月16日定价的影响。 预期本身口径冲突(核心 +0.4% 与 +0.22% 环比并存),因此先看整体同比是否落在 +3.3% 至 +3.4% 区间内:落在区间内且核心不超预期,则 FedWatch 的 56%–60.6% 区间大致维持,美元的日元通道继续主导;整体超过 +3.5% 则加息定价上修、美元反弹,对铝偏空;低于 +3.2% 则相反。同时观察 Brent 能否守住 100 美元。
  • ⚑ 第七,SMM 三地日度库存能否在连续第五期缺失后补发,以及上期所周五周度仓单。 本期全国数据已到位而区域分解仍缺,因此无法判断 79.6 万吨的减速发生在哪一地。若补发且华南占比上升,则与此前南海升水连续四日走阔互相印证,属区域物流解释;若华东累库,则减速的原因指向东部到货集中,性质不同。周度仓单若公布,观察其能否延续连续第十一周下降。

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、路透、CNAL 世铝网、AL Circle、Mysteel、SMM(上海有色网)、美国劳工统计局、中国国家统计局、中国海关总署(经转述)、Wood Mackenzie、ING、公开报道的高盛观点以及上市公司公开披露与美铝公司公告。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。本期作出两项自我更正并予以明确记录:其一,9月7日收盘价应为 24,420 而非此前四期沿用的 24,345(后者为当日结算价),据此重述受影响的均线;其二,本简报上一期公布的库存判别带未能完全覆盖数轴,导致 796,000 吨落入空隙,该缺陷已记录并自下一期起修正,同时对 LME 库存的「三位数抽库」反转条件补写了 −500 吨的实质门槛。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 and the nature of this edition

  • Written the morning of Friday, September 11, BEFORE the SHFE day session opens.
  • ⚑ NEW THIS EDITION: the complete SHFE September 10 day session for all twelve aluminium and all twelve alumina contracts; ⚑⚑ the COMPLETE September 10 NIGHT SESSION for six aluminium and six alumina contracts, whose ten prior-settlement fields reconcile EXACTLY against the September 10 day settlements; ⚑⚑ a 223-row retrieved SHFE close series, so ALL SIX SHFE averages are exact for the FIRST time on this ledger and the carried September 4 basis is verified to the cent; ⚑⚑ SMM’s Thursday September 10 national ingot print — the number this brief published a band for in advance; ⚑ the Mysteel national caliber for September 3, 7 and 10, which RESOLVES the long-running SMM–Mysteel spread; the Westmetall September 10 row and sixty retrieved settlement rows, so ALL SIX LME averages are exact for a THIRD consecutive edition; US August PPI; and the US August CPI consensus.
  • ⊘ NOT RETRIEVABLE: SMM’s own September 10 spot set (login-walled for an ELEVENTH consecutive edition — a third-party mirror caliber is listed in parallel and flagged); the CNAL five-caliber regional set for September 10, the FIRST break in that series on this ledger; SMM’s three-region daily inventory series, a FIFTH consecutive edition; the SHFE Friday weekly warehouse and warrant print; any AD contract data; SMM’s regional transaction premia and sentiment indices and Chalco’s five listings; LME volume and open interest; and the October fed funds strip, absent for a FOURTEENTH consecutive edition.

⚑⚑⊘ The top test resolved — on the exact bearish branch this brief named

  • ⚑⚑ AL2610, September 10: open **24,590, high 24,675, low 24,420, close 24,490, settlement 24,575, −70 (−0.28%) against the 24,560 prior settlement, volume 144,380 (−28,468, −16.5%), open interest 236,445 (−10,623, −4.30%). Close-to-close −70.**
  • ⚑⚑ The high of **24,675 is a NEW sixty-session high — 10 above September 9’s 24,665, 125 above the September 8 high and 155 above the June 5 high.**
  • ⊘⊘ It then closed **24,490, seventy points BELOW the 24,560 line yesterday’s test named. The pre-written text was: ‘a close above 24,665 makes the September 9 giveback consolidation; a close below 24,560 makes 24,665 the short-term top.’ The bearish branch fired on its own terms. There is no ‘both halves’ to record this time.**
  • ⊘ The giveback from the high **widened from 105 to 185.**
  • ⊘ **Twelve of twelve aluminium contracts closed BELOW their opens — a complete inversion of September 9’s twelve-for-twelve above — and twelve of twelve also closed below their highs, by 165 / 185 / 195 / 190 / 180 / 185 / 170 / 180 / 190 / 125 / 140 / 150, a band of 125–195 against 80–140 (derivation).**
  • ⚑⊘ **One counterweight recorded with equal weight: the settlement of 24,575 sits 85 ABOVE the close and +15 against the prior settlement. A settlement that far above the close means the volume-weighted session traded higher than it finished and the selling was concentrated at the end. On a settlement caliber September 10 was a small UP day; on a close caliber it was a down day (derivation).**

