math-skill 3.0.0 → 3.0.1

package/knowledge-base/probability/concentration-inequality.en.md

@@ -28,11 +28,11 @@ $$P(|f(X_1,\ldots,X_n) - \mathbb{E}[f]| \geq t) \leq 2\exp\left(-\frac{2t^2}{\su
 - **Gradient compression / communication efficiency**: Concentration bounds on gradient deviation after quantization or sparsification, ensuring convergence of distributed training
 
 ## Engineering Feasibility
-- **Dimension 1 Tensorization ✅**: The bounds themselves are scalar formulas with no tensor operations; zero overhead as an analytical tool
-- **Dimension 2 GEMM-mappability ✅**: Does not directly participate in GEMM, but can serve as a theoretical basis for hyperparameter selection (batch size, compression ratio)
-- **Dimension 3 Complexity ✅**: Computing the bound itself is $O(1)$ or $O(n)$, very low cost
-- **Dimension 5 Low Precision ✅**: Bounds can be computed in fp32; does not enter the training backbone
-- **Dimension 8 Operator Fusion ✅**: Does not enter the computation graph; no fusion overhead
+- **D1[v]**: The bounds themselves are scalar formulas with no tensor operations; zero overhead as an analytical tool
+- **D2[v]**: Does not directly participate in GEMM, but can serve as a theoretical basis for hyperparameter selection (batch size, compression ratio)
+- **D3[v]**: Computing the bound itself is $O(1)$ or $O(n)$, very low cost
+- **D5[v]**: Bounds can be computed in fp32; does not enter the training backbone
+- **D8[v]**: Does not enter the computation graph; no fusion overhead
 
 ## Risks and Failure Conditions
 - **Independence assumption violated**: In sequential data and autoregressive models, tokens are strongly correlated, and Hoeffding's exponential decay guarantee fails. Martingale versions (Azuma-Hoeffding) or mixing-time corrections are required.

package/knowledge-base/probability/concentration-inequality.md

@@ -28,11 +28,11 @@ $$P(|f(X_1,\ldots,X_n) - \mathbb{E}[f]| \geq t) \leq 2\exp\left(-\frac{2t^2}{\su
 - **梯度压缩 / 通信效率**：量化或稀疏化后梯度偏差的浓度界，确保分布式训练收敛
 
 ## 工程可行性
-- **维度 1 张量化 ✅**：界本身是标量公式，不涉及张量运算，作为分析工具零开销
-- **维度 2 GEMM 可映射 ✅**：不直接参与 GEMM，但可作为超参数选择（batch size、压缩率）的理论依据
-- **维度 3 复杂度 ✅**：计算界本身 $O(1)$ 或 $O(n)$，极低
-- **维度 5 低精度 ✅**：界用 fp32 计算即可，不进入训练主干
-- **维度 8 算子融合 ✅**：不进入计算图，无融合负担
+- **D1[v]**：界本身是标量公式，不涉及张量运算，作为分析工具零开销
+- **D2[v]**：不直接参与 GEMM，但可作为超参数选择（batch size、压缩率）的理论依据
+- **D3[v]**：计算界本身 $O(1)$ 或 $O(n)$，极低
+- **D5[v]**：界用 fp32 计算即可，不进入训练主干
+- **D8[v]**：不进入计算图，无融合负担
 
 ## 风险与失效条件
 - **独立性假设不成立**：序列数据、自回归模型中 token 间强相关，Hoeffding 的指数衰减保证失效。需用 Martingale 版本（Azuma-Hoeffding）或混合时间修正。

package/knowledge-base/probability/entropy.en.md

@@ -30,11 +30,11 @@ $$I(X; Y) = H(X) - H(X|Y) = H(Y) - H(Y|X) = \sum_{x,y} p(x,y) \log \frac{p(x,y)}
 - **Variational Autoencoder (VAE)**: ELBO = reconstruction likelihood $-$ KL regularization term; it fundamentally balances information compression (low $H(Z)$) against reconstruction fidelity
 
