feat(ssa): Complete Phase V-A editable analysis plan variables
Features: - Add editable variable selection in workflow plan (SingleVarSelect + MultiVarTags) - Implement 3-layer flexible interception (warning bar + icon + blocking dialog) - Add tool_param_constraints.json for 12 statistical tools parameter validation - Add PATCH /workflow/:id/params API with Zod structural validation - Implement synchronous parameter sync before execution (Promise chaining) - Fix LLM hallucination by strict system prompt constraints - Fix DynamicReport object-based rows compatibility (R baseline_table) - Fix Word export row.map error with same normalization logic - Restore inferGroupingVar for smart default variable selection - Add ReactMarkdown rendering in SSAChatPane - Update SSA module status document to v3.5 Modified files: - backend: workflow.routes, ChatHandlerService, SystemPromptService, FlowTemplateService - frontend: WorkflowTimeline, SSAWorkspacePane, DynamicReport, SSAChatPane, ssaStore, ssa.css - config: tool_param_constraints.json (new) - docs: SSA status doc, team review reports Tested: Cohort study end-to-end execution + report export verified Co-authored-by: Cursor <cursoragent@cursor.com>
This commit is contained in:
52
backend/src/modules/ssa/config/tool_param_constraints.json
Normal file
52
backend/src/modules/ssa/config/tool_param_constraints.json
Normal file
@@ -0,0 +1,52 @@
|
||||
{
|
||||
"ST_DESCRIPTIVE": {
|
||||
"variables": { "paramType": "multi", "requiredType": "any", "hint": "选择需要描述的变量" },
|
||||
"group_var": { "paramType": "single", "requiredType": "categorical", "hint": "分组变量（可选）" }
|
||||
},
|
||||
"ST_T_TEST_IND": {
|
||||
"group_var": { "paramType": "single", "requiredType": "categorical", "maxLevels": 2, "hint": "T检验要求二分类分组变量" },
|
||||
"value_var": { "paramType": "single", "requiredType": "numeric", "hint": "T检验要求连续型因变量" }
|
||||
},
|
||||
"ST_MANN_WHITNEY": {
|
||||
"group_var": { "paramType": "single", "requiredType": "categorical", "maxLevels": 2, "hint": "Mann-Whitney检验要求二分类分组变量" },
|
||||
"value_var": { "paramType": "single", "requiredType": "numeric", "hint": "要求连续型因变量" }
|
||||
},
|
||||
"ST_T_TEST_PAIRED": {
|
||||
"before_var": { "paramType": "single", "requiredType": "numeric", "hint": "前测变量应为连续型" },
|
||||
"after_var": { "paramType": "single", "requiredType": "numeric", "hint": "后测变量应为连续型" }
|
||||
},
|
||||
"ST_WILCOXON": {
|
||||
"before_var": { "paramType": "single", "requiredType": "numeric", "hint": "前测变量应为连续型" },
|
||||
"after_var": { "paramType": "single", "requiredType": "numeric", "hint": "后测变量应为连续型" }
|
||||
},
|
||||
"ST_CHI_SQUARE": {
|
||||
"var1": { "paramType": "single", "requiredType": "categorical", "hint": "卡方检验要求分类变量" },
|
||||
"var2": { "paramType": "single", "requiredType": "categorical", "hint": "卡方检验要求分类变量" }
|
||||
},
|
||||
"ST_FISHER": {
|
||||
"var1": { "paramType": "single", "requiredType": "categorical", "hint": "Fisher检验要求分类变量" },
|
||||
"var2": { "paramType": "single", "requiredType": "categorical", "hint": "Fisher检验要求分类变量" }
|
||||
},
|
||||
"ST_CORRELATION": {
|
||||
"var_x": { "paramType": "single", "requiredType": "numeric", "hint": "相关分析要求连续型变量" },
|
||||
"var_y": { "paramType": "single", "requiredType": "numeric", "hint": "相关分析要求连续型变量" }
|
||||
},
|
||||
"ST_LOGISTIC_BINARY": {
|
||||
"outcome_var": { "paramType": "single", "requiredType": "categorical", "maxLevels": 2, "hint": "二元Logistic回归要求二分类结局变量" },
|
||||
"predictors": { "paramType": "multi", "requiredType": "any", "hint": "预测变量" },
|
||||
"confounders": { "paramType": "multi", "requiredType": "any", "hint": "混杂因素（可选）" }
|
||||
},
|
||||
"ST_LINEAR_REG": {
|
||||