⊘⊘ Then the night session broke — the largest night move on this ledger

  • ⊘⊘ AL2610 night: open **24,310 (a 180-point gap below the day close), high 24,315, low 24,075, last 24,125 — −450 (−1.83%) against the 24,575 settlement and −365 against the close. Volume 122,943, open interest 218,334, a further −18,111 on the day session.**
  • ⊘ **It was a parallel shift, not a front-end event: six of six retrieved contracts fell 1.75% to 1.83% — AL2609 24,120 (−435), AL2610 24,125 (−450), AL2611 24,135 (−430), AL2612 24,105 (−445), AL2701 24,090 (−440), AL2702 24,100 (−440). A uniform move down the whole curve is repricing, not a squeeze release (interpretation).**
  • ⚑ **The night board is verified, not inferred: all ten prior-settlement fields match the September 10 day settlements exactly — AL 24,555 / 24,575 / 24,565 / 24,550 / 24,530 / 24,540 ✓ and AO 2,721 / 2,737 / 2,748 / 2,783 ✓ (derivation).**
  • ⊘ **And the night open interest is NOT a roll: across the six contracts, day 591,852 → night 571,416, −20,436, of which AL2610 alone is −18,111 while AL2611 fell 1,828 and AL2612 238, with AL2701 +139 and AL2702 +272. Position was reduced on a 1.83% decline and the back months did not absorb it (derivation).**

⚑⚑ All six SHFE averages are exact for the first time — and one carried number is corrected

  • ⚑⚑ A **223-row close series was retrieved: MA5 24,444.00 / MA10 24,279.50 / MA20 24,037.50 / MA30 24,009.33 / MA40 23,821.50 / MA60 23,599.08. The September 10 close clears all six by 46.00 / 210.50 / 452.50 / 480.67 / 668.50 / 890.92. MA20 through MA60 have been carried on a September 4 basis for six editions and are now live.**
  • ⚑ **The same series verifies the carried basis and thereby isolates the one error: recomputed at September 4 it returns 23,981.75 / 23,859.50 / 23,693.38 / 23,576.67 — exactly the four values this brief has carried. So the single defect is that the September 7 CLOSE is 24,420, not 24,345 — 24,345 was the SETTLEMENT. Restated: MA5 at September 9 was 24,420.00 (not 24,405.00); MA10 was 24,212.50 (not 24,204.50) (derivation).**
  • ⊘ **The day LOW of 24,420 pierced MA5 by 24.00; the night last of 24,125 sits 319.00 below MA5 and 154.50 below MA10, but 87.50 ABOVE MA20 (24,037.50) — and the night low of 24,075 came within 37.50 of MA20 and is EXACTLY the August 31 close (derivation).**
  • ⊘ **The close fell out of the 100th percentile for the first time in three sessions: 24,490 sits at the 95th percentile of the twenty-session distribution and the 98.3rd of the sixty-session, against 100th and 100th. Window high 24,560 (Sept 9); window low 22,445 (Jul 1) (derivation).**

⊘⚑ Open interest: a second consecutive fall, and this time on a falling price

  • ⊘ Main OI 247,068 → **236,445 (−10,623) on a DOWN day — the long-liquidation signature, the exact mirror of September 9’s short-covering read on a rising one.**
  • ⚑ Over two sessions: main **258,395 → 236,445, −21,950 (−8.5%), against a complex that went 628,099 → 623,067, −5,032 only. So roughly 16,900 lots rolled and roughly 5,000 left. The roll ruling survives a second session (derivation).**
  • ⊘ Complex volume **266,651 (−96,464, −26.6%); main volume 144,380 (−16.5%). The heaviest turnover was on the day the high was made; the day it failed traded materially lighter.**

⚑⊘ London: cash gave back, but the backwardation widened a fourth step

  • ⚑⊘ Cash settled **$3,343.00, −$9.00 (−0.27%) — still the second-highest settlement in twenty-one sessions.**
  • ⚑⚑ The 3-month settled **$3,318.00, −$20.00, so the backwardation widens a FOURTH consecutive step: −$0.50 → +$2.00 → +$6.00 → +$14.00 → +$25.00. Cash fell and prompt tightness increased on the same row (derivation).**
  • ⚑ **All six LME averages exact for a third consecutive edition: MA5 $3,324.50, MA10 $3,287.65, MA20 $3,256.30, MA30 $3,262.05, MA40 $3,241.44, MA60 $3,220.08. Cash clears all six by 18.50 / 55.35 / 86.70 / 80.95 / 101.56 / 122.92.**
  • ⊘ **The non-monotone stack WIDENED: MA30 is now $5.75 above MA20, against $2.99 on September 9 (derivation).**
  • ⚑ **The ratio fell a third session, by the smallest of the three margins: 24,490 ÷ $3,343.00 = 7.3257, against 7.3270 and 7.3564. SHFE −0.28% close-to-close against LME cash −0.27% (derivation).**
  • ⊘ Parallel calibers, listed and NOT folded into any average: a September 11 live 3M at **$3,287.00 (−$6.50, −0.20%); a second September 11 caliber at $3,296.00 (−0.65%), range $3,278–$3,321 on a $3,316 open. The two are not reconciled.**

