diff --git a/_includes/math/ai/OpenAIMath.html b/_includes/math/ai/OpenAIMath.html index 42e0b56..2bf1efa 100644 --- a/_includes/math/ai/OpenAIMath.html +++ b/_includes/math/ai/OpenAIMath.html @@ -89,11 +89,11 @@

{% include math/ai/openai-result.html num="109" %} {% include gh-link.html h diff --git a/_math_notes/ai/openai/integer-mult-bounds.md b/_math_notes/ai/openai/integer-mult-bounds.md index 7ba9216..eb2db0a 100644 --- a/_math_notes/ai/openai/integer-mult-bounds.md +++ b/_math_notes/ai/openai/integer-mult-bounds.md @@ -2,7 +2,7 @@ layout: page title: CrocSwap integer-mult-bounds date: 2026-10-21 -last_updated: 2026-10-29 +last_updated: 2026-11-16 pr_repo: CrocSwap/integer-mult-bounds redirect_from: - /articles/Math/ai/openai/integer-mult-bounds/ @@ -16,7 +16,7 @@ integer-mult-bounds 是改進 openai/math Result #109 的社群專案,建立 #109 給出整數乘法的 deterministic 時間界 $O(n(\log n)^{1-\kappa})$,其中 $\kappa=2^{-182}$。專案保留 #109 的框架,優化其中的有限建構、遞迴成本帳與 packing 等環節,把 $\kappa$ 往上推。多數進展不是全新的整數乘法演算法,而是在同一個框架內改善個別瓶頸。 -本文內容更新至 2026 年 10 月 29 日,其中的狀態與 $\kappa$ 以此為準。文中的日期與時間以 UTC 計。 +本文內容更新至 2026 年 11 月 16 日,其中的狀態與 $\kappa$ 以此為準。文中的日期與時間以 UTC 計。 ## OpenAI {% include math/ai/openai-result.html num="109" %} @@ -48,10 +48,10 @@ $\kappa$ 越大,$(\lg n)^{1-\kappa}$ 越小,所以改進這個結果等同 ## 目前狀態 -目前的 selected 是 DreamingOfClouds 的 {% include pr-link.html num=6030 %},於 2026 年 10 月 28 日升為 selected,$\kappa$ 為 +目前的 selected 是 CubeJerry(Jerry Chen)的 {% include pr-link.html num=7000 %},搭配同一作者的 BIT dependency {% include pr-link.html num=7001 %},於 2026 年 11 月 10 日升為 selected,$\kappa$ 為 $$ -\boxed{\kappa=781451/250000000=0.003125804} +\boxed{\kappa=363005817/100000000000=0.00363005817} $$ @@ -65,26 +65,24 @@ $$ - - + + - - + +
Previous selected{% include pr-link.html num=5599 %}$0.00310045${% include pr-link.html num=6030 %}$0.003125804$
Current selected{% include pr-link.html num=6030 %}$0.003125804${% include pr-link.html num=7000 %}$0.00363005817$
-{% include pr-link.html num=6030 %} 已 merge,是 maintainer-reviewed(維護者的 AI 協助 application review,限於保留的假設)的 conditional P104 twin-rung construction。它在 {% include pr-link.html num=5599 %} 之後累積的 P104 word 上移除 helper 470 在 frame 593 的 rung,完整的 caller 有 13,444 個 productive field 與 66,834 個 paid call,重放指令為 `make frontier6030-verify`。 +{% include pr-link.html num=7000 %} 已 merge,是 maintainer-reviewed(維護者的 AI 協助 application review,限於保留的假設)的 conditional shared-native construction。它在 DreamingOfClouds 的 shared-wire word 上再構造兩個 helper split,完整的 caller 有 23,392 個 productive field 與 117,612 個 paid call,重放指令為 `make frontier7000-verify`。 -與前一個 selected {% include pr-link.html num=5599 %} 相比,exponent saving 約大 0.81775%,這個比較不是實測的 running-time improvement。Selected result 仍保留既有的 conditional 假設,不是完整的 Lean multiplication theorem,也不是 independent human proof review。 - -{% include pr-link.html num=6030 %} 之後已有更高 $\kappa$ 的 merged candidate,例如 CubeJerry 的 {% include pr-link.html num=6145 %}($\kappa=0.003127804$),upstream ledger 標為 partially reviewed,尚未升為 selected。 +與前一個 selected {% include pr-link.html num=6030 %} 相比,exponent saving 約大 16.13%,這個比較不是實測的 running-time improvement。Selected result 仍保留同樣的五個 framework 假設,不是完整的 Lean multiplication theorem,也不是 independent human proof review。 ## 主要演進 -從 {% include pr-link.html num=1094 %} 到 {% include pr-link.html num=6030 %},selected $\kappa$ 經過四次主要的技術轉折。{% include pr-link.html num=4359 %} 讓 helper 完成的 transform 成為 productive output,改變了遞迴的帳;{% include pr-link.html num=4416 %} 到 {% include pr-link.html num=4470 %} 在這個 framework 上大幅減少 paid job;{% include pr-link.html num=4610 %} 到 {% include pr-link.html num=5413 %} 持續壓低 recursive cost;{% include pr-link.html num=5544 %} 起則轉向 P104 word,{% include pr-link.html num=5599 %} 與 {% include pr-link.html num=6030 %} 都在這條路線上。其中 {% include pr-link.html num=4416 %} 通過 maintainer review,但從未成為 selected,不在 selected progression 中。 +從 {% include pr-link.html num=1094 %} 到 {% include pr-link.html num=7000 %},selected $\kappa$ 經過五次主要的技術轉折。{% include pr-link.html num=4359 %} 讓 helper 完成的 transform 成為 productive output,改變了遞迴的帳;{% include pr-link.html num=4416 %} 到 {% include pr-link.html num=4470 %} 在這個 framework 上大幅減少 paid job;{% include pr-link.html num=4610 %} 到 {% include pr-link.html num=5413 %} 持續壓低 recursive cost;{% include pr-link.html num=5544 %} 起轉向 P104 word,{% include pr-link.html num=5599 %} 與 {% include pr-link.html num=6030 %} 都在這條路線上;{% include pr-link.html num=6030 %} 之後則改變 caller 與 root 的結構,經 query tree、mixed-query root 與 chain-nested root,走到一個 native word 服務兩條 root chain 的 shared-native 路線,{% include pr-link.html num=7000 %} 在這條路線上。其中 {% include pr-link.html num=4416 %} 通過 maintainer review,但從未成為 selected,不在 selected progression 中。 @@ -139,6 +137,12 @@ $$ + + + + + + @@ -149,7 +153,7 @@ $$ @@ -247,19 +251,41 @@ $$ - owned buffer:{% include pr-link.html num=5786 %} 的 asynchronous helper hub 讓 helper 在不同時間退出;{% include pr-link.html num=5802 %} 讓三個 pending sum 共用 native conversion,Mason Hale 的 {% include pr-link.html num=5811 %} 再加上第四個 accumulated update - path 與 rung:{% include pr-link.html num=5827 %} 把一條 paid rank-4 / rank-1 path 改成一個 rank-5 query;DreamingOfClouds 的 {% include pr-link.html num=5907 %} 移除 helper 468 在 frame 593 的 rung,並重新導向相關的讀取 -DreamingOfClouds 的 {% include pr-link.html num=6030 %} 是 {% include pr-link.html num=5907 %} 的 twin:helper 470 的路徑 $3\to593\to714\to875\to1001\to131\to127$ 去掉 frame 593,helper 704 的路徑由 $3\to593\to612\to127$ 改為 $3\to593\to714\to127$,在 frame 714 供 helper 470 讀取。Helper 560 不再於 frame 612 讀取 704,原本經由 560 取得這份內容的 helper 268 改在 frame 714 直接讀取 $2\cdot704$。另有四處 itinerary 變動:helper 115 的終點由 frame 105 改為 189,helper 713 的起點由 frame 3 改為 76,helper 736 的起點由 frame 424 改為 749,Y main 127 的第二個 chart 由 602 改為 595。其餘 ordinary event(ADD 與 SCALE)在值不變的前提下重寫,word 由 25,998 個 event 降到 22,803 個。Paid call 由 66,852 降到 66,834,productive field 維持 13,444,rank-mass surplus 維持 760。每個 local call 的 paid 變化為 $-2H_1+H_2-2H_3+H_4+H_6-H_7+H_8$,helper 變化為 $-H_1+H_2-H_3+H_6-H_7+H_8$,helper 變化減去 paid 變化得到 signed moment gain $H_1+H_3-H_4$,對 $0 @@ -290,7 +316,7 @@ $\kappa$ 由 #109 的 $2^{-182}$ 逐步上升到目前 selected 的 $0.003125804 - + @@ -314,7 +340,7 @@ $\kappa$ 由 #109 的 $2^{-182}$ 逐步上升到目前 selected 的 $0.003125804 - + @@ -338,7 +364,7 @@ $\kappa$ 由 #109 的 $2^{-182}$ 逐步上升到目前 selected 的 $0.003125804 - + @@ -422,11 +448,17 @@ $\kappa$ 由 #109 的 $2^{-182}$ 逐步上升到目前 selected 的 $0.003125804 - + + + + + + +
