2026-08-23. It started as an attempt to evaluate a direction: AI + E-commerce ERP startup. From the question “What systems do China’s largest e-commerce sellers actually use?”, it ended up as a software copyright application checklist. This post lays out the full research chain across four sections: market research, startup roadmap, platform API thresholds, and software copyright practical guide.

Part 1: ERP Market Research — Where the Money Is, Who’s Earning It

This section ran ~7.6 hours across 107 research agents. Every key finding survived adversarial verification from three independent perspectives; anything with 2/3 counter-arguments was discarded. The full report lives in ERP_RESEARCH_REPORT.md (50KB). Here I only include findings that survived scrutiny.

1.1 What Do China’s Top E-commerce Sellers Use?

The answer is more concentrated than you’d expect: Jushuitan (聚水潭), Wangdiantong (旺店通), and Jike Cloud (吉客云), segmented by order volume:

Jushuitan listed on the Hong Kong Stock Exchange in 2025, processing nearly 100 million orders daily. These aren’t marketing fluff — IPO filings carry legal liability.

1.2 Vendor Marketing Numbers Are Basically Unreliable

Adversarial verification knocked out several widely circulated figures:

• “KuaMai serves 4.5M merchants, 1.4M paying customers, 80M daily parcels” — 0/3 votes, pure vendor self-promotion, no third-party audit
• “Wangdiantong: 400K customers, 50K orders/sec during peak events” — 0/3 votes
• Various media outlets cite contradictory Jushuitan financials — only IPO-level data retained

Takeaway: Chinese web ERP market data is mostly vendor PR regurgitation. See any “XX万商家” claim and ask who audited it.

1.3 Open-Source ERP and Chinese E-commerce Exist in Parallel Universes

Odoo (Community, LGPL-3) and ERPNext (GPL-3.0) are mature open-source ERPs, but here’s the hard fact: neither natively supports Taobao, Tmall, JD.com, Pinduoduo, or 1688. The Odoo App Store’s e-commerce connectors only cover Amazon/Shopee/Lazada/TikTok.

This is precisely why profitable top sellers don’t use Odoo — not because open-source is bad, but because the gritty, relentless work of platform API integration has been done for ten years by domestic e-commerce ERPs, creating a moat. Traditional ERPs organize around finance and production; domestic e-commerce ERPs organize around order flow. The product DNA is fundamentally different from the start.

1.4 AI ERP: L3/L4 Don’t Exist (Yet)

Here’s the 2026 market through a maturity lens:

• L1 (chatbox added): Everywhere
• L2 (AI data querying): SAP Joule, Dynamics Copilot are working on it, but independent validated case studies are scarce
• L3 (AI analysis + API execution): No product has passed verification
• L4 (autonomous discover-decide-execute-validate loop): Nonexistent

Every “AI Agent operates your ERP” claim stops at L1/L2. This is a genuine gap — but honestly, a gap doesn’t equal opportunity. Here’s why below.

Part 2: Six Startup Routes — Why I Picked B

2.1 Six Routes at a Glance

2.2 Conclusion: B, but the platform mix got rewritten by reality

For a 2–5 person team, B is the only route satisfying “move fast, validate cheap”: zero licensing cost, find paying seed users after building just one module, and the downside of failure is controllable.

But the original “Taobao + Pinduoduo + 1688” starting combo had to be changed after checking each platform’s open API policies:

MVP = Pinduoduo + Douyin + 1688, with Taobao and JD.com coming later.

See Part 3 for the reasons.

Part 3: Platform API Thresholds — The Gatekeeper of the Entire Route

This section individually researched the latest 2025–2026 policies per platform. It consumed the most time and turned out to be the most valuable part.

3.1 Five-Platform Comparison

Three counter-intuitive findings:

1. The largest platform has the highest barrier. Intuitively you’d start with Taobao, but it’s precisely the platform individuals can’t enter at all.
2. “Start self-use, go tool-type later” is the only realistic path, and only works for Pinduoduo/Douyin/1688: first validate the full chain with your own store’s authorization, then apply for ISV marketplace entry once you have real cases. Tool-type applications face a chicken-and-egg problem — they may require evidence of already-listed apps and customer counts — so keep thorough records of real usage during the self-use period.
3. 1688 is critical on the procurement side. Its distribution/dropshipping APIs are free at the basic tier, but rate-limited. The throttling policy must be stress-tested in week one of development.