⊘⊘ Spot: the weakest page on this ledger — two series broke in the same edition

  • ⊘⊘ **SMM was login-walled for an ELEVENTH consecutive edition, AND the CNAL five-caliber regional set did not publish for September 10 — the FIRST break in that series on this ledger. Both spot sources this brief relies on failed at once, and that is stated plainly rather than papered over.**
  • ⊘ **Third-party mirror caliber, September 10, listed and NOT weighted: SMM A00 24,560, change ZERO; SMM A00 Foshan 24,780, change ZERO; ADC12 24,300–24,600, average 24,450, change ZERO. Three zero-change rows on one source on a day the board fell 70 is the signature of a CARRIED row, not a quote, and it is flagged as such.**
  • ⚑⊘ **If taken at face value the basis is 24,560 − 24,490 = +70, and the chain would read +70 (Sep 4) → +55 (Sep 7) → −20 (Sep 8) → −20 / 0 (Sep 9) → +70 (Sep 10) — ABOVE yesterday’s ‘0 to −20 is caught up’ band, i.e. the board fell and spot did not follow. Recorded as CONDITIONAL and given no weight until a real quotation confirms it (interpretation).**
  • ⊘ **Mysteel regional caliber at September 10, 22:30, listed in parallel: Gongyi 24,430; South China 24,800. Against the mirror’s Shanghai 24,560 that implies a South-China-minus-Shanghai spread of 240 versus CNAL’s 260 at September 9 — but the two come from different calibers and NO conclusion about spread narrowing is drawn.**
  • ⊘ Carried September 9 CNAL set: Shanghai 24,540; Changjiang 24,550; Zhongyuan 24,550; Nanhai/Foshan 25,070; Guangdong Nanchu 24,800; spreads 530 / 260 / +10.
  • ⊘ **SMM’s regional premia and sentiment indices are unavailable for an ELEVENTH edition, so this brief still cannot publish a sentiment index — and says so rather than substituting a proxy.**

⊘ Alumina: cost caliber one failed by three, and AO2701 enters as main contract

  • ⊘⊘ **AO2610, September 10: open 2,711, high 2,733, low 2,711, close 2,715, settlement 2,721, −15 (−0.55%) against the 2,736 prior settlement, volume 39,698, open interest 58,412 (−9,934).**
  • ⊘⊘ **COST CALIBER ONE FAILED: the close is three BELOW 2,718. The clearance chain is +39 (Sep 4) → +19 (Sep 8) → +2 (Sep 9) → −3 (Sep 10). The settlement of 2,721 still holds it by 3; AO2609 closed 2,688, thirty below. The ‘alumina cost supports the aluminium floor’ leg is DOWNGRADED on its own pre-written condition.**
  • ⚑⊘ **AO2701, first edition as main contract: open 2,778, high 2,787, low 2,753, close 2,758, settlement 2,770, −21 (−0.75%) against 2,791, volume 162,739 (−21.1%), open interest 195,391 (+7,733) — the only main contract on this page whose open interest ROSE.**
  • ⚑ **The alumina close curve went FULLY MONOTONE for the first time on this ledger: 2,688 / 2,715 / 2,730 / 2,742 / 2,758 / 2,771 / 2,787 / 2,815 / 2,833 / 2,850 / 2,857 / 2,863 — ZERO inversions, against one on September 9 and two on September 8.**
  • ⊘ But front-to-tail contango narrowed from +184 (6.83% of the front) to **+175 (6.51%); eleven of twelve contracts closed below their opens (only AO2610 above); complex OI 441,597 (−1,830); complex volume 243,792 (−116,187, −32.3%).**
  • ⊘ **The first close above cost caliber two (2,826.21) moved back OUT from AO2704 to AO2705 (2,833) — forward cost inflation priced eight months out again rather than seven. The September 9 improvement lasted exactly one session (derivation).**

⚑⚑ The largest divergence on the page: aluminium −1.8%, alumina −0.1% in the same night

  • ⚑⚑ Alumina night, six contracts: AO2609 2,688 (−8); AO2610 **2,719 (−2, −0.07%); AO2611 2,737 (0, FLAT); AO2612 2,744 (−4); AO2701 2,761 (−9, −0.32%); AO2702 2,771 (−12, −0.43%). AO2701 night open interest 203,668 (+8,277) on 90,883 lots.**
  • ⊘⊘ **Matched-pair night margin proxy (AL2610/AO2610, 1.93 t/t, assumption stated first): day 24,490 − 1.93 × 2,715 = 19,250.05; night 24,125 − 1.93 × 2,719 = 18,877.33. −372.72 in a single session, the largest on this ledger (derivation).**
  • ⚑ **The day-session curve compression actually NARROWED: front 19,307.16 (−61.49), tail 18,904.41 (−44.12), compression 402.75 against 420.12 — narrower by 17.37 (derivation).**
  • ⊘ **A metal that sells off while its cost input does not is a metal whose producers are absorbing the entire move (interpretation).**