{% include pr-link.html num=6030 %} $0.003125804$Past selected
shared-native 路線{% include pr-link.html num=7000 %}$0.00363005817$ Current selected
—
round two({% include pr-link.html num=62 %})round two
({% include pr-link.html num=62 %})
$5.1017\times10^{-5}$ interval strips、core-aware pair assembly Avi Eisenberg
eumemic
Alejandro Zarzuelo Urdiales
icekylinx
paired-cube({% include pr-link.html num=144 %})paired-cube
({% include pr-link.html num=144 %})
$4.6092\times10^{-4}$ paired cubes、shared completed cores icekylinxeumemic
gen4({% include pr-link.html num=276 %}、{% include pr-link.html num=279 %})gen4
({% include pr-link.html num=276 %}、{% include pr-link.html num=279 %})
$7.2357\times10^{-4}$ paired-cube bit word、concave-descent retiming DreamingOfClouds
Rohan Arun
Ali Romman
hexapode
{% include pr-link.html num=6030 %}(目前 selected){% include pr-link.html num=6030 %} $3.125804\times10^{-3}$ P104 twin-rung removal DreamingOfClouds
Ali Romman
Mason Hale
{% include pr-link.html num=7000 %}
(目前 selected)
$3.63005817\times10^{-3}$shared wires、helper splits on the shared-native callerCubeJerry
DreamingOfClouds
Ali Romman
@@ -453,6 +485,8 @@ $\kappa$ 由 #109 的 $2^{-182}$ 逐步上升到目前 selected 的 $0.003125804 - **Owned helper relation 與 backward caller**:{% include pr-link.html num=5377 %} 起,用 owned relation 去掉 backward call 的 rank-eight center({% include pr-link.html num=5377 %}、{% include pr-link.html num=5383 %}),再讓 backward call 改用縮減後的 supplier:{% include pr-link.html num=5392 %} 以 circuit transposition 轉換兩個 forward call,{% include pr-link.html num=5398 %} 的 direct backward caller 涵蓋全部 18 個 full call,{% include pr-link.html num=5413 %} 補完 coupled-output caller。 - **P104 word 與 helper-subspace split**:{% include pr-link.html num=5544 %} 起改用 P104 word 的 literal caller,再以 contact 重排、helper retiming 與 helper-subspace split 減少 paid call 或增加獨立輸入({% include pr-link.html num=5544 %} 至 {% include pr-link.html num=5599 %})。 - **Retained native copy、owned buffer 與 rung removal**:{% include pr-link.html num=5774 %} 起,在 P104 word 上以 paid native conversion 保留 source 值、以 owned buffer 累積 pending sum,再移除 helper 多餘的 rung({% include pr-link.html num=5907 %}、{% include pr-link.html num=6030 %})。 +- **Query tree、mixed-query root 與 chain-nested root**:{% include pr-link.html num=6348 %} 起改變 caller 與 root 的結構,依序以 binary query tree 取代 path caller、改用 mixed-query root({% include pr-link.html num=6398 %}),再以 chain-nested root 取代 supplier tree({% include pr-link.html num=6833 %},BIT 一側為 {% include pr-link.html num=6834 %})。 +- **Shared native word 與 helper split**:{% include pr-link.html num=6902 %} 起讓一個 native word 服務兩條 root chain,再以 causal helper split、gate scalar 的放寬與 shared wire 增加獨立輸出({% include pr-link.html num=6916 %} 至 {% include pr-link.html num=7000 %})。 ### 瓶頸 @@ -463,7 +497,9 @@ $\kappa$ 由 #109 的 $2^{-182}$ 逐步上升到目前 selected 的 $0.003125804 兩者中 complex 一側較小,是綁住 $\kappa$ 的一側。要讓 $\kappa>0.01$,兩個 effective primitive budget 都需要超過約 $1/99$。這描述的是目前建構的瓶頸,不是整數乘法的下界。 -PR {% include pr-link.html num=4359 %} 之後 complex 仍是綁住的一側:{% include pr-link.html num=4359 %} 中 complex 的 uniform saving 為 $0.002304$,BIT 為 $0.0027$。{% include pr-link.html num=4440 %} 起的建構保留一條 16-pair 的 BIT word,其 ordinary saving 為 $0.0034$,$\kappa$ 的提高來自 complex controller 的改進。{% include pr-link.html num=4535 %} 與 {% include pr-link.html num=4568 %} 沒有再動 controller,提高的幅度只有 $1.5\times10^{-8}$ 與 $9.7\times10^{-10}$,來自 compiler 與 outer 參數的收緊。{% include pr-link.html num=4610 %} 則改動 helper 與 native map,$\kappa$ 再提高到 $0.0030217758$。{% include pr-link.html num=4940 %} 另做了一項不主張 $\kappa$ 的計算:它以 {% include pr-link.html num=4929 %} 的 center-free port code $A$ 共軛 controller,即 $\text{Coded}=A\cdot\text{Standard}\cdot A^{-1}$,使 reverse stage 成為 center-free,只有 controller 邊界需要重新編碼。若這些重新編碼可被接受,BIT supplier 會把 conditional $\kappa$ 限制在約 $0.0034$,目前 selected 的 $0.003125804$ 低於這個值。這不是已建立的上界:重新編碼在實際 frame 上的 physical admission、計價與 compiler 常數都未建立,{% include pr-link.html num=4940 %} 也沒有 maintainer 的 proof review。 +PR {% include pr-link.html num=4359 %} 之後 complex 仍是綁住的一側:{% include pr-link.html num=4359 %} 中 complex 的 uniform saving 為 $0.002304$,BIT 為 $0.0027$。{% include pr-link.html num=4440 %} 起的建構保留一條 16-pair 的 BIT word,其 ordinary saving 為 $0.0034$,$\kappa$ 的提高來自 complex controller 的改進。{% include pr-link.html num=4535 %} 與 {% include pr-link.html num=4568 %} 沒有再動 controller,提高的幅度只有 $1.5\times10^{-8}$ 與 $9.7\times10^{-10}$,來自 compiler 與 outer 參數的收緊。{% include pr-link.html num=4610 %} 則改動 helper 與 native map,$\kappa$ 再提高到 $0.0030217758$。{% include pr-link.html num=4940 %} 另做了一項不主張 $\kappa$ 的計算:它以 {% include pr-link.html num=4929 %} 的 center-free port code $A$ 共軛 controller,即 $\text{Coded}=A\cdot\text{Standard}\cdot A^{-1}$,使 reverse stage 成為 center-free,只有 controller 邊界需要重新編碼。若這些重新編碼可被接受,當時的 BIT supplier 會把 conditional $\kappa$ 限制在約 $0.0034$。這不是已建立的上界:重新編碼在實際 frame 上的 physical admission、計價與 compiler 常數都未建立,{% include pr-link.html num=4940 %} 也沒有 maintainer 的 proof review。 + +{% include pr-link.html num=7000 %} 的 $\kappa=0.00363005817$ 已超過 $0.0034$,因為 BIT 一側也換了 supplier:{% include pr-link.html num=6834 %} 的 chain-nested root caller 與 {% include pr-link.html num=7001 %} 的 paired QUERY-only caller 給出 ordinary saving $0.00385248$。這個值由 selected 的 $0.0034$ 經 $0.0034\to0.0038507\to0.00385248$ 兩步 bootstrap 取得,每一步只使用前一步已完成的 backend;complex 的 outer ledger 保守地採用較弱的 BIT saving $0.00376669$。 ### 可能的改進方向 @@ -480,7 +516,7 @@ PR {% include pr-link.html num=4359 %} 之後 complex 仍是綁住的一側:{% $\kappa$ 衡量的是漸近 exponent saving,不是實測的執行時間。 -所有結果都以 #109 的框架與一般歸約為前提,專案不是 OpenAI 的官方釋出或背書。