 ## Engineering Feasibility
-- **Dimension 1 Tensorization ✅**: $-\sum p \log p$ is an element-wise operation, perfectly vectorizable
-- **Dimension 2 GEMM-mappability ⚠️**: Entropy itself is not a GEMM, but the gradient computation of the cross-entropy loss involves a softmax-to-matmul chain
-- **Dimension 3 Complexity ✅**: $O(|\mathcal{X}|)$ linear; acceptable for vocabulary-level computation
-- **Dimension 5 Low Precision ✅**: $\log$ and exp are stable in bf16; softmax benefits from the log-sum-exp trick
-- **Dimension 8 Operator Fusion ✅**: softmax + cross-entropy is a classic fused operator (FusedSoftmaxCrossEntropy)
+- **D1[v]**: $-\sum p \log p$ is an element-wise operation, perfectly vectorizable
+- **D2[~]**: Entropy itself is not a GEMM, but the gradient computation of the cross-entropy loss involves a softmax-to-matmul chain
+- **D3[v]**: $O(|\mathcal{X}|)$ linear; acceptable for vocabulary-level computation
+- **D5[v]**: $\log$ and exp are stable in bf16; softmax benefits from the log-sum-exp trick
+- **D8[v]**: softmax + cross-entropy is a classic fused operator (FusedSoftmaxCrossEntropy)
 
 ## Risks and Failure Conditions
 - **Continuous entropy can be negative**: Differential entropy $h(X)$ is not constrained by $H(X) \geq 0$; directly comparing differential entropies of different dimensions can be misleading. Mutual information or KL divergence (which are non-negative) should be used instead.

package/knowledge-base/probability/entropy.md

@@ -30,11 +30,11 @@ $$I(X; Y) = H(X) - H(X|Y) = H(Y) - H(Y|X) = \sum_{x,y} p(x,y) \log \frac{p(x,y)}
 - **变分自编码器 VAE**：ELBO = 重构似然 $-$ KL 正则项，本质是在信息压缩（低 $H(Z)$）与重构保真之间取平衡
 
 ## 工程可行性
-- **维度 1 张量化 ✅**：$-\sum p \log p$ 是逐元素运算，完美向量化
-- **维度 2 GEMM 可映射 ⚠️**：熵本身不是 GEMM，但交叉熵损失的梯度计算涉及 softmax → matmul 链
-- **维度 3 复杂度 ✅**：$O(|\mathcal{X}|)$ 线性，vocab 级计算可接受
-- **维度 5 低精度 ✅**：$\log$ 和 exp 在 bf16 下稳定，softmax 有 log-sum-exp 技巧
-- **维度 8 算子融合 ✅**：softmax + cross-entropy 是经典融合算子（FusedSoftmaxCrossEntropy）
+- **D1[v]**：$-\sum p \log p$ 是逐元素运算，完美向量化
+- **D2[~]**：熵本身不是 GEMM，但交叉熵损失的梯度计算涉及 softmax → matmul 链
+- **D3[v]**：$O(|\mathcal{X}|)$ 线性，vocab 级计算可接受
+- **D5[v]**：$\log$ 和 exp 在 bf16 下稳定，softmax 有 log-sum-exp 技巧
+- **D8[v]**：softmax + cross-entropy 是经典融合算子（FusedSoftmaxCrossEntropy）
 
 ## 风险与失效条件
 - **连续熵可为负**：微分熵 $h(X)$ 不受 $H(X) \geq 0$ 约束，直接比较不同量纲的微分熵可产生误导。应改用互信息或 KL 散度（非负）。

package/knowledge-base/probability/fisher-information.en.md

@@ -28,13 +28,13 @@ $$D_{KL}(p_\theta \| p_{\theta + d\theta}) \approx \frac{1}{2} d\theta^T \mathca
 - **Sensitivity analysis for pretraining-finetuning**: Parameter directions with high Fisher information = parameters sensitive to data; these should be handled more carefully during fine-tuning
 