3 · Inventory

⚑⚑ SMM’s Thursday print arrived at 796,000 mt — and landed in this brief’s own band gap

  • ⚑⚑ **September 10: 796,000 mt, −6,000 on Monday September 7’s 802,000 and −19,000 on Thursday September 3’s 815,000. Full chain: Aug 24 860,000 → Aug 27 852,000 → Aug 31 837,000 → Sept 3 815,000 → Sept 7 802,000 → Sept 10 796,000.**
  • ⊘⊘ **BAND RESOLUTION (the bands were published one edition BEFORE the number): 775,000–785,000 = confirm; 786,000–795,000 = deceleration; above 800,000 = stall. 796,000 lands in the 5,000-wide GAP between the second and third. RULING: confirmation FAILED, stall NOT reached, and the print sits 1,000 above the deceleration band. A set of bands that does not partition the line is a method defect, it is recorded as one, and it is corrected from the next edition.**
  • ⊘ **PACE (derivation): Thursday-to-Thursday −19,000 = 52.8% of the 36,000 two-observation pace and 79.4% of August’s own ~23,935; Monday-to-Thursday only −6,000 against the prior Thursday-to-Monday −13,000. The destock did not stop. It halved, and it is now running below August.**
  • ⚑ Cumulative from the early-July 1.13 million mt phase high: **−334,000 mt, −29.6%.**

⚑⚑ The real progress in this column is a caliber, not a level

  • ⚑⚑ **Mysteel prints 768,000 mt the same day, with its own dated chain: Aug 31 809,000 → Sept 3 789,000 (−20,000) → Sept 7 774,000 (−15,000) → Sept 10 768,000 (−6,000).**
  • ⚑⚑ **SMM minus Mysteel across four dates: 28,000 / 26,000 / 28,000 / 28,000 — a CONSTANT DEFINITIONAL OFFSET, not a data conflict. The ‘unreconciled spread’ flag this brief has carried since August 31 is RETIRED.**
  • ⚑ **Two independent national calibers agree on the direction: Mysteel Thursday-to-Thursday −21,000 against SMM’s −19,000 (derivation).**

⊘ London: the row moved for the first time in four — by 175 tonnes

  • ⊘ **244,350 mt at September 10, −175. The pre-written reversal condition was ‘a new three-figure draw re-opens the event ruling.’ −175 meets the LETTER and not the SUBSTANCE — 0.07% of stock, one eighth of September 4’s −1,450. The ‘event, not trend’ ruling STANDS, and the condition is recorded as poorly specified, the same class of defect as the inventory band above.**
  • ⊘ **SIXTEEN-INTERVAL FLOW (derivation): August 18 246,925 → September 10 244,350 = −2,575 mt (−1.04%); ten of sixteen intervals exactly zero; September 4’s −1,450 is still 56.3% of the whole.**
  • ⊘ **SIXTY-SESSION FLOW (derivation): June 17 316,525 → 244,350 = −72,175 mt (−22.80%); ~1,582 mt per interval to August 18 and ~161 mt per interval since — 10.2% of the prior pace.**

Interpretation

  • ⊘ **The hardest bullish datum on this page is downgraded one notch but not voided: the destock continues, at half the pace and now below August’s, and two independent calibers confirm it simultaneously (interpretation).**
  • ⚑⊘ **London remains the clearest contradiction on the page: the warrant row moved 175 tonnes in four sessions while the cash premium widened a fourth time to $25.00. When warrants barely move and prompt premia keep widening, the warrant series is the more likely candidate for measuring the wrong thing (interpretation).**
  • ⚑ **A methodological gain and a methodological defect are both recorded this edition: the SMM–Mysteel caliber problem is solved; the band that failed to partition the line is not. Both go on the page (interpretation).**

4 · Macro Factors

⚑ A fourth consecutive dollar decline — through a channel that is not the Fed

  • ⚑ **DXY ~98.70 in Asian hours on September 10, a fourth consecutive decline and a fourth session below 99, in a four-month low area; a parallel caliber has it steady near 98.8. Month −1.03% (carried).**
  • ⊘ **The named driver remains the yen and the Bank of Japan’s September 18 meeting, not the Fed. Bullish for aluminium, for a reason that is not about US rates (derivation).**

