Skip to content

GitLab

J
jingxiaoguan
Open
Opened by weidongying@weidongyingMaintainer

规格：合同台账 AI 证据与混合业务统计

Problem Statement

合同台账尚未完整实现 v1.3 原型的模块 AI 标记、可追溯关键词证据、详情页来源标识与多原文件预览。综合检索也无法可靠处理口语化的“业务/业绩”问题：它需要按单项合同与已确认关联的框架订单计算金额，不能由模型猜测关联或跨库计算。

Solution

以合同模块命中作为台账 AI 标记和关键词解析的唯一事实来源，动态展示适用于合同的已启用模块；详情页提供六个业务页签、字段来源、人工关键词覆盖和多文件预览。为混合业务统计建立已确认合同—订单关联，由受控只读工具完成关联、去重、日期归属和金额计算；CoreMind 只抽取意图条件。页面展示业务步骤、公式、合同/订单明细和数据质量提示，不展示 SQL 或思维链。

User Stories

  1. As a 合同管理员, I want every enabled contract module displayed in the ledger as AI or —, so that I can identify keyword-related contracts quickly.
  2. As a 合同管理员, I want the list columns, filters and keyword cards to follow module configuration and sort order, so that newly enabled contract modules work without fixed front-end fields.
  3. As a 合同管理员, I want multi-module filters to use AND semantics, so that every returned contract meets all selected module conditions.
  4. As a 合同管理员, I want “维护状态” to mean the existing review status, so that no second competing status exists.
  5. As a 合同管理员, I want unreviewed contracts to show verify and reviewed contracts to show view, so that actions match review status.
  6. As a 合同管理员, I want contract-number clicks to open a detail page with basic information, amount and settlement, business clauses, risk, keyword analysis and original-file preview, so that I can review complete evidence.
  7. As a 合同管理员, I want each detail field marked AI, manual-confirmed or system, so that I understand its provenance.
  8. As a 合同管理员, I want authorised corrections of AI fields to preserve their AI candidate and audit trail, so that manual changes are traceable.
  9. As a 合同管理员, I want keyword analysis to display module hit count, keyword, category, source excerpt and location, so that every list marker has evidence.
  10. As a 合同管理员, I want manual keyword include/exclude to update markers, contract AI status and searches immediately, so that review does not require re-parsing.
  11. As a 合同管理员, I want the primary PDF previewed by default and all source files switchable, so that multi-file contracts are reviewable.
  12. As a 合同管理员, I want Word sources downloadable with converted Markdown shown when available, so that non-PDF sources remain useful.
  13. As a business user, I want oral expressions such as business, performance, framework contract and contract-plus-order recognised as mixed business statistics, so that I do not need database vocabulary.
  14. As a business user, I want AI business amount calculated as single-contract AI amount plus confirmed-framework-order AI amount, so that totals follow the agreed rule.
  15. As a business user, I want single contracts assigned by signing year and framework orders assigned by start date with created-date fallback, so that yearly statistics have a stable rule.
  16. As a business user, I want framework order amount counted only through confirmed effective contract-order links, so that guessed and duplicate amounts cannot affect totals.
  17. As a business user, I want results grouped by assessment line with both components and their sum, so that I can audit business amount.
  18. As a business user, I want expandable separate contract and order ledgers, so that totals are traceable to source records.
  19. As a business user, I want data-quality notices for unlinked frameworks, date fallback and assessment-line mismatches, so that exclusions are visible.
  20. As a system administrator, I want one order to have at most one confirmed effective contract link, so that it cannot be counted twice.
  21. As a system administrator, I want link imports validated against both ledgers with source IDs, confirmer, time, method and status, so that cross-database links are governable.
  22. As a customer, I want a business-step summary but no SQL, database errors or model chain-of-thought, so that results are transparent and safe.

Implementation Decisions

  • Contract modules are configuration-driven; every enabled, contract-applicable module becomes a ledger marker, filter and keyword-analysis card. Default modules remain project name, service content, technical requirements and personnel requirements.
  • Module AI marker is only AI or —; module evidence belongs in keyword analysis, never the ledger cell.
  • Review status is the sole meaning of prototype maintenance status. Detail provenance is AI, manual-confirmed or system.
  • Manual keyword overrides take immediate effect and do not re-split, re-parse or rebuild vectors.
  • Source preview defaults to primary PDF, supports source switching and uses Word download plus Markdown fallback.
  • Confirmed contract-order links are imported with readable identifiers, validated and stored with stable IDs. Only confirmed effective links count; one framework may have many orders, one order at most one confirmed link.
  • Mixed business amount includes only single contracts and framework agreements. It equals single-contract keyword amount plus confirmed-framework-order keyword amount; supplemental, change and termination agreements are excluded by default.
  • AI means aggregate AI tag; specified keyword/category filters concrete keywords and subwords. Missing year defaults to current year. Framework order grouping uses order assessment line and exposes conflicts with framework line.
  • Dedicated read-only business-statistics tool owns cross-ledger validation, association, de-duplication and arithmetic. CoreMind supplies structured conditions only and never cross-database SQL.
  • Customer-visible processing is fixed: recognise intent, filter single contracts, filter confirmed framework orders, group/de-duplicate, produce details and quality notices.

Testing Decisions

  • Primary seam: authenticated gateway API behavior for ledger rows, detail provenance, keyword evidence, source selection, mixed statistics, references and quality notices.
  • Tool tests: confirmed-link inclusion, unlinked exclusion, one-order-one-count, date fallback, assessment-line grouping and keyword filtering.
  • Existing rich response, contract-module, confirm-draft and faithful-export tests are prior art.
  • Assert externally visible records, totals, labels and exclusions; never assert SQL text, prompt wording or model reasoning.

Out of Scope

  • Inferring contract-order links from contract numbers, customers, projects or model similarity.
  • Counting draft, invalid or unconfirmed links in financial totals.
  • Re-parsing, re-chunking or vector rebuild after manual keyword override.
  • Exposing SQL, database errors or chain-of-thought.
  • Default inclusion of supplemental, change or termination agreements.

Further Notes

The real dataset currently has no confirmed contract-order links. Until imported and confirmed, framework-order amount must be shown as pending association and excluded from mixed totals. Initial storage and controlled-tool work has started; this issue governs completion, UI integration and regression coverage.