 ## Engineering Feasibility
-- **Dimension 1 Tensorization ⚠️**: The full FIM is a $d \times d$ matrix ($d$ = number of parameters); direct materialization is infeasible (LLM parameter counts $10^{10}+$). Approximations are required.
-- **Dimension 2 GEMM-mappability ✅**: K-FAC uses Kronecker factors $A \otimes B$; $A$ and $B$ can each be computed and inverted using GEMM
-- **Dimension 3 Complexity ⚠️**: Exact FIM computation is $O(nd^2)$; K-FAC reduces this to $O(d)$ scale but requires per-layer maintenance
-- **Dimension 4 Memory ⚠️**: K-FAC's Kronecker factors require additional memory, though significantly compressed compared to the full FIM
-- **Dimension 5 Low Precision ✅**: FIM estimation can use fp32; fp64 is not required
-- **Dimension 6 Parallelism ✅**: K-FAC's Kronecker factors naturally decompose by layer, enabling parallel computation
-- **Dimension 8 Operator Fusion ✅**: The EWC penalty term is element-wise and can be fused into the parameter update kernel
+- **D1[~]**: The full FIM is a $d \times d$ matrix ($d$ = number of parameters); direct materialization is infeasible (LLM parameter counts $10^{10}+$). Approximations are required.
+- **D2[v]**: K-FAC uses Kronecker factors $A \otimes B$; $A$ and $B$ can each be computed and inverted using GEMM
+- **D3[~]**: Exact FIM computation is $O(nd^2)$; K-FAC reduces this to $O(d)$ scale but requires per-layer maintenance
+- **D4[~]**: K-FAC's Kronecker factors require additional memory, though significantly compressed compared to the full FIM
+- **D5[v]**: FIM estimation can use fp32; fp64 is not required
+- **D6[v]**: K-FAC's Kronecker factors naturally decompose by layer, enabling parallel computation
+- **D8[v]**: The EWC penalty term is element-wise and can be fused into the parameter update kernel
 
 ## Risks and Failure Conditions
 - **Full FIM is intractable**: For LLM-scale parameter counts ($d > 10^9$), even K-FAC's Kronecker approximation may be too costly. In practice, diagonal Fisher ($O(d)$) or low-rank approximations are commonly used.

package/knowledge-base/probability/fisher-information.md

@@ -28,13 +28,13 @@ $$D_{KL}(p_\theta \| p_{\theta + d\theta}) \approx \frac{1}{2} d\theta^T \mathca
 - **预训练-微调的敏感度分析**：Fisher 信息大的参数方向 = 对数据敏感的参数，微调时应更谨慎
 
 ## 工程可行性
-- **维度 1 张量化 ⚠️**：完整 FIM 是 $d \times d$ 矩阵（$d$ = 参数量），直接物化不可行（LLM 参数量 $10^{10}+$）。必须用近似。
-- **维度 2 GEMM 可映射 ✅**：K-FAC 用 Kronecker 因子 $A \otimes B$，$A$ 和 $B$ 各自可用 GEMM 计算和求逆
-- **维度 3 复杂度 ⚠️**：精确 FIM 计算 $O(nd^2)$，K-FAC 降至 $O(d)$ 量级但需逐层维护
-- **维度 4 显存 ⚠️**：K-FAC 的 Kronecker 因子需额外显存，但相比完整 FIM 已大幅压缩
-- **维度 5 低精度 ✅**：FIM 估计用 fp32 即可，不需 fp64
-- **维度 6 并行 ✅**：K-FAC 的 Kronecker 因子天然按层分解，可并行计算
-- **维度 8 算子融合 ✅**：EWC 惩罚项为逐元素运算，可融入参数更新 kernel
+- **D1[~]**：完整 FIM 是 $d \times d$ 矩阵（$d$ = 参数量），直接物化不可行（LLM 参数量 $10^{10}+$）。必须用近似。
+- **D2[v]**：K-FAC 用 Kronecker 因子 $A \otimes B$，$A$ 和 $B$ 各自可用 GEMM 计算和求逆
+- **D3[~]**：精确 FIM 计算 $O(nd^2)$，K-FAC 降至 $O(d)$ 量级但需逐层维护
+- **D4[~]**：K-FAC 的 Kronecker 因子需额外显存，但相比完整 FIM 已大幅压缩
+- **D5[v]**：FIM 估计用 fp32 即可，不需 fp64
+- **D6[v]**：K-FAC 的 Kronecker 因子天然按层分解，可并行计算
+- **D8[v]**：EWC 惩罚项为逐元素运算，可融入参数更新 kernel
 