⊘ US August CPI lands tonight, into a Fed priced across three calibers

  • ⊘ **CPI at 08:30 ET September 11 = 20:30 Beijing — the last inflation print before September 16.**
  • ⊘ **Consensus, with an unreconciled caliber conflict: headline +0.4% m/m and +3.4% y/y (a second caliber says it EASES to +3.3%); core +0.4% m/m / +2.4% y/y on one caliber and +0.22% m/m on another. The two core calibers contradict each other and are listed in parallel.**
  • ⊘ **CME FedWatch, three calibers: ~56%, 59%, 60.6% (September 8) priced for a 25 bp HIKE. The blackout is running. The October fed funds strip is absent for a FOURTEENTH edition.**
  • ⚑⊘ **US August PPI (released September 10): +0.4% m/m, +5.4% y/y against a +5.3% consensus and a +4.7% prior; core +0.2% against +0.3%. Final demand goods +1.1% against services +0.1%; final demand energy +4.2%; diesel +24.1% m/m.**
  • ⊘ **Brent $101.25 at September 10, +0.04% — a second consecutive session above $100. ⊘ A second caliber carries $105.71 for the same date, a 4.4% gap, listed, not reconciled, not weighted.**

Interpretation

  • ⚑ A fourth straight dollar decline is **bullish for aluminium, but its driver is the yen rather than the Fed and should not be read as a dovish Fed repricing (interpretation).**
  • ⊘ **The structure of the PPI matters more than its headline: final demand energy +4.2% and diesel +24.1% on the month is an energy cost shock that has not yet reached core. For aluminium it raises the smelting cost floor and makes September 16 harder to turn dovish — two-sided, and weighted as such (interpretation).**
  • ⊘ **Tonight’s CPI is the only event that can move the macro column inside 24 hours — and the night session already fell 1.83% ahead of it (interpretation).**

5 · Fundamentals

⚑⊘ Supply, cost and demand

  • ⚑ **The cost column turned bearish this edition: alumina cost caliber one failed (AO2610 2,715 against 2,718) and cost caliber two’s first close above moved back out to AO2705, so the cost-support narrative under the metal is downgraded one notch.**
  • ⊘ But the matched-pair night margin proxy **compressed 372.72 in one session, which says the decline was absorbed by smelter margin rather than conceded by falling costs (derivation).**
  • ⚑ **Alunorte, cut to 50% of capacity in August, is confirmed as since restored to full production following a temporary terminal-access agreement with its gas supplier.**
  • ⊘ An **undated market note attributes London’s recent strength to US–Iran supply uncertainty and LME stocks near a 36-year low; a separate undated note has LME aluminium coming off four-year highs on the possibility of a US–Iran deal. The two are the same fact read in both directions, neither is dated, and both are listed without weight.**
  • ⊘ A carried **Rusal item citing a 6% / ~250,000 mt annual primary cut on alumina cost is UNDATED and is listed alongside the separately sourced 2.1 million mt alumina mothball proposal, without weight.**
  • ⚑ Carried supply ledger: Guinea’s ~150 million mt/yr export cap target (−25%); Rio Tinto Yarwun −1.2 million mt/yr from October; Alcoa’s 2026 guidance cut of 200,000–300,000 mt; **Alcoa’s acquisition of South32’s aluminium assets, announced JUNE 30, 2026 at ~USD 4.1 billion total with up to USD 750 million of CVRs and closing expected H1 2027; Qatalum ~60% of nameplate; Alba lines 1–3 down (19%); EGA’s Al Taweelah 18% of reduction cells restarted by mid-August, with the 25% vs 50% alumina refinery conflict still unreconciled.**
  • ⊘ **The demand side contributed NO new verifiable data this edition: August unwrought aluminium and product exports of 626,000 mt (Jan–Aug +16.7%) are carried, and China’s August social financing and new loans are still unpublished.**
  • ⊘ **Spot transaction sentiment cannot be given this edition: SMM’s sentiment indices are login-walled for an eleventh edition and the CNAL regional set did not publish. This brief leaves the field blank rather than substituting a proxy.**

6 · Primary Aluminum Spot Market

⊘⊘ September 10: both spot sources failed at once, so only a conditional reading is possible

  • ⊘⊘ **This is the weakest spot page on this ledger, for the reason given in section 2: SMM’s eleventh login wall plus the first break in the CNAL five-caliber series.**
  • ⊘ What is available: the mirror caliber A00 **24,560 (zero change), Foshan 24,780 (zero change), ADC12 average 24,450 (zero change); and the Mysteel caliber Gongyi 24,430, South China 24,800 at 22:30 on September 10.**
  • ⚑⊘ From that, **only a conditional basis of +70 can be stated (24,560 − 24,490), and it is noted that its validity depends entirely on whether the mirror row is a quotation or a carry (interpretation).**
  • ⊘ **East China, Central China and South China transaction activity, shipment willingness and purchasing sentiment: NO verifiable data this edition, and none is described.**
  • ⊘ The **ADC12 caliber spread widens to SIX — 24,450 / 24,750 (undated) / 23,950 / 24,000 / 23,615 / 23,500–23,800 — no weight given; and there is no AD contract data of any kind for September 10.**

