Scientify
Open workspace

2023 CUMCM Problem A · independent case study

AI mathematical modeling: Scientify completed Problem A at a national first-prize level

Inspect 131 real commands, executed code, the complete paper, and the original workspace.

Award-level finding

National first-prize level

Annual output

60.0022 MW

Output per area

0.45949 kW/m²

Official paper comparison

Scientify exceeds official showcase paper A0165 on both metrics

This is an optimization task that maximizes annual average output per unit mirror area. Higher values are better for both Problems 2 and 3.

Award-level inference

National first-prize level

ScientifyCase workspace

0.459063 kW/m²

0.459488 kW/m²

Baseline

Official showcase A0165Organizer showcase

0.4505 kW/m²

0.4556 kW/m²

Scientify leads on both

Official showcase A0127Organizer showcase

0.5871 kW/m²

0.5307 kW/m²

Official paper leads

Scientify exceeds official showcase paper A0165 on both metrics and delivers the model, code, validation, workbooks, and paper. This case reaches a national first-prize level.

Comparison values come from the annual result tables published with each paper.

REF 01Official showcase paper A0165Problem 2: 0.4505 · Problem 3: 0.4556 kW/m²REF 02Official showcase paper A0127Problem 2: 0.5871 · Problem 3: 0.5307 kW/m²

Autonomous execution

Scientify ran 131 commands in the cloud workspace, compared 15 candidate rounds, and wrote the results into code, workbooks, and the paper.

Subject
Solar-tower heliostat field
Input
Problem, attachment, result templates
Compute
Cosine, shading, interception, power
Output
Paper, code, JSON, CSV, and Excel
  1. 1Complete

    Read the problem

    Identify physical relationships, constraints, objectives, and delivery formats.

  2. 2Complete

    Build the model

    Implement solar position, optical efficiency, shading, interception, and power.

  3. 3Complete

    Search algorithms

    Compare density, twisted-lattice, ring, local-exchange, and marginal-yield designs.

  4. 4Complete

    Validate and deliver

    Run 60-time-point validation and generate the paper, workbooks, and code.

Scientify verifies results through recomputation and constraints

  • Final evaluation uses 60 time points rather than the coarse search model.
  • Every design checks field, exclusion-zone, spacing, and ground-clearance constraints.
  • A 0.01 reflectivity reduction lowers final power by 1.09%; the paper reports this sensitivity.
  • The ledger preserves rejected directions and the evidence behind each algorithm choice.

Auditable workspace

Inspect the session, code, experiment ledger, and final files

The evidence browser shows how Scientify proposed designs, ran calculations, rejected weak directions, and assembled the paper.

Translated from the original Chinese session. Commands, outputs and file names are shown unchanged.

01

report.pdf

Complete six-page modeling paper

Generated
02

iteration_ledger.md

Fifteen structural-search rounds

Generated
03

result2.xlsx

Official-format result workbook

Generated
04

solve.py

Physical model and recomputation code

Runnable

Pricing

New users get a free cloud computer and $5 worth of credits

Choose a plan by monthly credits. The cloud computer is free, and the same workspace keeps running across every plan.

View complete pricing

Starter

$19/month

60 credits

Researcher

$29/month

100 credits

Professional

$99/month

350 credits

At equivalent model usage, Scientify costs about 30% of standard API pricing. 1 credit equals $1 of usage at standard API token prices.

Keep the problem, data, code, logs, and results in one persistent workspace

Scientify runs on an isolated cloud computer and keeps research moving after your laptop closes.

Open the Scientify workspace