light-competition

Competition and Project Application Materials

First identify the competition/project type (first step of the process, must produce the rules_checklist artifact)

The review priorities of different competitions vary greatly. First confirm which competition/group it is, andalign with the current edition's official rules— rules change year by year. First download the current edition's review rules package, check off each item according to the references/competition_rules_cn.md rules_checklist,save to disk material_checklist.md before starting to write(verbal confirmation does not count; using an old edition's group names/missing the current edition's key areas are high-frequency elimination points anti_patterns D2/E1). One-sentence positioning for each competition:

• Mathematical Modeling National Competition CUMCM: The abstract (a separate page, determines initial screening) is most important; must include sensitivity analysis + reproducible code in the appendix; there are rules on AI usage statements.
• Mathematical Modeling MCM/ICM: English Summary Sheet on the first page, extremely high weight;hard 25-page limit for the entire paper(including appendix code; only the AI report does not count toward pages); 99-hour three-person team.
• Statistical Modeling Competition: Emphasizes real data + rigorous statistical inference > algorithmic showmanship.
• China International College Students' Innovation Competition (formerly Internet+): Undergraduate/graduate group × Creative Group/Entrepreneurship Group (the old "Startup/Growth Group" has been eliminated); submit BP + roadshow PPT + defense.
• Challenge Cup: First distinguish whether it is the "Big Challenge" (extracurricular academic, emphasizes research depth) or the "Small Challenge" (entrepreneurship plan, emphasizes commercial viability).
• College Students' Innovation and Entrepreneurship Training Program/research training: Three categories: innovation/entrepreneurship training/entrepreneurship practice; emphasizes student-centeredness and educational value; follows application → mid-term inspection → final acceptance.

Group subdivisions, page limits, current edition's key areas, scoring dimensions, AI rules, etc. for the six major competitionshard rule details that change easily year by yearare all pushed down to references/competition_rules_cn.md(including rules_checklist template and last_checked); for the backend high-scoring skeleton and review dimensions, see databases/db08-ip-materials/case_skeletons.md。

Material modules (as needed)

Application form, project plan, business plan, roadshow PPT (submit m16), defense script, project abstract, technical roadmap, innovation points, feasibility analysis, market analysis (scale/competitors/pain points/barriers), research foundation, expected outcomes, budget estimate (reasonable categorization), team division of labor.

General application form skeleton(isomorphic for College Students' Innovation and Entrepreneurship Training Program/Challenge Cup/research training/funds, and can be borrowed across them): project basis and significance → domestic and international status → research content and objectives → technical roadmap/plan → feasibility analysis → innovation points → schedule (milestones) → expected outcomes → budget estimate → team division of labor and foundation. Aligned with the NSF/NIH paradigm, this is Specific Aims→Significance→Innovation→Approach→Preliminary Data→Timeline→Budget→Broader Impacts. Key hard points borrowed from the Research Grants skill:

• Specific Aims one-page structure: Knowledge gap opening → long-term + near-term goals → central hypothesis → 2–4 aims (each containing rationale/working hypothesis/method summary/expected output, independent yet complementary, completable within the time budget) → closing payoff paragraph.
• Write Innovation across five dimensions: concept/method/integration/translation/scale, and explain how to overcome the limitations of existing solutions.
• Approach: pair each aim with: design · sample size/data volume · analysis plan · quality control · contingency plan。
• Overall persuasion line: Hook→Problem→Solution→Evidence→Impact→Team, ensuring internal consistency among budget↔activities↔timeline↔personnel.

Before the business plan, first fill in the Lean Canvas (nine boxes) to self-check the logic loop: Problem (top 3 pain points)·Customer Segments·Unique Value Proposition·Solution·Channels·Revenue Streams·Cost Structure·Key Metrics·Unfair Advantage (hard-to-replicate barrier). Note that "first-mover/team/resources" are often mistaken for barriers; UVP and Unfair Advantage are the most critical and hardest to fill in.

Core pages first(borrowing NIH Specific Aims, adapted for CN): judges have limited time,the first page/opening 300–500 wordsdetermine whether they read through with interest. Write in the following order:

• Hook (specific numerical hook) → Gap (why existing solutions are insufficient) → long-term goal/project goal → central hypothesis (entrepreneurship-oriented = UVP in one sentence).
• Rationale: move one piece of preliminary evidence forward to reduce perceived risk.
• 2–4 sub-goals: each containing rationale/hypothesis/method/output, mutually independent and completable within the cycle.
• Payoff: value closing.