7 · Key Price Drivers

⚑ Bullish factors

  • ⚑⚑ **London’s backwardation widened a FOURTH consecutive step to +$25.00 (−0.50 → +2.00 → +6.00 → +14.00 → +25.00) — on the same day cash fell. The strongest evidence of prompt physical tightness on the page.**
  • ⚑⚑ **LME cash at $3,343.00 still clears all six averages (by 18.50 to 122.92) and is the second-highest settlement in twenty-one sessions.**
  • ⚑⚑ **MA20 (24,037.50) was NOT tested: the night low of 24,075 came within 37.50 and did not break it. The whole medium-term structure is still intact.**
  • ⚑ **On a settlement caliber September 10 was an UP day (24,575, +15 against the prior settlement, 85 above the close) — the selling was concentrated at the close, not distributed through the session.**
  • ⚑ **The two-session open interest reduction is mostly a ROLL: main −21,950 against a complex −5,032, so roughly 16,900 lots stayed inside the system.**
  • ⚑ **The destock continues: SMM’s 796,000 mt is the lowest on this ledger, −334,000 mt (−29.6%) from the July phase high; Mysteel’s 768,000 agrees.**
  • ⚑ **Two independent national inventory calibers are now proven to differ by a CONSTANT (28,000), which raises the confidence attached to both.**
  • ⚑ **The dollar fell a fourth consecutive session and held below 99 a fourth, in a four-month low area.**
  • ⚑ **Brent held above $100 a second session ($101.25), lifting the energy cost floor under smelting; US PPI’s final demand energy +4.2% and diesel +24.1% point to a cost impulse not yet passed through.**
  • ⚑ **Alumina complex open interest fell only 1,830 and AO2701’s ROSE 7,733, and the close curve went fully monotone for the first time on this ledger.**
  • ⚑ **Alumina barely moved in the night (−0.07% to −0.43%, AO2611 flat): the cost side did not confirm aluminium’s decline.**
  • ⚑ **LME stocks remain in a 36-year-low area, and the carried disruption ledger (Guinea cap, Yarwun, Alcoa guidance, Qatalum, Alba, EGA) shows no relief.**
  • ⚑ **Alunorte’s restoration is bearish on its own, but its August cut to 50% is a reminder that single-point fragility in the global alumina chain is unresolved.**
  • ⚑ China’s August PPI non-ferrous smelting and rolling at **+0.8% m/m and +20.8% y/y (carried) — producer prices still confirm the metal complex repricing.**
  • ⚑ **This brief’s own data quality took a large step forward: all six SHFE averages exact, ten-for-ten night reconciliation, and the SMM–Mysteel spread resolved — which makes every test written below more reliable.**

⊘ Bearish factors

  • ⊘⊘ **The named top test fired on its bearish branch: a new high at 24,675, then a close at 24,490, seventy below the 24,560 line.**
  • ⊘⊘ **The night session fell 450 (−1.83%), the largest on this ledger, with six contracts down 1.75%–1.83% in parallel — repricing, not a technical unwind.**
  • ⊘⊘ **MA5 and MA10 are both LOST (the night sits 319.00 and 154.50 below them), and the day low already pierced MA5.**
  • ⊘ **Twelve of twelve contracts closed below their opens AND below their highs, with the giveback band widening from 80–140 to 125–195.**
  • ⊘ **The close fell from the 100th/100th percentile to the 95th/98.3rd.**
  • ⊘ **Main open interest fell 10,623 on a DOWN day (long liquidation), and the night shed a further 20,436 across six contracts with the back months NOT absorbing it (AL2610 alone −18,111).**
  • ⊘ **Volume contracted 16.5% on the break day (complex −26.6%). Light volume on a decline is normally bullish, but set against the heaviest turnover of the run occurring on the day the high was made and failed, it reads as high-level churn that could not clear at the high (interpretation).**
  • ⊘⊘ **The destock HALVED: Thursday-to-Thursday −19,000, only 52.8% of the 36,000 pace and below August’s own ~23,935; Monday-to-Thursday only −6,000.**
  • ⊘ **796,000 landed in this brief’s band gap and the confirmation FAILED — the hardest bullish datum on the page did not upgrade to a three-observation series.**
  • ⊘⊘ **Alumina cost caliber one failed by three (2,715 against 2,718), clearance chain +39 → +19 → +2 → −3; the cost-support leg is downgraded on its own pre-written condition.**
  • ⊘ **Cost caliber two’s first close above moved back OUT from AO2704 to AO2705 — forward cost inflation repriced further away.**
  • ⊘ **Alumina contango narrowed from +184 to +175, eleven of twelve contracts closed below their opens, and complex volume fell 32.3%.**
  • ⊘⊘ **The matched-pair margin proxy compressed 372.72 in one night — the decline was absorbed entirely by smelter margin.**
  • ⊘ **LME stocks moved just −175 in four sessions, ten of sixteen intervals are exactly zero, and the draw is running at 10.2% of its prior pace.**
  • ⊘ **Both spot sources failed in the same edition, so this page can publish no verifiable sentiment, premium or regional spread — uncertainty is higher, not lower.**
  • ⊘ **The mirror’s three zero-change rows are highly suspect; if they are a carry, the +70 basis reading is void.**
  • ⊘ **US August CPI lands tonight into 56%–60.6% hike pricing six days before the decision — the event risk is concentrated on one side of one session.**
  • ⊘ **The October fed funds strip is absent for a FOURTEENTH edition, so the term structure of Fed pricing remains unavailable.**
  • ⊘ **The CPI core consensus itself is two contradictory calibers (+0.4% and +0.22% m/m) — the expectation is unreliable before the print is.**
  • ⊘ **China’s demand side contributed zero new verifiable data, and August social financing and new loans are still unpublished.**
  • ⊘ Carried: Goldman Sachs’ reported USD 2,350/mt Q4 2026 projection (undated, not weighted) — now ~29.7%** below the September 10 settlement.**

