TIP

Real-world business data operations are often complex. Simple Instant APIs that write and update single tables can't handle complex business orchestration — especially financial calculations, multi-table consistency, and transaction guarantees. These require a 100% deterministic engineering approach to keep business processes reliable.

Make business actions write back reliably

This page uses "generate an order from a quotation" in the CRM to show when single-table writes are not enough. This action writes the order master, writes the order line items, and updates the quotation and the opportunity — all at once. To keep the data 100% accurate, the execution must either succeed entirely or fail entirely, so it needs to be wrapped in a Backend Function as a stable service.

The CRM requirement: generate an order from a quotation

Sales has already quoted a price and the customer has confirmed acceptance. When a rep runs "Generate Order" in the CRM, the system can't just insert one order record — the order amount, line items, quotation status, and opportunity stage all have to change together.

In this tutorial's CRM example, the requirement reads:

Generate an order from a confirmed quotation: create the order master, copy the quotation line items into order line items, update the quotation status to Converted, and sync the opportunity stage to Won or Awaiting Fulfillment; before executing, show which tables will be written; after executing, return the order ID, line-item count, order amount, and the update result for each table.

Business requirement	Action in the CRM	Why a single table isn't enough
Generate the order	Insert into the order master	An order master without line items can't support later fulfillment, shipping, or reconciliation
Keep the purchase details	Copy product, quantity, unit price, and discount from quotation items to order items	Missing or short line items make the amount and the actual purchase inaccurate
Prevent duplicate conversion	Update the quotation status to Converted	Without updating the status, the same quotation can be converted again
Sync the sales stage	Update the opportunity stage and the close time	Without syncing, the sales funnel, manager dashboards, and actual wins drift apart
Roll back on failure	No half-completed data survives a failed step	Customer, sales, and finance must all see the same business fact

Who this is for

Implementation engineers, administrators, and technical leads.

Map out what one action touches

In this tutorial's CRM example, a single "quotation to order" action touches at least these datasets and fields:

Dataset	Physical table	Key fields	Purpose
Quotation master	crm_quotation	id, customer_id, opportunity_id, status, total_amount	Validate that the quotation can convert, and update its status
Quotation items	crm_quotation_item	quotation_id, product_id, quantity, unit_price, discount	Read the line items to copy into the order
Order master	crm_order	id, customer_id, quotation_id, total_amount, status	Create the order record
Order items	crm_order_item	order_id, product_id, quantity, unit_price, discount	Write the order line items
Opportunities	crm_opportunity	id, stage, won_time, order_id	Sync the stage and close information
Customer Information	crm_customer	id, owner_id	Validate the customer and owner

If your own system involves tickets, contracts, or project settlement, list it the same way: which records this action creates, which statuses it updates, and which logs it writes.

Option 1: implement with rabetbase in Claude Code

Paste the prompt below into Claude Code. The prompt describes only the business requirement and delivery constraints — no rabetbase commands needed. Claude Code breaks the task down on its own and uses rabetbase to handle the app, datasets, Backend Function, dry-run, and push checks.

请帮我在当前 Lovrabet 应用中实现一个 Backend Function：createOrderFromQuotation。

业务场景：
我们以 CRM 为案例。销售已经给客户出过报价，客户确认接受报价后，需要把这张报价单生成正式订单。

需要实现的业务动作：
1. 基于一张已确认的报价单创建订单主表。
2. 把报价单明细复制为订单明细。
3. 把报价单状态更新为已转订单。
4. 同步销售机会阶段和成交信息。
5. 防止同一张报价单被重复生成订单。
6. 任一步失败时，不留下半成功数据。

实现要求：
- 先识别当前应用里的真实数据集、字段、必填项、枚举值和关联关系，不要猜字段名。
- 先检查是否已经存在同名函数或可复用的公共函数。
- 函数需要支持 preview。preview 为 true 时，只返回将要写入和更新的计划，不修改业务数据。
- 正式写入前必须做字段校验、状态校验、金额校验和重复提交校验。
- 写入过程要保证订单主表、订单明细、报价单状态、销售机会阶段的数据一致性。
- 返回结果要包含 orderId、detailCount、totalAmount、updatedTables、operator、executeTime。
- 推送到平台前，先给出 dry-run 预览和影响范围。
- 未经我确认，不要正式推送或执行真实写入。

请最后输出：
1. 你识别到的数据集、字段和关联关系。
2. createOrderFromQuotation 的入参和返回结构。
3. preview 模式会返回的写入计划。
4. 正式写入时的执行顺序。
5. 一致性、防重复提交和失败处理方式。
6. dry-run 结果摘要。
7. 需要我确认后才能继续的操作。

What rabetbase normally goes through

flowchart TD
  A[确认应用和认证] --> B[读取数据集清单]
  B --> C[读取字段和关联关系]
  C --> D[检查已有 BFF / COMMON]
  D --> E[创建或同步本地 BFF 脚本]
  E --> F[编辑 createOrderFromQuotation]
  F --> G[自检字段、事务、幂等、返回结构]
  G --> H[查看 bff status]
  H --> I[push --dry-run 预览]
  I --> J{用户确认}
  J -->|否| K[继续修改本地脚本]
  J -->|是| L[bff push --yes 推送]
  L --> M[回平台核对 Backend Function]
  M --> N[按需运行态验证]

Phase	What rabetbase does	What to check
Confirm the environment	Reads the command contract, auth status, and current app	rabetbase runs on your machine; the current app is correct
Read the data model	Pulls the dataset list, fields, operations, and relations	No guessed field names — real datasets are the source of truth
Check existing services	Reviews ENDPOINT and COMMON functions	Avoids creating a duplicate service
Generate the local script	Scaffolds the Backend Function locally	Function name, type, and script path are correct
Write the service logic	Implements preview, validation, multi-table writes, status sync, duplicate protection	A failed step returns a clear reason
Check local status	Shows whether the script is new, modified, or in sync	You know which function a push will affect
Dry-run preview	Lists only the functions to create or update; nothing is pushed	Confirm the push scope and risk points
Push	Syncs the script to the platform after you confirm	The service appears under Backend Function on the platform
Runtime verification	Calls the service from a page, a Skill, or the runtime CLI	Returns the order ID, line-item count, and update results

Option 2: do it manually in the platform

You can do the same thing in the platform without Claude Code. Manual steps suit a small number of hand adjustments; for complex write services, generate and check the script with rabetbase first.

1. Open App Management.
2. Click Dataset Overview and confirm that the quotation master, quotation items, order master, order items, opportunities, and customer information datasets are recognized.
3. Record each dataset's primary keys, foreign keys, required fields, enum values, and amount fields.
4. Click Backend Function.
5. Create or edit a write service. In the CRM example, name it createOrderFromQuotation.
6. Implement preview first: return the order master to create, the line-item count, the order amount, and the quotation and opportunity fields to update.
7. Execute the writes only after the user confirms.
8. When the writes finish, return the order ID, line-item count, order amount, datasets updated, and the failure reason or rollback result.

Screenshot reference: the Backend Function entry is shown below. Multi-table writes, transaction control, preview, and rollback logic are configured here.

Next steps

Continue to "Connect SQL, BFF, and Backend Functions to Skills".