The "project rationale + research content and goals + innovation points" of the College Students' Innovation and Entrepreneurship Training Program/Challenge Cup, the 30-second opening of an Internet+ roadshow, and the abstract of a mathematical modeling competition are all de facto core pages. For the complete eight-paragraph skeleton + paragraph-by-paragraph Chinese examples + self-check list, see assets/aims_zh_guide.md。

Writing key points

• Innovation points: differentiate from existing solutions, quantifiable and demonstrable (reuse the judgments from m03/m04).
• Feasibility: technology (has a foundation), resources (compute/data/funding), time (milestones), team (division of labor matches).
• Progress/technical route(borrowing the writing-plans approach): break phases into verifiable small steps, each with deliverables, avoiding vague "phase one investigation". Milestone JSON→Gantt chart (phase bars + deliverables + go/no-go diamond nodes)/technical roadmap (phase blocks + arrow flow) use scripts/roadmap_gen.py(run synthetic self-test without parameters,--emit-sample export sample JSON; matplotlib automatically attaches Chinese fonts; the chart is a reproducible artifact, delete after use).System architecture diagram/business model flowchart/technical solution block diagram(reviewers often revise repeatedly) can use Draw.io MCP(diagram-as-code，.drawio XML editable and version-controllable, export vector), more suitable for structural diagrams than Gantt scripts (for configuration see the recommended MCP table in README).
• Judge's perspective: highlight bright spots, social/economic/academic value, closed logic loop, data support.
• Market analysis(entrepreneurship-oriented, borrowing the Market Research Reports skill and the 7-step method):
  • Three-tier estimation：TAM→SAM→SOM，put each layer's assumptions into an "assumption register" so judges can directly challenge the inputs, with sensitivity analysis; if no ready-made report is available, use first principles (population base × conversion filtering × realistic ARPU), then use competitors' revenue/public financial reportsCross-validation, to eliminate top-down inflation.
  • Competitor four-lens: Porter's Five Forces rated H/M/L per force + rationale, Top share, positioning 2×2 matrix, strategic groups; 5–7 companies on a unified template, distilling table stakes/common gaps/underserved segments.
  • Data discipline: must cite sources and be ≤2 years old; customer profiles based on real language from reviews and forums rather than conjecture; use real, verifiable data, do not fabricate market size.
  • Market data charts(TAM/SAM/SOM layered concentric circles, competitor 2×2 positioning matrix, Porter's Five Forces grading, risk probability×impact heatmap) using scripts/market_charts.py(run synthetic self-test without parameters,--emit-sample export sample JSON; its TAM/SAM/SOM three values share the same set of numbers as the db08 budget/financial forecast, with a07 consistency gatekeeping; matplotlib automatically applies Chinese fonts and colorblind-safe semantic colors; charts are reproducible artifacts, deleted once materials are done).

Roadshow PPT and defense

• Roadshow deck follows the m16 (light-slides) programmatic route: python-pptx produces editable pptx, unified visuals per the db06 theme, suitable for startup/Challenge Cup roadshow venues; data charts go through m11 real-data charting then embedded, no generative models to create charts. When you needfaster production of design-forward roadshow materials/posters, you can use Canva MCP(natural language generation + brand template batch filling, export PPTX/PDF/PNG; core features free, autofill/brand templates require Enterprise) for startup/Challenge Cup on-site decks and posters—data charts still go through m11 real-data charting then embedded, no AI-generated data charts.
• Academic/defense slides use the assertion-evidence method (Michael Alley): each page title states a complete concluding sentence, body uses charts/diagrams as evidence, avoid piling up bullets; one page, one message. Set upper and lower bounds for content density (borrowing beamer-skill: ≤7 bullets/2 formulas/5 symbols per page, and every page must be worth existing); Why before What. For formula-dense scenarios, Beamer can be used (block/alertblock for key points), but use overlays cautiously for handouts/printing—beamer-skill simply abandons \pause/\onslide, switching to multiple pages + colorblind-safe semantic colors (contrast ≥4.5:1) for progressive emphasis.
• After the PPT draft is produced, run the visual QA loop(borrowing PPTX skill / beamer-skill): convert to PDF→convert page by page to images (pdftoppm -jpeg -r 150)→use "fresh eyes" to check overlap/contrast/overflow/residual placeholders (grep for xxxx/lorem/unreplaced text)→fix→re-verify only affected pages→until no new issues. Beamer swallows overflow warnings inside blocks, so you must rely on visual audit to catch them.Separate review from revision: first only read out a structured report of "location/current state/suggestion/severity", then make changes in a separate step; you can add a devil's-advocate adversarial round specifically hunting for design flaws.

