sap-rpt1

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

SAP-RPT-1-OSS FI/CO Predictor Skill

SAP-RPT-1-OSS FI/CO 预测器技能

Use SAP-RPT-1-OSS for local, source-reviewed tabular prediction experiments on FI/CO CSV extracts. Keep this skill scoped to prototype and research workflows; do not present predictions as production finance, credit, audit, payment, or compliance decisions.

使用SAP-RPT-1-OSS对FI/CO CSV提取文件进行本地、经源审核的表格预测实验。本技能仅限用于原型和研究工作流；不得将预测结果用于生产环境中的财务、信贷、审计、支付或合规决策。

When to Use This Skill

何时使用本技能

Use this skill when preparing or reviewing SAP finance tabular data for local SAP-RPT-1-OSS experiments, especially:

FI-AR payment default or late-payment risk classification/regression.
FI-AP cash discount leakage or payment timing risk prediction.
FI-GL journal anomaly or unusual posting review prioritization.
Time-safe feature preparation, target leakage checks, and governance review for FI/CO CSV datasets.
Local SAP-RPT-1-OSS setup planning where Hugging Face access, Python 3.11, and GPU limits need to be explained before execution.

Do not use this skill as a hosted SAP-RPT API integration guide, production scoring service, SAP AI Core deployment guide, or live SAP extraction workflow.

在为本地SAP-RPT-1-OSS实验准备或审核SAP财务表格数据时使用本技能，尤其适用于以下场景：

FI-AR（应收账款）付款违约或逾期付款风险的分类/回归分析。
FI-AP（应付账款）现金折扣损失或付款时机风险预测。
FI-GL（总账）日记账异常或异常过账的审核优先级排序。
FI/CO CSV数据集的时间安全特征准备、目标数据泄露检查和治理审核。
在执行前需要说明Hugging Face访问权限、Python 3.11环境和GPU限制的本地SAP-RPT-1-OSS设置规划。

请勿将本技能用作托管SAP-RPT API集成指南、生产评分服务、SAP AI Core部署指南或实时SAP提取工作流。

Quick Start

快速开始

Confirm the task is a local prototype using synthetic or approved masked FI/CO CSV data.
Read
```
references/data-governance.md
```
before touching real finance data.
Choose one prediction point and one target column. Define what was known at that point in time.
Select the Python executable from an approved Python 3.11 environment. On Windows this might be
```
py -3.11
```
or
```
.venv\Scripts\python.exe
```
; on macOS/Linux this might be
```
python3.11
```
or
```
.venv/bin/python
```
.

Use

scripts/fico_data_prep.py --dry-run --input "<file.csv>" --target <column>

to inspect schema and leakage risks.

Use

scripts/rpt1_oss_predict.py --dry-run --input "<file.csv>" --target <column>

to review local SAP-RPT-1-OSS prerequisites.

Add
```
--encoding <encoding>
```
or
```
--delimiter ';'
```
when enterprise CSV exports are not UTF-8 comma-delimited.
Run local inference only after explicit user request, Hugging Face access, license review, and an explicit output path or stdout mode.

确认任务为使用合成或已批准的掩码FI/CO CSV数据进行的本地原型实验。
在处理真实财务数据前，阅读
```
references/data-governance.md
```
。
选择一个预测点和一个目标列，定义该时间点已知的信息。
从已批准的Python 3.11环境中选择Python可执行文件。在Windows系统中可能是
```
py -3.11
```
或
```
.venv\Scripts\python.exe
```
；在macOS/Linux系统中可能是
```
python3.11
```
或
```
.venv/bin/python
```
。

使用

scripts/fico_data_prep.py --dry-run --input "<file.csv>" --target <column>

检查数据 schema 和数据泄露风险。

使用

scripts/rpt1_oss_predict.py --dry-run --input "<file.csv>" --target <column>

检查本地SAP-RPT-1-OSS的前置条件。

当企业CSV导出文件不是UTF-8逗号分隔格式时，添加
```
--encoding <encoding>
```
或
```
--delimiter ';'
```
参数。
仅在获得明确的用户请求、Hugging Face访问权限、许可证审核通过，并指定明确的输出路径或标准输出模式后，再运行本地推理。

