The Playbook进阶攻略
FLOOR × DIFFERENTIATORS
底线 × 差异化
Built by reading complete top-tier submissions spanning the whole scored range — 18.6/20 down to the 16.5/20 floor — plus a 19.33/20 calibration exemplar. First clear a non-negotiable floor on all four criteria, then stack differentiators.
本模型来自通读覆盖整个分数段的完整入选申请——从 18.6/20 到 16.5/20 的底线案例——外加一份 19.33/20 的校准范本。先在四项标准上全部清过一条不可谈判的底线，再叠加差异化优势。
floor first · then stack moves先清底线 · 再叠亮点
The floor every winner cleared每位入选者都清过的底线
Entry Form 申请材料
Floor底线SAT 1510–1600 (most ≥1550) or ACT 35–36; 6–8 AP 5s; genuine college-level coursework; at least one founded or led activity.SAT 1510–1600（多数 ≥1550）或 ACT 35–36；6–8 门 AP 满分；货真价实的大学水平课程；至少一项自己创办或主导的活动。
Scientific Merit 科学价值
Floor底线Graduate-level methods — PDE / spectral solvers, DFT, Monte-Carlo, real statistics — correctly executed. Not a science-fair demo.研究生水平的方法——偏微分方程/谱方法求解器、密度泛函、蒙特卡洛、真正的统计分析——并正确执行。不是科学展演示。
Contribution 学生贡献
Floor底线The student demonstrably did the core work (wrote the code, ran the analysis, derived the result) — and the mentor letter corroborates it specifically.学生可证明地完成了核心工作（写代码、跑分析、推导结果）——且导师推荐信给出具体的佐证。
Potential 科学潜力
Floor底线An accessible layperson summary with a real hook or analogy, plus a distinctive personal essay tying identity to science.一篇有真实钩子或类比、外行也读得懂的摘要，加上一篇把个人身份与科学连起来、令人记住的个人陈述。
If you cannot honestly check all four cards, the priority is fixing the gap — not adding polish to a strength.
如果四张卡里有任何一张无法诚实打勾，第一优先级是补短板——而不是继续打磨长板。
The seven moves that separate 18+ from 16.5把 16.5 拉到 18+ 的七个动作
-
Cross-validate every key claim with a second, independent method每个关键结论都用第二种独立方法交叉验证
The strongest rigor signal in the corpus: theory checked against simulation checked against experiment; a new fitting method benchmarked against the field standard; an analytically solvable toy problem verified to <0.1% error. One result confirmed two ways beats two unconfirmed results.这是整个语料中最强的严谨性信号：理论对照模拟再对照实验；新拟合方法与领域标准方法对标；用可解析求解的玩具问题把误差验证到 0.1% 以内。一个被两种方法确认的结果，胜过两个未经确认的结果。
-
Contribute something new — don't just run an existing tool贡献新东西——而不只是运行现成工具
Add a term, a method, a generalization: a new energy term derived by hand and added to a simulator; a published correspondence proven under far more general conditions. The floor cases merely reproduced results already known analytically.加一个项、一种方法、一次推广：亲手推导一个新能量项并加入模拟器；把一条已发表的对应关系证明到更一般的条件下。而底线案例只是复现了解析上已知的结果。
-
A rigorous null result is a winning result严谨的阴性结果同样是赢面
Ruling out a candidate explanation for a famous anomaly, or honestly refuting your own hypothesis with full statistical hygiene, both scored high — because the rigor, not the "yes," carries the merit. Frame a null as shrinking the field's search space.排除某著名反常信号的一种候选解释，或以完备的统计规范诚实推翻自己的假设，都拿到了高分——因为承载科学价值的是严谨，不是「是」。把阴性结果表述为帮领域缩小了搜索空间。
-
Get an external validation stamp拿到一枚外部认证的图章
A first-author abstract, a single-author arXiv paper, a peer-reviewed publication, an ISEF Grand Award, an Olympiad medal — each removes the "is this really the student's work, and is it any good?" doubt in one stroke.第一作者摘要、独立作者 arXiv 论文、同行评审发表、ISEF 大奖、奥赛奖牌——任何一枚都能一举打消「这真是学生自己做的吗、水平到底如何」的疑虑。
-
Own the question — and bound the mentor's role explicitly拥有你的问题——并明确圈定导师的角色
Top scorers either found the question themselves or took an assigned seed and visibly extended it — adding analyses the mentor had not suggested. Name exactly what help was given (a dataset, stability tips, starter code) so the independent core is unmistakable.高分者要么自己找到问题，要么把导师给的种子课题做出看得见的延伸——补上导师从未提过的分析。准确写明得到过哪些帮助（数据集、数值稳定性建议、起步代码），让独立完成的核心部分无可置疑。
-
Quantify significance — and repeat the number量化你的贡献——并反复重复那个数字
"76% less power per chip." "Mass error improved from 100 keV to 10 keV." "8.6% median error, continent-wide." A crisp figure threaded through the paper, summary, and essays is what evaluators remember and re-cite.「单芯片功耗降低 76%」「质量误差从 100 keV 压到 10 keV」「全大陆中位误差 8.6%」。一个干脆利落的数字贯穿论文、摘要与短文——这正是评审记得住、还会转述的东西。
-
Win the communication axis赢下表达这条轴
Many strong quantitative applicants lose points here, so it is cheap differentiation: a journalist-quotable layperson summary, named takeaways, and a personal essay with a genuine stake — not a résumé in paragraph form.许多数理强手恰恰在这里丢分，所以这是性价比最高的差异化：一篇记者愿意引用的外行摘要、点得出名的结论要点，以及一篇有真实利害关系的个人陈述——而不是段落版简历。