 ## 风险与失效条件
 - **完整 FIM 不可计算**：对于 LLM 级别的参数规模（$d > 10^9$），即使 K-FAC 的 Kronecker 近似也可能代价过高。实践中常用对角 Fisher（$O(d)$）或低秩近似。

package/knowledge-base/probability/information-bottleneck.en.md

@@ -30,12 +30,12 @@ where $q_\phi(y|z)$ is the classifier/decoder, $r(z)$ is the prior distribution
 - **Information-theoretic interpretation of attention sparsification / routing**: Sparse Attention and MoE routing can be understood as implicit information bottlenecks — selectively allowing "useful" tokens to pass while discarding noise
 
 ## Engineering Feasibility
-- **Dimension 1 Tensorization ✅**: The VIB encoder/decoder are standard networks; $D_{KL}$ is computed element-wise
-- **Dimension 2 GEMM-mappability ✅**: The main computation is a standard feedforward network + GEMM
-- **Dimension 3 Complexity ✅**: Only adds $O(d)$ computation for the KL term compared to the original network
-- **Dimension 4 Memory ⚠️**: Requires additional parameters for the prior distribution $r(z)$ and intermediate quantities for KL computation
-- **Dimension 5 Low Precision ✅**: Reparameterization trick + analytical KL solution are stable in bf16
-- **Dimension 8 Operator Fusion ✅**: No conflict with the standard training pipeline; normal fusion applies
+- **D1[v]**: The VIB encoder/decoder are standard networks; $D_{KL}$ is computed element-wise
+- **D2[v]**: The main computation is a standard feedforward network + GEMM
+- **D3[v]**: Only adds $O(d)$ computation for the KL term compared to the original network
+- **D4[~]**: Requires additional parameters for the prior distribution $r(z)$ and intermediate quantities for KL computation
+- **D5[v]**: Reparameterization trick + analytical KL solution are stable in bf16
+- **D8[v]**: No conflict with the standard training pipeline; normal fusion applies
 
 ## Risks and Failure Conditions
 - **Accurate estimation of $I(X;Z)$ is difficult**: Mutual information estimation between high-dimensional continuous variables is itself an open problem (estimators such as MINE and NWJ have high variance). In practice, the VIB variational lower bound is used as a workaround, but the bound may be loose.

package/knowledge-base/probability/information-bottleneck.md

@@ -30,12 +30,12 @@ $$\mathcal{L}_{VIB} = \mathbb{E}_{p(x,y)}[-\log q_\phi(y|z)] + D_{KL}(p_\theta(z
 - **注意力稀疏化 / 路由的信息论解释**：Sparse Attention 和 MoE 路由可理解为隐式信息瓶颈——选择性地让"有用" token 通过，丢弃噪声
 