Selected result 的 scope 保留以下 conditional 假設({% include pr-link.html num=4470 %} 起的五個框架假設,{% include pr-link.html num=4610 %}、{% include pr-link.html num=5413 %}、{% include pr-link.html num=5599 %} 與 {% include pr-link.html num=6030 %} 沿用): +所有結果都以 #109 的框架與一般歸約為前提,專案不是 OpenAI 的官方釋出或背書。Selected result 的 scope 保留以下 conditional 假設({% include pr-link.html num=4470 %} 起的五個框架假設,{% include pr-link.html num=4610 %}、{% include pr-link.html num=5413 %}、{% include pr-link.html num=5599 %}、{% include pr-link.html num=6030 %} 與 {% include pr-link.html num=7000 %} 沿用): - native / all-column movement:native operator、phase 慣例、spectator 與 whole-job adapter 在每個正的 column multiplicity 下都成立 - owned erasure:owned stream 與 suspended center 遵守保留的 allocation、erasure 與 ownership 規則 @@ -488,7 +524,7 @@ $\kappa$ 衡量的是漸近 exponent saving,不是實測的執行時間。 - precision 與 fixed-tape:保留的 fixed-tape movement 與 precision / storage 規定適用於新的 finite template - analytic decomposition、fine-prime avoidance 與 exact recovery -Selected result 的驗證是 maintainer 的 AI 協助 proof review({% include pr-link.html num=6030 %} 為 application review),加上 scoped、serial 的變動建構重放({% include pr-link.html num=4470 %} 為八段,{% include pr-link.html num=5599 %} 在 Linux 上以 1 GiB 取樣記憶體上限通過,{% include pr-link.html num=6030 %} 沿用 {% include pr-link.html num=5960 %} 的獨立 interpreter 檢查 word / invoice / native map),並明確重用前一個 selected 的 foundation,並非完整的 Lean multiplication theorem,也不是獨立的人類 review。Winning uniform template 由演算法定義,沒有實體化,專案不主張實用上的加速。其餘 candidate 則各自附帶作者提交的證據與獨立檢查。 +Selected result 的驗證是 maintainer 的 AI 協助 proof review({% include pr-link.html num=6030 %} 與 {% include pr-link.html num=7000 %} 為 application review),加上 scoped、serial 的變動建構重放({% include pr-link.html num=4470 %} 為八段,{% include pr-link.html num=5599 %} 在 Linux 上以 1 GiB 取樣記憶體上限通過,{% include pr-link.html num=6030 %} 沿用 {% include pr-link.html num=5960 %} 的獨立 interpreter 檢查 word / invoice / native map,{% include pr-link.html num=7000 %} 以獨立的 application、native catalogue 與 exact scalar checker 分別檢查 caller contact、BIT frame、native map 與 scalar arithmetic),並明確重用前一個 selected 的 foundation,並非完整的 Lean multiplication theorem,也不是獨立的人類 review。Winning uniform template 由演算法定義,沒有實體化,專案不主張實用上的加速。其餘 candidate 則各自附帶作者提交的證據與獨立檢查。 ## References @@ -506,6 +542,9 @@ Selected result 的驗證是 maintainer 的 AI 協助 proof review({% include - {% include pr-link.html num=5599 %} [selection / integration receipt](https://github.com/CrocSwap/integer-mult-bounds/blob/main/docs/research/pr-intake-evidence/pr5599-completion/INTEGRATION.json) - {% include pr-link.html num=6030 %} [application review](https://github.com/CrocSwap/integer-mult-bounds/blob/main/docs/research/pr-intake-evidence/pr6030-audit/PROOF-REVIEW.md) - {% include pr-link.html num=6030 %} [selection / integration receipt](https://github.com/CrocSwap/integer-mult-bounds/blob/main/docs/research/pr-intake-evidence/pr6030-audit/INTEGRATION.json) +- {% include pr-link.html num=7000 %} / {% include pr-link.html num=7001 %} [admission