3.2 Confirming the Competitive Gap

On the open-source side: “Pinduoduo + 1688 full dropshipping pipeline” has no existing solution on ERPNext or Odoo. The strongest SaaS competitor, Dianxiaomi (店小秘), has quite complete free-tier features, but the dropshipping step still requires manual operation — automation depth and data ownership (open-source base enables self-hosting) represent the differentiation space.

Part 4: Software Copyright Application — the Only Tangible Thing Worth Pushing During the Freeze

Here’s why we end up here: Pinduoduo’s tool-type ISV application lists software copyright as a hard prerequisite, and the copyright approval cycle is 1–3 months — perfect for consuming time during the project freeze period. Below are the latest requirements as verified in 2026-08.

4.1 Checklist (4 Core Categories)

① Registration Application Form: Fill online at the Copyright Protection Center website. Software name must follow the standard format “XX System / XX Software”, version like V1.0, completion date must precede the publication date, and all fields must strictly match the code and manual.

② Software Description Manual: ≥15 pages (30 recommended), must include real running screenshots, clearly explain features and operational workflows with images and text, and include runtime environment documentation. For backend systems with no UI, supplement with API docs and processing flowcharts.

③ Source Code Document (highest rejection rate, accounting for 35% of corrections):

• First 30 pages + last 30 pages of continuous code, 60 pages total
• ≥50 lines per page (count excluding blank lines and comments)
• Total recommended: 6,000–10,000 lines, prioritize business-unique logic
• Header: software name + version number (matching the application form) + page number in the upper right
• Start with the opening statement, end with the closing statement
• No other party’s Copyright/Author notices allowed anywhere
• If open-source code is used: annotate the source + attach the original license + describe modifications

④ Identity Document: Individual = photocopy of ID card (both sides); Enterprise = business license copy stamped with official seal.

4.2 Process and Cost

Entirely online (paperless reform in June 2023 means only electronic certificates are issued):

1

Real-name verification (1–3 business days) → Online preliminary review → Formal substantive review (30–40 business days standard) → Electronic certificate

Cost truth: the official registration fee was abolished in 2017; DIY = ¥0. The few hundred to thousand yuan charged by agents is purely service fee. “Expedited” processing is an agent upsell, not an official channel. If time is tight, buying once is fine; otherwise just do it yourself.

4.3 2026 “Strictest Review Year” Red Lines

Rejection rates are up 40% year-over-year in 2026. Eight categories of behavior trigger immediate rejection + credit blacklist inclusion. The four most relevant to us:

1. Pure AI-generated content submitted as-is will be rejected — AI-assisted coding is fine, but there must be substantive human modification, and the modification records must be kept
2. Code similarity comparison system is live — it cross-references against the historical registration database + all public open-source projects; shell-jobbing open-source code will fail
3. Boilerplate manual text = 100% rejection — sentences like “the system is stable and easy to operate” are not allowed; every feature must map to a specific code module
4. Max 3 applications per day; don’t fragment to pad numbers — one genuine system, one application

The best护身符 is really just a good habit: git commit history, project initiation docs, test reports — these genuine R&D process artifacts are the bedrock that withstands any scrutiny.

4.4 Our Execution Plan

• Applicant: The independent integration service from Route B, named something like “XX E-commerce Order & Procurement Integration System V1.0”
• Timing: Organize materials immediately after developing a genuinely runnable feature; all screenshots come from real running interfaces
• Freeze period: Individual real-name verification + git commit trail only. Nothing else.

Appendix

• Full market report: ERP_RESEARCH_REPORT.md
• Route B landing plan: erpnext-route-b-plan-2026-08-23.md
• Software copyright checklist: ruanzhu-checklist-2026-08-23.md
• Methodology: Market research section was adversarially verified across multiple agents (3-vote system). Platform API and software copyright sections are first-party data verified online on 2026-08-23; please cite the date when referencing.