 ## 工程可行性
-- **维度 1 张量化 ✅**：VIB 的编码器/解码器均为标准网络，$D_{KL}$ 为逐元素计算
-- **维度 2 GEMM 可映射 ✅**：主体计算为标准前向网络 + GEMM
-- **维度 3 复杂度 ✅**：相比原始网络仅增加 KL 项的 $O(d)$ 计算
-- **维度 4 显存 ⚠️**：需要额外维护先验分布 $r(z)$ 的参数和 KL 计算中间量
-- **维度 5 低精度 ✅**：重参数化技巧 + KL 解析解在 bf16 下稳定
-- **维度 8 算子融合 ✅**：与标准训练流程无冲突，可正常融合
+- **D1[v]**：VIB 的编码器/解码器均为标准网络，$D_{KL}$ 为逐元素计算
+- **D2[v]**：主体计算为标准前向网络 + GEMM
+- **D3[v]**：相比原始网络仅增加 KL 项的 $O(d)$ 计算
+- **D4[~]**：需要额外维护先验分布 $r(z)$ 的参数和 KL 计算中间量
+- **D5[v]**：重参数化技巧 + KL 解析解在 bf16 下稳定
+- **D8[v]**：与标准训练流程无冲突，可正常融合
 
 ## 风险与失效条件
 - **$I(X;Z)$ 的精确估计困难**：高维连续变量间的互信息估计本身是开放问题（MINE、NWJ 等估计器方差大），实际用 VIB 变分下界绕过，但下界可能很松。

package/knowledge-base/probability/kl-divergence.en.md

@@ -33,12 +33,12 @@ $$D_{KL}(p \| q) = H(p, q) - H(p)$$
 - **PPO / RLHF**: $D_{KL}(\pi_\theta \| \pi_{\text{ref}})$ serves as a penalty term for the policy deviating from the reference policy
 
 ## Engineering Feasibility
-- **Dimension 1 Tensorization ✅**: Element-wise $p \log(p/q)$ is fully vectorizable
-- **Dimension 2 GEMM-mappability ⚠️**: KL itself is not a GEMM, but its inputs (logits) come from GEMM layers
-- **Dimension 3 Complexity ✅**: $O(|\mathcal{X}|)$ linear
-- **Dimension 4 Memory ⚠️**: For large vocabularies, both $p$ and $q$ probability vectors must be held simultaneously; chunked computation is possible
-- **Dimension 5 Low Precision ✅**: Log-softmax differences are stable in bf16; note that $\log q$ diverges as $q \to 0$, requiring clamping
-- **Dimension 8 Operator Fusion ✅**: Can be fused with softmax into FusedKLDivLoss
+- **D1[v]**: Element-wise $p \log(p/q)$ is fully vectorizable
+- **D2[~]**: KL itself is not a GEMM, but its inputs (logits) come from GEMM layers
+- **D3[v]**: $O(|\mathcal{X}|)$ linear
+- **D4[~]**: For large vocabularies, both $p$ and $q$ probability vectors must be held simultaneously; chunked computation is possible
+- **D5[v]**: Log-softmax differences are stable in bf16; note that $\log q$ diverges as $q \to 0$, requiring clamping
+- **D8[v]**: Can be fused with softmax into FusedKLDivLoss
 
 ## Risks and Failure Conditions
 - **KL diverges to infinity when $q(x)=0$ but $p(x)>0$**: In practice, label smoothing or temperature scaling must be applied to $q$ to avoid zero probabilities. The mode-seeking behavior of reverse KL can exacerbate this issue — the student model "drops" low-probability regions of the teacher distribution.

package/knowledge-base/probability/kl-divergence.md

@@ -33,12 +33,12 @@ $$D_{KL}(p \| q) = H(p, q) - H(p)$$
 - **PPO / RLHF**：$D_{KL}(\pi_\theta \| \pi_{\text{ref}})$ 作为策略偏离参考策略的惩罚项
 