FI/CO Use-Case Matrix

FI/CO 用例矩阵

Detailed v1 recipes live in

references/fico-use-cases.md

Area	V1 depth	Typical target	First reference
FI-AR	Detailed	`paid_late` , `days_late` , `default_flag`	Payment default / late payment
FI-AP	Detailed	`discount_lost` , `discount_amount_lost`	Cash discount leakage
FI-GL	Detailed	`manual_review_flag` , `reversal_flag` , `outlier_flag`	Journal anomaly
CO/PS/CO-PA/FI	Starter only	Overrun, margin, cash, dispute, credit risk	Use-case matrix rows only

详细的v1方案请查看

references/fico-use-cases.md

。

领域	V1 深度	典型目标	首个参考场景
FI-AR	详细	`paid_late` , `days_late` , `default_flag`	付款违约/逾期付款
FI-AP	详细	`discount_lost` , `discount_amount_lost`	现金折扣损失
FI-GL	详细	`manual_review_flag` , `reversal_flag` , `outlier_flag`	日记账异常
CO/PS/CO-PA/FI	仅入门级	超支、利润、现金流、争议、信贷风险	仅用例矩阵行

Data Preparation Patterns

数据准备模式

Prefer source extracts already flattened to one row per prediction object, such as one invoice, one vendor invoice, one journal line, or one journal document. Rename technical SAP fields into semantic column names before inference; SAP-RPT-1-OSS uses column names and values as part of the tabular context.

Use S/4HANA starting points such as ACDOCA plus relevant master data and process extracts where available. Use ECC fallback tables only as starting points, not universal truth. Never assume table availability, field semantics, or release behavior without checking the target system.

优先使用已扁平化为每个预测对象对应一行的源提取数据，例如一张发票、一张供应商发票、一条日记账分录或一份日记账凭证。在推理前将SAP技术字段重命名为语义列名；SAP-RPT-1-OSS会将列名和值作为表格上下文的一部分使用。

在可用情况下，使用S/4HANA的起始点，如ACDOCA加上相关主数据和流程提取数据。仅将ECC fallback表作为起始点，而非通用标准。在未检查目标系统的情况下，切勿假设表的可用性、字段语义或版本行为。

Target Leakage Rules

目标数据泄露规则

Define the as-of date before selecting features. Exclude fields created or updated after that prediction point.

Common leakage examples:

Clearing date, clearing document, payment run result, or final payment status when predicting before payment.
Dunning, dispute, collection, write-off, or audit outcomes created after the prediction point.
Reversal or investigation flags created after a journal posting.
Actuals, settlements, or period-close adjustments posted after a forecast date.

在选择特征前定义截止日期，排除在该预测点之后创建或更新的字段。

常见的数据泄露示例：

在预测付款情况时，使用清账日期、清账凭证、付款运行结果或最终付款状态。
在预测点之后生成的催款、争议、催收、核销或审计结果。
日记账过账后创建的冲销或调查标记。
在预测日期之后过账的实际数据、结算或期间调整。

Governance Checklist

治理检查清单

Before using real FI/CO data, confirm business owner approval, legal/compliance approval, field minimization, masking of personal and bank-related data, time-based validation splits, documented target definitions, and human review. Use

references/data-governance.md

as the minimum checklist.

Do not use predictions as the sole basis for payment blocking, credit decisions, collections action, write-offs, audit conclusions, or control sign-off.

For Windows, macOS, Linux, and managed non-admin workstations, use

references/enterprise-portability.md

before suggesting setup or execution commands. Prefer user-writable virtual environments, approved cache locations, quoted paths, and explicit output locations.