8 · One-Sentence Market View

Six of yesterday’s seven tests resolved and four resolved bearish: the named top test fired cleanly on its bearish branch — AL2610 first printed a new sixty-session high at 24,675, ten above September 9’s and 155 above the June 5 high, then closed 24,490, seventy below the 24,560 line that yesterday’s text said would make 24,665 the top, with the giveback from the high widening from 105 to 185, twelve of twelve contracts closing below their opens in a complete inversion of September 9 and twelve of twelve also closing below their highs on a band that widened from 80–140 to 125–195; the night session then broke, AL2610 gapping 180 points lower at the open, trading to 24,075 — exactly the August 31 close — and finishing 24,125, down 450 or 1.83% against settlement, the largest night move on this ledger, with six of six contracts down 1.75% to 1.83% in a parallel shift, MA5 and MA10 both lost and MA20 at 24,037.50 held by only 87.50 with the night low 37.50 away from it; a 223-row retrieved close series makes all six SHFE averages exact for the first time on this ledger and, because the September 4 basis reproduces to the cent, isolates a single carried error — the September 7 close is 24,420, not 24,345, which was that day’s settlement, so September 9’s MA5 was 24,420.00; on inventory, SMM’s Thursday print landed at 796,000 mt, falling into the 5,000-tonne gap this brief’s own bands left between deceleration at 786,000–795,000 and stall above 800,000, so the honest ruling is that confirmation failed and stall was not reached, while the arithmetic is unambiguous — Thursday-to-Thursday minus 19,000 is 52.8% of the 36,000 pace and below August’s own 23,935, and Monday-to-Thursday is only minus 6,000, meaning the destock did not stop but halved — though the real progress in that column is a caliber rather than a level, because Mysteel printed 768,000 the same day and SMM minus Mysteel across four dates is 28,000, 26,000, 28,000, 28,000, proving a constant definitional offset rather than a data conflict and retiring a flag this brief has carried for two weeks, with both calibers agreeing on the deceleration; alumina downgraded on its own pre-written condition, AO2610 closing 2,715 against the 2,718 cost line on a clearance chain of plus 39, plus 19, plus 2, minus 3, with cost caliber two’s first close above moving back out from AO2704 to AO2705, the contango narrowing from 184 to 175 and eleven of twelve contracts closing below their opens; and the largest divergence on the page is that in the same night aluminium fell 1.75% to 1.83% while alumina fell 0.07% to 0.43% with AO2611 outright flat, compressing the matched-pair margin proxy by 372.72 in one session, the largest on this ledger, which means the decline was absorbed entirely by smelter margin; London split again, cash settling 3,343.00 dollars, down 9.00, still the second-highest in twenty-one sessions and clearing all six averages, while the three-month fell 20.00 so the backwardation widened a fourth consecutive step to plus 25.00, and the stock row moved for the first time in four by just 175 tonnes — meeting the letter of the three-figure-draw reversal condition and nowhere near its substance, so the event ruling stands and that condition, like the inventory band, is recorded as badly drawn; on macro the dollar fell a fourth session and held below 99 at roughly 98.70 through a yen channel rather than a Fed one, US August PPI beat on the headline at plus 5.4% and missed on the core at plus 0.2% with final demand energy at plus 4.2% and diesel at plus 24.1% on the month as a cost shock not yet in core, Brent held above 100 dollars a second session at 101.25, and US August CPI lands tonight at 20:30 Beijing as the last inflation print before September 16 into FedWatch calibers of 56, 59 and 60.6 per cent; so the net is a board that failed on its named condition, a night that shifted the whole curve down and erased two averages, a destock running at half pace, an alumina cost line broken by three and a metal falling while its input did not — against an untested MA20, a settlement caliber that still says the day was up, a two-session position reduction that is mostly a roll, a fourth widening of London’s prompt premium, a fourth dollar decline, and a material step up in this brief’s own data quality — a distinctly weaker page than yesterday’s, and the thing most likely to change it is whether today’s day session goes and tests MA20 (interpretation).