review](https://github.com/CrocSwap/integer-mult-bounds/blob/main/docs/research/pr-intake-evidence/pr7000-admission/PROOF-REVIEW.md) +- {% include pr-link.html num=7000 %} / {% include pr-link.html num=7001 %} [application review](https://github.com/CrocSwap/integer-mult-bounds/blob/main/docs/research/pr-intake-evidence/pr7000-application/PROOF-REVIEW.md) +- {% include pr-link.html num=7000 %} [selection / integration receipt](https://github.com/CrocSwap/integer-mult-bounds/blob/main/docs/research/pr-intake-evidence/pr7000-application/INTEGRATION.json) - [current-status.md](https://github.com/CrocSwap/integer-mult-bounds/blob/main/docs/research/current-status.md) - [bottleneck-brief.md](https://github.com/CrocSwap/integer-mult-bounds/blob/main/docs/research/bottleneck-brief.md) - [pr-ledger.md](https://github.com/CrocSwap/integer-mult-bounds/blob/main/docs/research/pr-ledger.md) @@ -514,7 +553,7 @@ Selected result 的驗證是 maintainer 的 AI 協助 proof review({% include ### Pull requests -除 {% include pr-link.html num=1094 %}、{% include pr-link.html num=4359 %}、{% include pr-link.html num=4470 %}、{% include pr-link.html num=4610 %}、{% include pr-link.html num=5413 %}、{% include pr-link.html num=5599 %} 與 {% include pr-link.html num=6030 %} 曾是 selected,以及 {% include pr-link.html num=4416 %} / {% include pr-link.html num=4433 %} 的 joint p7 reviewed checkpoint 外,標題中的 $\kappa$ 都是作者的宣稱,不代表已被 maintainer 認可,各 PR 的處置見 [pr-ledger.md](https://github.com/CrocSwap/integer-mult-bounds/blob/main/docs/research/pr-ledger.md)。 +除 {% include pr-link.html num=1094 %}、{% include pr-link.html num=4359 %}、{% include pr-link.html num=4470 %}、{% include pr-link.html num=4610 %}、{% include pr-link.html num=5413 %}、{% include pr-link.html num=5599 %}、{% include pr-link.html num=6030 %} 與 {% include pr-link.html num=7000 %} 曾是 selected,以及 {% include pr-link.html num=4416 %} / {% include pr-link.html num=4433 %} 的 joint p7 reviewed checkpoint 外,標題中的 $\kappa$ 都是作者的宣稱,不代表已被 maintainer 認可,各 PR 的處置見 [pr-ledger.md](https://github.com/CrocSwap/integer-mult-bounds/blob/main/docs/research/pr-ledger.md)。{% include pr-link.html num=7001 %} 的 BIT dependency(ordinary saving $0.00385248$)已納入 {% include pr-link.html num=7000 %} 的 maintainer application review,但 {% include pr-link.html num=7001 %} 本身不是獨立的 selected checkpoint。 - {% include pr-link.html num=1094 %} Conditional kappa 0.001672660197076918: rare-pattern T16 exchanges - {% include pr-link.html num=2848 %} T248 residue exchange: conditional κ ≈ 0.00171865850830493 @@ -567,7 +606,30 @@ Selected result 的驗證是 maintainer 的 AI 協助 proof review({% include - {% include pr-link.html num=5827 %} Conditional κ = 0.00312476680 (+5.8046936e-7 over #5811): unread path - {% include pr-link.html num=5907 %} Conditional κ .003125282 (+5.22e-7 over #5827): remove helper 468's frame-593 rung with rerouted reads - {% include pr-link.html num=6030 %} Conditional κ .003125804 (+5.22e-7 over #5907): remove helper 470's frame-593 rung, the twin of #5907's removal +- {% include pr-link.html num=6144 %} Conditional κ .003128386 (+2.582e-6 over #6030): owned registers 770, 771 and 772 descend through the frames of helpers 705, 706 and 731, which no longer relay them; helper 732 reads 771 at frame 117 - {% include pr-link.html num=6145 %} Conditional κ = 0.003127804 (+6.67e-7 over #6143): remove the 772/731 relay +- {% include pr-link.html num=6167 %} Conditional κ = 0.003128645 (+2.59e-7 over #6144): retain an operand to remove a root stop +- {% include pr-link.html num=6348 %} Conditional κ = 0.00314933 (+0.661%): replace the path caller with a binary query tree +- {% include pr-link.html num=6350 %} Compose native input lifts with the query tree at conditional κ 0.00314980 +- {% include pr-link.html num=6375 %} Conditional κ = 0.00315120 (+3.3e-7 over #6372): three shared-carrier input lifts +- {% include pr-link.html num=6398 %} Mixed-query root: conditional κ = 0.00328621476572836015146308 (+4.28455% over #6375) +- {% include pr-link.html num=6417 %} Conditional κ = 0.003286849: clean native merge with mixed caller +- {% include pr-link.html num=6545 %} Improve conditional κ to 0.00329394 by retiring an independent Gaussian +- {% include pr-link.html num=6726 %} Conditional κ = 0.00330720 (+3.000e-6, +0.09079% over #6716): two paid idle fragments unlock independent outputs +- {% include pr-link.html num=6833 %} Conditional κ = 0.00337180 (+6.36e-5, +1.92% over #6824): replace the supplier tree by a chain-nested root +- {% include pr-link.html num=6834 %} Chain-nested root caller on the BIT native word: ordinary BIT saving 0.00385248 (dependency bound, no overall κ) +- {% include pr-link.html num=6902 %} Conditional κ = 0.00351406 (+1.652% over #6889): one native word serves two root chains +- {% include pr-link.html num=6911 %} Conditional κ = 0.003519487 (+0.154% over #6902): 8 ports keep the hosted form +- {% include pr-link.html num=6916 %} Conditional κ = 0.00357027 (+1.443% over #6911): 68 causal helper splits +- {% include pr-link.html num=6918 %} Conditional κ = 0.00357906 (+0.246% over #6916): flipped-word helper splits +- {% include pr-link.html num=6931 %} Conditional κ = .00360410 (+0.245% over #6919): ten more helper splits +- {% include pr-link.html num=6932 %} Conditional κ = .00362106 (+0.470% over #6931): allow 1/3 and 1/5 as gate scalars +- {% include pr-link.html num=6934 %} Conditional κ = .00362386 (+0.077% over #6932): rebuilt word, smaller caller +- {% include pr-link.html num=6936 %} Conditional κ = .00362702 (+0.087% over #6934): seven main contacts on shared wires +- {% include pr-link.html num=6937 %} Two-chain family: remaining proofs, objects, backup words and open leads (no new κ) +- {% include pr-link.html num=6980 %} Research: QUERY-only caller +- {% include pr-link.html num=7000 %} Conditional κ = 0.00363005817 (+0.0838% over #6936): shared wires and two helper splits +- {% include pr-link.html num=7001 %} Remove empty-root credits from the paired BIT supplier at 0.00385248 (no new κ) #### Independent checks