在使用真实FI/CO数据前，确认已获得业务所有者批准、法律/合规部门批准、字段最小化处理、个人和银行相关数据的掩码处理、基于时间的验证拆分、文档化的目标定义以及人工审核。将

references/data-governance.md

作为最低检查清单。

不得将预测结果作为付款冻结、信贷决策、催收行动、核销、审计结论或控制签字的唯一依据。

对于Windows、macOS、Linux和受管非管理员工作站，在建议设置或执行命令前，请查看

references/enterprise-portability.md

。优先使用用户可写的虚拟环境、已批准的缓存位置、带引号的路径和明确的输出位置。

Bundled Resources

捆绑资源

```
references/source-review-2026-06-18.md
```
: public source/model/product-page review and open upstream issue list.
```
references/fico-use-cases.md
```
: detailed FI-AR, FI-AP, and FI-GL recipes plus starter matrix rows.
```
references/data-governance.md
```
: minimum data governance and model-card checklist.
```
references/enterprise-portability.md
```
: Windows, macOS, Linux, non-admin, proxy/cache, and CSV export guidance.
```
scripts/fico_data_prep.py
```
: read-only CSV schema, target, and leakage inspection helper.
```
scripts/rpt1_oss_predict.py
```
: opt-in local inference wrapper for SAP-RPT-1-OSS.
```
assets/*.csv
```
: synthetic FI/CO sample datasets only.

```
references/source-review-2026-06-18.md
```
：公开源/模型/产品页面审核以及上游公开问题列表。
```
references/fico-use-cases.md
```
：详细的FI-AR、FI-AP和FI-GL方案以及入门级矩阵行。
```
references/data-governance.md
```
：最低数据治理和模型卡片检查清单。
```
references/enterprise-portability.md
```
：Windows、macOS、Linux、非管理员、代理/缓存和CSV导出指南。
```
scripts/fico_data_prep.py
```
：只读CSV schema、目标和数据泄露检查辅助工具。
```
scripts/rpt1_oss_predict.py
```
：SAP-RPT-1-OSS的可选本地推理封装器。
```
assets/*.csv
```
：仅包含合成FI/CO样本数据集。

Known Issues

已知问题

Treat upstream SAP-RPT-1-OSS issues as source-reviewed limitations, not fixed behavior. See

references/source-review-2026-06-18.md

for issue numbers and titles reviewed from

SAP-samples/sap-rpt-1-oss

Local inference can download gated model artifacts, populate local caches, require Hugging Face authentication, and require substantial memory. Commands bundled with this plugin must not run inference directly.

将上游SAP-RPT-1-OSS问题视为经源审核的限制，而非已修复的行为。请查看

references/source-review-2026-06-18.md

，获取从

SAP-samples/sap-rpt-1-oss

中审核的问题编号和标题。

本地推理可能会下载 gated 模型工件、填充本地缓存、需要Hugging Face身份验证，并且需要大量内存。本插件捆绑的命令不得直接运行推理。

Source and Verification Notes

源和验证说明

Sources reviewed: SAP-samples

sap-rpt-1-oss

, Hugging Face

SAP/sap-rpt-1-oss

, and the SAP product page for SAP-RPT.

Verification scope is public source/model/product-page review only. Live SAP tenant validation, live SAP system validation, hosted SAP-RPT API validation, local inference benchmark validation, and production finance workflow validation were not performed.

Product boundary:

SAP-RPT-1-OSS is the local open model workflow documented here.
Hosted SAP-RPT-1 playground is product context only and is not bundled as a client.
SAP-RPT-1.5 is future product context from the 2026-06-18 source review and must not be described as currently available in this skill.

已审核的源：SAP-samples的

sap-rpt-1-oss

、Hugging Face的

SAP/sap-rpt-1-oss

以及SAP-RPT的SAP产品页面。

验证范围仅限公开源/模型/产品页面审核。未执行实时SAP租户验证、实时SAP系统验证、托管SAP-RPT API验证、本地推理基准验证和生产财务工作流验证。

产品边界：

SAP-RPT-1-OSS是本文档所述的本地开源模型工作流。
托管SAP-RPT-1 playground仅作为产品背景，未作为客户端捆绑。
SAP-RPT-1.5是2026-06-18源审核中的未来产品背景，不得在本技能中描述为当前可用。