 ## 工程可行性
-- **维度 1 张量化 ✅**：逐元素 $p \log(p/q)$ 完全向量化
-- **维度 2 GEMM 可映射 ⚠️**：KL 本身不是 GEMM，但输入（logits）来自 GEMM 层
-- **维度 3 复杂度 ✅**：$O(|\mathcal{X}|)$ 线性
-- **维度 4 显存 ⚠️**：大 vocab 下需同时保留 $p$ 和 $q$ 的完整概率向量，可 chunk 计算
-- **维度 5 低精度 ✅**：log-softmax 差值在 bf16 下稳定；注意 $q \to 0$ 时 $\log q$ 发散，需 clamp
-- **维度 8 算子融合 ✅**：可与 softmax 融合为 FusedKLDivLoss
+- **D1[v]**：逐元素 $p \log(p/q)$ 完全向量化
+- **D2[~]**：KL 本身不是 GEMM，但输入（logits）来自 GEMM 层
+- **D3[v]**：$O(|\mathcal{X}|)$ 线性
+- **D4[~]**：大 vocab 下需同时保留 $p$ 和 $q$ 的完整概率向量，可 chunk 计算
+- **D5[v]**：log-softmax 差值在 bf16 下稳定；注意 $q \to 0$ 时 $\log q$ 发散，需 clamp
+- **D8[v]**：可与 softmax 融合为 FusedKLDivLoss
 
 ## 风险与失效条件
 - **$q(x)=0$ 但 $p(x)>0$ 时 KL 发散为无穷**：实践中必须对 $q$ 做 label smoothing 或温度缩放，避免零概率。反向 KL 的 mode-seeking 行为可加剧此问题——学生模型"丢弃"教师分布的低概率区域。

package/lenses/algorithmic.en.md

@@ -1,4 +1,4 @@
-# 🖥️ Algorithmic Lens
+# Algorithmic Lens
 
 > Algorithms are the automation of thought — converting insights into precise, repeatable steps.
 

package/lenses/algorithmic.md

@@ -1,4 +1,4 @@
-# 🖥️ 算法透镜 / Algorithmic Lens
+# 算法透镜 / Algorithmic Lens
 
 > 算法是思想的自动化——将洞察转化为可重复执行的精确步骤。
 

package/lenses/axiomatization.en.md

@@ -1,4 +1,4 @@
-# 📐 Axiomatization Lens
+# Axiomatization Lens
 
 > Build from minimal assumptions using rigorous logic — scrutinizing premises matters more than checking conclusions
 

package/lenses/axiomatization.md

@@ -1,4 +1,4 @@
-# 📐 公理化透镜
+# 公理化透镜
 
 > 从最少假设出发，用严格逻辑构建——审查前提比检查结论更重要
 

package/lenses/categorical.en.md

@@ -1,4 +1,4 @@
-# 🧩 Categorification Lens
+# Categorification Lens
 
 > Strip away incidental details to reveal essential structure — problems across different domains often share the same abstract skeleton
 

package/lenses/categorical.md

@@ -1,4 +1,4 @@
-# 🧩 范畴化透镜
+# 范畴化透镜
 
 > 忽略偶然细节，揭示必然结构——不同领域的问题往往共享相同的抽象骨架
 

package/lenses/causal.en.md

@@ -1,4 +1,4 @@
-# 🔗 Causal Lens
+# Causal Lens
 
 > Correlation does not imply causation — but causation can be clarified. The crucial distinction: "how Y behaves when X is observed" is not the same as "what would happen if X were done."
 

package/lenses/causal.md

@@ -1,4 +1,4 @@
-# 🔗 因果透镜 / Causal Lens
+# 因果透镜 / Causal Lens
 
 > 相关不等于因果——但因果可以理清。关键区别："看到 X 时 Y 如何"不等于"如果做了 X 会怎样"。
 

package/lenses/duality.en.md

@@ -1,4 +1,4 @@
-# 🔄 Duality Lens
+# Duality Lens
 
 > A hard problem is equivalent to an easy problem — the key is finding the right transform and its inverse
 

package/lenses/duality.md

@@ -1,4 +1,4 @@
-# 🔄 对偶透镜
+# 对偶透镜
 
 > 复杂问题等价于简单问题——关键是找到合适的变换和它的逆
 

package/lenses/game.en.md

@@ -1,4 +1,4 @@
-# 🎯 Game Lens
+# Game Lens
 
 > Your optimal choice depends on others' choices — thinking must be not only deep but interactive.
 