Red flags that visibly cost points肉眼可见扣分的红旗
- "Limitations: none / N/A." Every real study has limitations; omitting them signals weak self-assessment and lowers merit. Always name 2–4 concrete ones and why they matter.「研究局限：无 / N/A」。任何真实研究都有局限；写「无」暴露的是自我评估能力的欠缺，直接拉低科学价值分。永远列出 2–4 条具体局限并说明其影响。
- Confirmatory results dressed as discovery — re-deriving a known result (or reproducing the mentor's own preprint) caps significance.把验证性结果包装成新发现——重推一个已知结论（甚至复现导师自己的预印本）会给重要性封顶。
- Thin data relative to the claims — five samples or three one-hour trials cannot support strong conclusions.数据量撑不起结论——五个样本、三次一小时的实验，托不住任何强结论。
- Close-relative mentorship or an assigned-and-not-extended question — properly disclosed, it is legal; but it visibly caps perceived independence.近亲担任导师，或课题被指派后未做延伸——如实披露并不违规，但会明显压低评审对独立性的判断。
- Thin or lukewarm recommendations — a generic "Top 15%" letter, or a near-content-free mentor form, undercuts the Contribution criterion even when the science is fine.单薄或不冷不热的推荐信——一封泛泛的「前 15%」评价，或几乎没有内容的导师表格，即使科研本身不错也会拖垮贡献分。
- Overclaiming impact — "revolutionize," "democratize" — ahead of demonstrated results, without staging who must do what next.夸大影响——在结果尚未证明之前就「颠覆」「普惠」，却不说明下一步该由谁做什么。
No lab? The home-based route to the top tier没有实验室？在家也能进高分段
- Choose a topic where a laptop is the entire instrument — theory, simulation, mathematical proof. Both fully home-based top-tier projects in our corpus were exactly this.选一个笔记本电脑就是全部仪器的课题——理论、模拟、数学证明。语料中两个完全在家完成的高分项目正是如此。
- Mine public archives — space-telescope data, satellite imagery, open scientific databases — and make rigor your moat: justify sample selection, validate your pipeline against an established tool, test statistical assumptions, and report nulls honestly.深挖公开数据档案——太空望远镜数据、卫星遥感影像、开放科学数据库——并把严谨性筑成护城河：论证样本选择、用成熟工具校验自己的流程、检验统计假设、诚实报告阴性结果。
- Turn "no mentor" into a documented strength — narrate the self-teaching and every analysis decision, so ownership is uncontested. The 19.33/20 calibration exemplar was a home-based, public-data project, later peer-reviewed and published.把「没有导师」写成有据可查的优势——把自学过程与每个分析决策讲清楚，让所有权无可争议。那份 19.33/20 的校准范本正是在家完成的公开数据项目，后来通过同行评审正式发表。
Pre-submission scorecard提交前自检清单
A submission that checks most of these lands in Top-400 territory; checking nearly all of them is what 18+ looks like.
大部分能打勾，就到了前 400 的水位线；几乎全部打勾，才是 18 分以上的样子。
Scientific Merit 科学价值
- Central claim validated by a second independent method or benchmark核心结论有第二种独立方法或基准的验证
- A specific novel element, stateable in one sentence有一句话就能说清的具体创新点
- 2–4 concrete limitations named — never "none"列出 2–4 条具体局限——绝不写「无」
- Dataset / result count defensible for the claims made数据量与结果数量撑得起所下的结论
Student Contribution 学生贡献
- Per phase, you can narrate exactly what you did vs what the mentor provided每个阶段都能讲清哪些是你做的、哪些来自导师
- The mentor's letter will corroborate independence with specifics导师推荐信会用具体事例佐证你的独立性
- You originated or visibly extended the question课题由你提出，或经你做出看得见的延伸
Entry Form 申请材料
- Academics at the 1550+ / 35+, 6+ AP-5, college-course bar — or resource context made clear so achievement is judged fairly学业达到 SAT 1550+/ACT 35+、6 门以上 AP 满分、大学课程的水位——或把资源背景写清，让成就被放在语境中评判
- At least one founded / led activity, plus specific Top-1%-style recommendations至少一项自己创办或主导的活动，以及具体到事例的「前 1%」式推荐信
Scientific Potential 科学潜力
- The layperson summary has a hook a journalist could quote外行摘要里有记者愿意引用的钩子
- The essay conveys a distinctive identity and genuine stake — not a résumé个人陈述传递出独特的身份与真实的利害——而非简历
- One quantified significance number repeated across paper, summary, and essays一个量化的关键数字贯穿论文、摘要与短文
One structural reality to hold onto: scoring carries real category and evaluator variance (the two ranking lenses correlate at only r ≈ 0.18), and structural advantages are real correlates. So maximize the things you control — rigor, independence, communication, honesty — and don't over-index on topic glamour.
最后记住一个结构性现实：评分带有真实的类别与评审员方差（两种排名镜头的相关性仅 r ≈ 0.18），结构性优势也是客观存在的相关项。所以，把力气花在你能控制的事上——严谨、独立、表达、诚实——不要过度迷信课题的光环。
Next: the tools that operationalize all of this — Skills →
下一步：把这一切落地的工具——技能助手 →