田二
理工科升学规划
Admissions advising read the way an editor reads a manuscript — for the argument underneath. We build school lists from the Common Data Set, not the brochure; rank universities with an ensemble of twelve global systems, not one; and plan the summers, essays, and timeline that get a STEM applicant from a first question to an offer.
Most admissions advice is marketing in a cardigan — a confident voice, a brochure number, a list of "safeties" assembled from rumor. We work the other way. Every school a student considers begins as a row in its Common Data Set: the standardized, audited filing every US college submits in the same format. When a school's glossy materials and its CDS disagree, the CDS is the one we believe.
Reading that data honestly changes the list. A published "1500–1570" band at a test-optional school is not the band of the admitted class — it is the band of the strongest quarter who chose to submit. We correct for that self-selection, we know which schools have quietly returned to test-required (where the numbers are finally honest again), and we know that for most strong STEM applicants from a Chinese curriculum, a 780+ Math is a baseline, not a distinguisher. The work is to find the schools where a student's real strengths are a positive signal rather than a coin flip.
The same humility runs through how we rank schools. No single ranking is correct — each measures something real and is biased by its own method — so we aggregate twelve of them and report not just a consensus but the disagreement. A university ranked between #5 and #15 everywhere is a surer thing than one that swings from #2 to #80. We would rather hand a family an honest range than a falsely precise integer.
And we tell students what we do not know. The numbers say what is competitive; the application says whether a reader finds the student interesting. The most under-leveraged advantage a STEM applicant has is rarely another fifty points — it is a distinct context, a real project, a question they actually chased. Nothing is hurried here; the long way is always shown.
We start from each school's Common Data Set — admit rate, score bands, the admission-factor matrix, financial-aid reality — and translate it into what it actually means for one student's profile. The Yale walkthrough on our Data & Insights page is the method, shown in full.
A reach / match / safety list corrected for test-policy bias and TOEFL gating, weighted toward programs where STEM strength is read as a positive — and informed by our own meta-ranking, not US News alone.
From a curated database of elite STEM summer programs — Harvard SSP, MIT MITES & Beaver Works, Stanford, UPenn ESAP, JHU EEI, CMU SAMS, and more — matched to grade, deadline, language requirement, and intent.
Working only from each school's official supplemental prompts, we coach the argument and the voice — never a template, never a translation-tool cadence that readers discount on sight.
A junior-to-senior-year timeline — testing windows, early vs. regular deadlines, CSS Profile, the F-1 interview — so nothing that matters is discovered a week too late.
Our differentiator is original, statistical analysis of the 2024–25 Common Data Set universe — 135 institutions, read line by line. Four findings shape almost every list we build.
Test-required schools see a submission rate ~1.7× higher than test-optional ones (57% vs. 34%). A test-optional school's published band reflects only the strong third who chose to submit — so the true admit floor sits 30–60 points below the printed 25th percentile.
At STEM-leaning schools the Math 75th percentile runs 40–50 points above EBRW. For applicants from a Chinese math curriculum who score 780–800 reliably, that asymmetry is a structural advantage — strongest where math is a signal, not a saturated baseline.
Despite the folklore of "you need a 100," the median hard TOEFL minimum across 100 schools is 80. The binding constraint is real but lower and more school-specific than most families assume — we map it tier by tier.
A section-by-section walkthrough of a Common Data Set, using Yale as the worked example — admit math, the C7 factor matrix, the C9 score profile, and the financial-aid section that quietly matters most.
Every major ranking measures something real and is biased by its own method: ARWU rewards elite research output, QS leans ~45% on opinion surveys, Leiden is rigorous but research-only. So instead of trusting one, we aggregate up to twelve global systems into a single transparent, reproducible consensus — the same way an ensemble model beats any single estimator.
At the very top the order is robust to any reasonable weighting — the genuinely elite lead every system. Weighting refines the contested middle; it does not manufacture the order.
Two views are published. Tian2 gives every system one vote. Tian2-Weighted weights each system by its real-world acceptance, scored on four axes:
Public & student awareness, audience size.
Methodological rigor & trust among academics.
Use by governments, universities & employers.
Track record, first-mover status, stability.
Built on 100% public-data inputs with open, reproducible code. The detailed per-system weighting rubric is part of Tian2's internal methodology and is summarized, not published, here.
A curated, verified directory of roughly fifty elite STEM summer programs — drawn from a database of 412 programs and re-checked for current dates, deadlines, grade eligibility, and language requirements. A sample of the flagship engineering and science offerings:
| School | Program | Deadline | Focus |
|---|---|---|---|
| Harvard | Secondary School Program (SSP) | Feb 11 | Engineering science, CS & math for credit |
| MIT | MITES Summer · Beaver Works (BWSI) | Feb 1 · Mar 31 | Intensive research; robotics, AI, aerospace |
| Yale | YYGS — Innovations in Science & Tech | Jan 14 | Interdisciplinary STEM |
| Stanford | Pre-Collegiate · Biomedical Engineering | Mar 13 | Medical technology & neural interfaces |
| UPenn | Engineering Summer Academy (ESAP) | Mar 1 | Biotech, nanotech, computer graphics |
| JHU | Explore Engineering Innovation (EEI) | Mar 14 | First-year engineering, simulated |
| CMU | Summer Academy for Math & Science (SAMS) | March | STEM project learning & research |
| Princeton | Algorithmic Thinking (PACT) | Rolling | Theoretical CS & algorithms |
We coach essays the way we coach a derivation: find the argument, then set it down cleanly in the student's own voice. We maintain a reference of the official supplemental prompts across 131 institutions — word limits, prompt counts, the question behind the question — and work strictly from those.
We do not hand a student a model essay to imitate, and we do not republish anyone else's article corpus. Admissions readers see thousands of template-cadenced, translation-tool essays a year and discount them on sight — so we never make one.
Outcomes are presented as anonymized archetypes — field, method, and venue only. We publish no names, no scores, and no application records.
A visual results gallery is in preparation. Until then, outcomes are described in text only, fully anonymized.
The numbers tell you what is competitive. The application tells a reader whether you are interesting.
— The Tian2 EditorsA free editorial showcase. We share method and analysis openly; there is no storefront here.