Review simulation

After the draft is produced, self-review from the judges' perspective (analogous to m14): is the innovation sufficient, is it feasible, does the business logic hold, is the budget reasonable, what might be asked in the defense.

1. First use references/anti_patterns.md five categories of anti-patterns to sweep for pitfalls item by item(concept/writing/technical/format/strategy, each with symptom→why points are deducted→how to fix + a one-page quick-reference table).
2. Then use scripts/scorecard.py to produce a machine-checkable self-review scorecard: fill in a per-dimension self-assessment JSON (each dimension 1–10), weight by the chosen competition's (innovation/dachuang/dating/mcm) review dimensions, output total score + weak dimensions + red flags against anti_patterns + release decision (--emit-sample check format,--list see dimension). ⚠ Weights areempirical relative references, unofficial scores(the exact weights for Chinese competitions are mostly undisclosed), used only for ranking "which dimension is holding things back", subject to the current edition's Review Rules and may pass weights_override override.
3. Competition-specific review: For grant/academic categories, go item by item through the NIH five criteria (Significance/Investigator/Innovation/Approach/Environment) or the NSF two criteria (Intellectual Merit + Broader Impacts); for slides, use the beamer-skill deduction system + devil's-advocate adversarial rounds; for mathematical modeling, review the abstract/sensitivity, MCM 25 pages and the AI statement (details in references/competition_rules_cn.md); for entrepreneurship categories, review whether the UVP/barriers/market sizing can withstand questioning.
4. For multiple selling points, use content-strategy weighted scoring (customer impact 40% + fit 30% + potential 20% + resources 10%) to rank them; in the pitch, present the strongest one or two first (anti_patterns E2).
5. Finally do a templates/defense_qa.md defense QA rehearsal, and prepare honest fallback talking points for weak spots.

Linkage with other outputs

Technical content ← m05/a03/a04; paper output ← m07; software copyright/patent ← m15; PPT ← m16; charts ← m11; consistency is ensured by a07 (the same project's statements are unified across materials).

Outputs

Artifact names to write to disk(CONVENTIONS §6.1 single source of truth; the downstream orchestrator dispatches by name, a07 cross-checks across materials, a08 self-checks): use according to the selected material module competition/application_draft.md(application form) ∣ business_plan.md(business plan) ∣ pitch_deck_outline.md(pitch outline, handed to m16 to produce pptx) ∣ defense_qa.md(defense QA rehearsal) ∣ budget_table.md(budget table), and always produce material_checklist.md(materials checklist verification + current edition rules confirmation). Fillable templates are in templates/(application form eight-section skeleton/business plan/defense QA rehearsal/project abstract); for the end-to-end walkthrough see examples/worked_example.md。

Supporting guides and assets: The application core page guide (NIH Specific Aims → CN Grand Innovation Program/Challenge Cup core page, eight-section skeleton + Chinese examples + self-check) is in assets/aims_zh_guide.md; the cross-competition anti-pattern handbook (five categories, symptom → why points were deducted → how to fix) is in references/anti_patterns.md; the hard rules and confirmation checklist for each edition of the competitions are in references/competition_rules_cn.md; the judge-perspective self-review scorecard (weighted by dimension + red flags for weak items, weights are empirical, not official scores) is in scripts/scorecard.py; the progress chart generator (milestone JSON → Gantt chart/technical roadmap, supports --granularity week|month) is in scripts/roadmap_gen.py; the market data chart generator (TAM/SAM/SOM concentric circles/competitor 2×2/Porter's Five Forces/risk heatmap) is in scripts/market_charts.py(JSON and the budget/financial forecast share the same TAM/SAM/SOM).

Backend knowledge base (db08-ip-materials): The budget table templates (research expenditure budget/entrepreneurship financial forecast + assumptions register + self-review checklist) are in databases/db08-ip-materials/budget_template.md; the high-scoring structural skeletons for each competition (Internet+/Challenge Cup Grand Challenge/Grand Innovation Program/mathematical modeling + review dimensions + common elimination points) are in databases/db08-ip-materials/case_skeletons.md; the subcategory cards for competition/software copyright/patent materials are in databases/db08-ip-materials/material_extended_cards.md(material_cards.md is the template and canonical index shell; the actual cards have been moved to extended). For per-tool research notes (real endpoints/methods/links), see references.md.

Handoff

The whole process is recorded in db09; business/data authenticity risks are reported to a10 (do not exaggerate, do not fabricate). Before delivery, pass the a08 (light-self-review) self-check gate.