Programs · Mathematical Modeling · 数学建模

HiMCM, read closely and coached end to end.

The High School Mathematical Contest in Modeling (HiMCM), run by COMAP, gives teams a fortnight to turn an open, real-world question into a complete English paper. Tian2 coaches the whole arc — from decoding the prompt to a submission that passes every compliance gate — with the same standard the books and competition playbooks keep.

See a worked demo paper → Watch the assistant work

The Contest

COMAP · 2026 = 29th HiMCM

HiMCM is an open-ended modeling contest: there is no single correct answer, and the deliverable is a full English research paper. Everything below is sourced from COMAP's published rules for the 2026 season.

Format Open modeling. Real-world problem, no fixed answer; the deliverable is a complete English paper.

Teams 2–4 students, same school, one advisor. No cap on teams per school.

Window 14 days (336 h). Each team picks its own continuous working stretch. Entirely online.

Paper Single PDF ≤ 25 pages — summary, figures, and code appendix included. Fully anonymous; the summary page is decisive.

Awards By global percentile: Outstanding (~top 1%), Finalist (~7%), Meritorious (~15%), Honorable Mention (~30%), Successful Participant.

Entry fee ~$100 per team (international). Results by ~February 2027.

The 2026 Calendar

All times US Eastern · the EST deadline is authoritative

Nov 4 · 14:00 EST

Registration deadline (Nov 5, 03:00 Beijing).

Nov 4 · 15:01 EST

Problems released — the 14-day window opens.

Nov 17 · 20:00 EST

Problem window closes.

Nov 17 · 21:00 EST

Paper deadline. The system auto-closes — there is no grace period (Nov 18, 10:00 Beijing).

~Feb 2027

Results released.

Beijing runs 13–14 hours ahead of US Eastern depending on daylight saving. The deadline is governed by EST; coaching plans are anchored to it, not to the local clock.

How We Coach

Seven stages · 建模辅导七阶

The work is a resumable, gated pipeline — the modeling loop (understand, assume, model, solve, validate, write) carried stage by stage so nothing is hand-waved and every figure traces back to real, checked computation. The insight stays the students'; the structure is ours.

Decode

Restate the problem, name the sub-questions, pin down the key terms.

Strategy

Choose models and justify them from first principles — not by reflex.

Data & Code

Acquire real data, write and run the model, produce honest figures.

Sensitivity

Parameter sweeps and named stress scenarios — the judges' differentiator.

Draft

The full paper, with embedded figures and a clear results narrative.

Summary

A one-page, numeric summary written for the panel — the decisive page.

Submit-check

Anonymity scan, ≤25 pages, AI-Use Declaration, single final PDF.

What the judges reward

From COMAP's published weights

Modeling25%

Creativity & completeness15%

Sensitivity analysis10%

Solution process & writingalso

Screening advances roughly the top half to a preliminary round, then to detailed and final judging, where Outstanding and Finalist teams may defend. A model that is built well but never stress-tested leaves the sensitivity points — and the credibility they signal — on the table.

A Worked Demo Paper

Proof of standard · 示范论文

2021 HiMCM Problem B · Award-standard · Real USBR data

Tackling the Drought — Forecasting Lake Mead & Sizing Water Recycling

A complete 22-page paper produced end to end at the award-winning standard: a coupled pipeline from reservoir level to storage to budget to recycling, validated against real U.S. Bureau of Reclamation data, with a Monte-Carlo coverage analysis and a named decision flip under stress.

Open the demo →

22 ppwithin the 25-page limit

17 figuresevery one from checked computation

0.19%volume curve validated vs. official tables

7 checksdeterministic compliance — all pass

Watch the Assistant Work

A real run · 2025 HiMCM Problem A

The coaching pipeline is a working application, not a slide deck. Below is unretouched output from an actual run — the assistant was handed 2025 HiMCM Problem A (Emergency Evacuation Sweeps), and across the seven stages it acquired the layout, wrote and executed Python in a sandbox, produced these figures, ran a parameter sweep and named stress scenarios, assembled a .docx paper, and then audited its own submission. Every number below traces back to code that ran.

School floor plan with three colour-coded responder sweep routes down a nine-room corridor; total sweep time 117 seconds. — Stage 3 · Data & Code The model built the building as a weighted graph and solved the sweep as a makespan problem: three responders, nearest-neighbour routing from the entrance, **117 s** to clear a nine-room floor. Rendered by the Python the assistant ran, not drawn by hand.

Line chart of total sweep time against responder count, showing an elbow at three responders at 1.95 minutes. — Stage 3 · Result Sweep time versus crew size, with the diminishing-returns **elbow at three responders** (1.95 min) and the travel-overhead gap against a perfect-parallel lower bound.

Tornado chart of parameter sensitivity beside a bar chart comparing baseline, stress and extreme scenarios. — Stage 4 · Sensitivity The differentiator: a tornado chart ranking each input (walk speed dominates) and three **named scenarios** — Baseline 2.21 min → Stress 2.84 → Extreme 3.80 — so the recommendation is stress-tested, not asserted.

Stage 7 · Submit-Check — the assistant audits its own paper Verdict: not yet submittable — 3 blockers

Pass Anonymity. No student, school, or advisor name in the body, tables, headers, or metadata; author field is a safe placeholder.

Pass Length & structure. ~12 pages (≤ 25); Summary → TOC → Introduction → … → References → Appendix, matching the Outstanding-paper order; code confined to the appendix.

Pass Figures & tables. All three figures and four tables captioned and referenced in text; no orphans.

Blocker Summary is a placeholder. The Q1–Q4 numbers live in the Conclusions but were never written into the Summary judges read first — flagged, with the exact numbers to paste in.

Blocker AI-Use Declaration missing. Required as a separate document — the checker drafts a truthful one for the team to confirm and attach.

Flag Baseline inconsistency. Results cite 1.95 min; the sensitivity table cites 2.21 min. The checker caught the 0.26 min discrepancy and named the two places to reconcile.

This is the point: a good coach does not rubber-stamp. The assistant flagged its own draft as not-yet-submittable and told the team exactly what to fix — the modeling, the missing Summary, and the honest AI-Use Declaration all stay the students' responsibility. The full machine-generated report ships with the run (read the raw Submit-Check report).

On AI Use

COMAP policy, since 2023

※ Compliance

COMAP permits AI for grammar, translation, and coding assistance, but forbids it as the core modeling idea or as a ghost-writer for the body of the paper. An AI Use Declaration is mandatory, and any AI-generated text counts toward the 25-page limit. Tian2's coaching is built for drafting, structure, and quality assurance — the modeling insight must be, and stays, the students' own.