@@ -15,7 +15,7 @@ This is a "player's" perspective — in environments where multiple decision-mak
 
 ## What Problems It Is Not Suited For
 
-- Single-agent decision problems with no interaction with others — optimization thinking is more appropriate
+- Single-agent decision problems with no interaction with others — the variational lens is more appropriate
 - Purely cooperative problems where all participants share identical objectives — strategic analysis is unnecessary
 - Deterministic problems where outcomes are uniquely determined by one's own actions — no others' responses are involved
 - Outcomes determined purely by luck, with no strategic choices available to any party

package/lenses/game.md

@@ -1,4 +1,4 @@
-# 🎯 博弈透镜 / Game Lens
+# 博弈透镜 / Game Lens
 
 > 你的最优选择取决于他人的选择——思考不仅要深入，还要互动。
 
@@ -15,7 +15,7 @@
 
 ## 不适合什么问题
 
-- 单人决策问题，与他人无互动——优化思想更适用
+- 单人决策问题，与他人无互动——变分透镜更适用
 - 纯合作问题，参与者目标完全一致——无需战略分析
 - 确定性问题，结果由自身行动唯一决定——不涉及他人反应
 - 纯粹运气决定的结果，各方无策略选择

package/lenses/geometric.en.md

@@ -1,4 +1,4 @@
-# 🌉 Geometric Lens
+# Geometric Lens
 
 > Metric, curvature, and spatial structure on manifolds — seeing problems through the eyes of differential geometry, where distance, bending, and optimal paths reveal the intrinsic shape of data
 

package/lenses/geometric.md

@@ -1,4 +1,4 @@
-# 🌉 几何透镜
+# 几何透镜
 
 > 度量/曲率/流形上的空间结构——用微分几何的眼光看问题，距离、弯曲与最优路径揭示数据的内在形状
 

package/lenses/local-to-global.en.md

@@ -1,4 +1,4 @@
-# 📈 Local-to-Global Lens
+# Local-to-Global Lens
 
 > Patching local properties into global, sheaf cohomology obstructions — can local solutions be seamlessly assembled into a global solution? Where do the obstructions lie?
 

package/lenses/local-to-global.md

@@ -1,4 +1,4 @@
-# 📈 局部到全局透镜
+# 局部到全局透镜
 
 > 局部性质拼接为全局、层上同调障碍——局部解能否无缝组装成全局解？障碍在哪里？
 

package/lenses/perturbation.en.md

@@ -1,4 +1,4 @@
-# 🧠 Perturbation Lens
+# Perturbation Lens
 
 > Propagation of small perturbations, stability, and robustness — a system's sensitivity to tiny changes determines its reliability
 

package/lenses/perturbation.md

@@ -1,4 +1,4 @@
-# 🧠 微扰透镜
+# 微扰透镜
 
 > 小扰动的传播、稳定性、鲁棒性——系统对微小变化的敏感度决定了它的可靠性
 

package/lenses/probabilistic.en.md

@@ -1,4 +1,4 @@
-# 🎲 Probabilistic Lens
+# Probabilistic Lens
 
 > Uncertainty is quantifiable — probability is quantified belief, data is quantified evidence
 

package/lenses/probabilistic.md

@@ -1,4 +1,4 @@
-# 🎲 概率透镜
+# 概率透镜
 
 > 不确定性是可量化的——概率是量化的信念，数据是量化的证据
 

package/lenses/projection.en.md

@@ -1,4 +1,4 @@
-# 🧮 Projection & Decomposition Lens
+# Projection & Decomposition Lens
 
 > Complex wholes can be orthogonally decomposed into independent components — conflicts are exposed in subspaces, signal and noise are separated under projection.
 

package/lenses/projection.md

@@ -1,4 +1,4 @@
-# 🧮 投影与分解透镜 / Projection & Decomposition Lens
+# 投影与分解透镜 / Projection & Decomposition Lens
 