9 · Watchlist for Today’s Day and Night Sessions (September 11)

  • ⚑⚑ First, whether MA20 (24,037.50) gets tested, and whether the night low of 24,075 breaks. This is the single most important level on the page: the night came within 37.50 of MA20 without breaking it, and that low is exactly the August 31 close. The test is simple: hold 24,075 and close above 24,279.50 (MA10) and the night is an overshoot; break 24,037.50 and close below it and the sixty-session up-structure is materially broken and should be recorded as a trend change rather than a pullback. References: 24,444.00 (MA5) / 24,279.50 (MA10) / 24,125 (night last) / 24,075 (night low = Aug 31 close) / 24,037.50 (MA20) / 24,009.33 (MA30).
  • ⚑⚑ Second, whether a REAL spot quotation appears, and whether the basis is genuinely +70. Both spot sources failed this edition and the mirror’s three zero-change rows are highly suspect. The discriminating bands are written in advance: a genuine SMM or CNAL A00 above 24,400 confirms ‘the board fell and spot did not follow’ and turns the basis into a real premium — bullish; 24,150–24,300 means spot simply followed with a one-day lag and the basis returns to a normal 0 to −50; below 24,100 means spot is leading the decline — bearish.
  • ⚑⚑ Third, whether alumina confirms aluminium’s decline or keeps refusing to. Aluminium fell 1.83% overnight while AO2701 fell 0.32% and AO2611 was flat, compressing the matched-pair proxy by 372.72. The band: AO2610 closing above 2,718 makes the cost-caliber-one failure one-session noise and the downgrade can be withdrawn; a second consecutive close below 2,718 confirms it — and watch whether AO2701 breaks 2,753, its September 10 low. Also watch whether alumina complex open interest keeps falling after −1,830.
  • ⚑ Fourth, whether main-contract open interest keeps falling, and whether the complex still supports the ‘roll’ ruling. Two sessions have taken the main contract down 21,950 against a complex down only 5,032, and the night shed a further 20,436 across six contracts. Bands, carried and tightened: a complex change within ±5,000 is still a roll; −15,000 or worse voids the roll reading and it becomes an exit; and if main open interest RISES on a falling price that is new short entry rather than long liquidation — a different animal, recorded separately.
  • ⚑ Fifth, which gives way first — London’s $25.00 backwardation or its motionless stock row. The clearest contradiction on this page is now in its fifth edition. Watch whether cash holds $3,324.50 (MA5), which it clears by only $18.50; whether the backwardation widens a fifth step, narrows, or flips to contango; and whether stocks return to consecutive zeros (which voids this edition’s literal trigger entirely) or produce a second three-figure draw — with the threshold now written explicitly to avoid repeating this edition’s defect: a single draw of −500 mt or larger is required to count as substantive.
  • ⚑ Sixth, US August CPI tonight at 20:30 Beijing and its effect on September 16 pricing. The consensus itself carries contradictory core calibers (+0.4% and +0.22% m/m), so watch the headline year-on-year against the 3.3%–3.4% range first: inside the range with core in line leaves the 56%–60.6% FedWatch band roughly intact and the dollar’s yen channel dominant; above +3.5% lifts hike pricing and the dollar — bearish for aluminium; below +3.2% does the reverse. Watch alongside whether Brent holds $100.
  • ⚑ Seventh, whether SMM’s three-region daily series returns after five missing editions, and the Friday SHFE warrant print. The national figure arrived but the regional decomposition did not, so it is impossible to say where the deceleration to 796,000 occurred. If it publishes with South China’s share rising, that corroborates the four-session widening in the Nanhai premium and supports a regional-logistics reading; if East China is building, the cause points to concentrated eastern arrivals — a different explanation with different implications. If the weekly warrant print appears, watch whether it extends an eleventh consecutive weekly decline.

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, Reuters, CNAL, AL Circle, Mysteel, SMM, the US Bureau of Labor Statistics, China’s National Bureau of Statistics, China Customs (as reported), Wood Mackenzie, ING, publicly reported Goldman Sachs views, and listed-company and Alcoa corporate disclosures. 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 falsifies one. Derivations that rely on an assumed coefficient (such as 1.93 t alumina per tonne of aluminium) state the assumption before the calculation. THIS EDITION MAKES TWO SELF-CORRECTIONS AND RECORDS BOTH EXPLICITLY: first, the September 7 close is 24,420, not the 24,345 carried for four editions (24,345 was that day’s settlement), and the affected averages are restated; second, the inventory bands this brief published one edition in advance failed to partition the number line, so the 796,000 mt print fell into a gap — the defect is recorded, the band method is corrected from the next edition, and the LME ‘three-figure draw’ reversal condition has been rewritten above with an explicit −500 mt substantive threshold. This brief does not constitute investment advice and accepts no responsibility for any decision made on the basis of it.**