The Data数据洞察

What the scored docket reveals

评分名单透露了什么

Findings from an original analysis of the 463-project docket plus a close read of 37 complete Physics applications (11 on the table, 26 not). All applicant work is described as anonymized archetypes — field, method, setting, result — never by name.

以下发现来自对 463 项评分名单的原创分析，以及对 37 份完整物理类申请（11 份入选、26 份未入选）的逐页精读。所有申请者的工作一律以匿名原型呈现——只讲领域、方法、场所与结果，绝不点名。

463 scored projects · 2,471 entries463 项评分 · 2,471 份申请

Geography地域分布

Half the table comes from two states半张名单来自两个州

Share of the 463 "On The Table" projects, by state463 项入选项目的州别占比

New York

35%

California

14%

New Jersey

Texas

North Carolina

All other states其余各州合计

~35%

Bars scaled to New York = 100%. About 12 research high schools alone account for ~25% of the entire table.条形以纽约州为 100% 基准。仅约 12 所科研强校就占了整张名单的约 25%。

Research Setting研究场所

Where selected vs unselected Physics projects were done入选与未入选的物理项目分别在哪里完成

On the table (n = 11)入选组（n = 11）

University lab大学实验室

64%

Home / online在家 / 线上

18%

Not on the table (n = 26)未入选组（n = 26）

University lab大学实验室

~4%

Home / online在家 / 线上

46%

Small, non-random sample — directional, not exact. Remaining shares are regional colleges and other settings.样本小且非随机——只看方向，不看精确值。其余份额为地区院校及其他场所。

Readings解读

Four honest readings四条诚实的解读

Structural advantage is real. Research-magnet schools, mentor pipelines, and university proximity feed exactly what the rubric rewards — sophisticated methodology and credentialed recommendations. Institution access is a correlate, not a criterion: it stands in for the things actually scored.结构性优势真实存在。科研特色高中、导师渠道与大学资源，恰好喂养了量规所奖励的东西——高深的方法论与有分量的推荐信。但机构资源只是相关项，不是评分项：它是真正被打分的那些量的替身。
Home-based does not mean weak. The two fully home-based projects that reached the top tier were theory / computational / public-data work — topics where a laptop is the entire instrument, so lab access carries zero methodological penalty.在家做研究不等于弱。两个完全在家完成却跻身高分段的项目，都是理论 / 计算 / 公开数据类课题——笔记本电脑就是全部仪器，实验室的缺席不构成任何方法论劣势。
The real lesson is matching. Pick a question your setting can credibly execute: lab-dependent topics need real lab access; without it, choose a rigorous theory, simulation, or public-archive question instead.真正的教训是「匹配」。选一个你的条件能够可信执行的问题：依赖实验室的课题就要有真实的实验室；没有的话，就转向能严谨完成的理论、模拟或公开数据课题。
Gender is near-even on the table (roughly 51% / 47%) and tracks the entrant pool — not a visible selection lever.入选名单的性别比例接近均衡（约 51% / 47%），与申请者整体一致——不是可见的筛选变量。

How to read these numbers这些数字该怎么读 The setting comparison is Physics-only and small-sample, drawn from a coaching collection where both groups were already strong. Non-selected projects carry no scores in the dataset, so selected-vs-not is a content inference, not a score regression. Treat percentages as directional. 场所对比仅限物理类且样本很小，取自一个两组都已相当强的辅导样本库。数据集中未入选项目没有分数，因此「入选与否」的对比是内容层面的推断，不是分数回归。百分比只看方向。

Next: turn the findings into moves — The Playbook →

下一步：把发现变成行动——进阶攻略 →