 > 复杂整体可以正交分解为独立分量——冲突在子空间中暴露，信号与噪声在投影下分离。
 

package/lenses/spectral.en.md

@@ -1,4 +1,4 @@
-# 📡 Spectral Decomposition Lens
+# Spectral Decomposition Lens
 
 > Any linear operator can be decomposed into a superposition of eigencomponents — eigenvalues reveal dominant structure, eigenvectors reveal dominant directions.
 

package/lenses/spectral.md

@@ -1,4 +1,4 @@
-# 📡 谱分解透镜 / Spectral Decomposition Lens
+# 谱分解透镜 / Spectral Decomposition Lens
 
 > 任何线性算子都可以分解为特征分量的叠加——特征值揭示主导结构，特征向量揭示主导方向。
 

package/lenses/symmetry.en.md

@@ -1,4 +1,4 @@
-# ⚛️ Symmetry Lens
+# Symmetry Lens
 
 > Seek the invariant amid change — every symmetry corresponds to a conserved quantity, and every invariant is a key to simplifying the problem
 

package/lenses/symmetry.md

@@ -1,4 +1,4 @@
-# ⚛️ 对称透镜
+# 对称透镜
 
 > 在变化中寻找不变——每一个对称性都对应一个守恒量，每一个不变量都是简化问题的钥匙
 

package/lenses/topological.en.md

@@ -1,4 +1,4 @@
-# 🌀 Topological Lens
+# Topological Lens
 
 > Stretching and bending are allowed, but tearing is not — the truly important properties are those invariant under continuous deformation
 

package/lenses/topological.md

@@ -1,4 +1,4 @@
-# 🌀 拓扑透镜
+# 拓扑透镜
 
 > 可以拉伸弯曲但不能撕裂——真正重要的性质在连续变形中不变
 

package/lenses/variational.en.md

@@ -1,4 +1,4 @@
-# ⚖️ Variational Lens
+# Variational Lens
 
 > Pursue the best among the feasible under constraints — convexity determines difficulty, duality reveals structure
 

package/lenses/variational.md

@@ -1,4 +1,4 @@
-# ⚖️ 变分透镜
+# 变分透镜
 
 > 在约束下追求可行中的最优——凸性决定难度，对偶揭示结构
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "math-skill",
-  "version": "3.0.0",
+  "version": "3.0.1",
   "description": "面向 AI 架构创新的数学研究操作系统 v3：三层正交架构（思想透镜 / 数学知识库 / 设计翻译层），自动诊断用户意图并路由。Math Research OS for AI architecture design: thinking lenses + math knowledge base + design translation layer with auto intent diagnosis.",
   "keywords": [
     "math",

package/references/agentic-workflow.en.md

@@ -38,7 +38,7 @@ Record research tasks in a table that the Agent can continuously update:
 ```markdown
 | Candidate Structure | Math Correctness | GPU 8-Dim | Complexity | Status | Notes / Next Steps |
 |--------------------|-----------------|-----------|------------|--------|-------------------|
-| Tropical gating replacing TopK | ✅ Differentiable relaxation | 1✅2❌3✅… | Sub-quadratic (per-token) | Verifying | Compare with SWA baseline |
+| Tropical gating replacing TopK | [v] Differentiable relaxation | 1[v]2[x]3[v]… | Sub-quadratic (per-token) | Verifying | Compare with SWA baseline |
 | …                  |                 |           |            | Todo   |                   |
 ```
 

package/references/agentic-workflow.md

@@ -40,7 +40,7 @@
 ```markdown
 | 候选结构 | 数学正确性 | GPU 八维 | 复杂度 | 状态 | 备注/下一步 |
 |---------|-----------|---------|-------|------|-----------|
-| 热带门控替代 TopK | ✅ 可微松弛 | 1✅2❌3✅… | 亚二次(逐token) | 验证中 | 对比 SWA 基线 |
+| 热带门控替代 TopK | [v] 可微松弛 | 1[v]2[x]3[v]… | 亚二次(逐token) | 验证中 | 对比 SWA 基线 |
 | …            |           |         |       | 待办 |           |
 ```