"outcome_var": { "paramType": "single", "requiredType": "numeric", "hint": "线性回归要求连续型结局变量" },
|
||||
"predictors": { "paramType": "multi", "requiredType": "any", "hint": "预测变量" },
|
||||
"confounders": { "paramType": "multi", "requiredType": "any", "hint": "混杂因素（可选）" }
|
||||
},
|
||||
"ST_ANOVA_ONE": {
|
||||
"group_var": { "paramType": "single", "requiredType": "categorical", "minLevels": 3, "hint": "ANOVA要求3组及以上分组变量" },
|
||||
"value_var": { "paramType": "single", "requiredType": "numeric", "hint": "要求连续型因变量" }
|
||||
},
|
||||
"ST_BASELINE_TABLE": {
|
||||
"group_var": { "paramType": "single", "requiredType": "categorical", "minLevels": 2, "maxLevels": 5, "hint": "基线表需要分类分组变量" },
|
||||
"analyze_vars": { "paramType": "multi", "requiredType": "any", "hint": "选择需要分析的变量" }
|
||||
}
|
||||
}
|
||||
@@ -9,6 +9,7 @@
|
||||
*/
|
||||
|
||||
import { FastifyInstance, FastifyRequest, FastifyReply } from 'fastify';
|
||||
import { z } from 'zod';
|
||||
import { logger } from '../../../common/logging/index.js';
|
||||
import { workflowPlannerService } from '../services/WorkflowPlannerService.js';
|
||||
import { workflowExecutorService } from '../services/WorkflowExecutorService.js';
|
||||
@@ -372,6 +373,143 @@ export default async function workflowRoutes(app: FastifyInstance) {
|
||||
}
|
||||
);
|
||||
|
||||
/**
|
||||
* PATCH /workflow/:workflowId/params
|
||||
* Phase V: 批量更新 workflow step 参数（变量可编辑化）
|
||||
*
|
||||
* Zod 结构校验防火墙：
|
||||
* - 结构非法（字段类型错误、必填字段缺失）→ 400 Bad Request
|
||||
* - 统计学不合理但结构合法 → 放行，交给 R 引擎 tryCatch
|
||||
*/
|
||||
app.patch<{ Params: { workflowId: string }; Body: { steps: Array<{ stepOrder: number; params: Record<string, any> }> } }>(
|
||||
'/:workflowId/params',
|
||||
async (request, reply) => {
|
||||
const { workflowId } = request.params;
|
||||
const { steps } = request.body;
|
||||
|
||||
const PatchStepSchema = z.object({
|
||||
stepOrder: z.number().int().positive(),
|
||||
params: z.record(z.string(), z.unknown()),
|
||||
});
|
||||
const PatchBodySchema = z.object({
|
||||
steps: z.array(PatchStepSchema).min(1),
|
||||
});
|
||||
|
||||
const validation = PatchBodySchema.safeParse({ steps });
|
||||
if (!validation.success) {
|
||||
return reply.status(400).send({
|
||||
success: false,
|
||||
error: 'Invalid request body',
|
||||
details: validation.error.flatten(),
|
||||
});
|
||||
}
|
||||
|
||||
try {
|
||||
const workflow = await prisma.ssaWorkflow.findUnique({
|
||||
where: { id: workflowId },
|
||||
select: { id: true, status: true, sessionId: true },
|
||||
});
|
||||
|
||||
if (!workflow) {
|
||||
return reply.status(404).send({ success: false, error: 'Workflow not found' });
|
||||
}
|
||||
|
||||
if (workflow.status !== 'planned' && workflow.status !== 'pending') {
|
||||
return reply.status(409).send({
|
||||
success: false,
|
||||
error: `Cannot modify params for workflow in '${workflow.status}' state`,
|
||||
});
|
||||
}
|
||||
|
||||
// Validate variable names exist in session data schema
|
||||
const session = await prisma.ssaSession.findUnique({
|
||||
where: { id: workflow.sessionId },
|
||||
select: { dataSchema: true },
|
||||
});
|
||||
const schema = session?.dataSchema as any;
|
||||
const validColumnNames = new Set<string>(
|
||||
(schema?.columns || []).map((c: any) => c.name)
|
||||
);
|
||||
|
||||
for (const stepPatch of validation.data.steps) {
|
||||
for (const [key, value] of Object.entries(stepPatch.params)) {
|
||||
if (typeof value === 'string' && value && !key.startsWith('_')) {
|
||||
if (['group_var', 'outcome_var', 'value_var', 'var_x', 'var_y',
|
||||
'before_var', 'after_var', 'var1', 'var2'].includes(key)) {
|
||||
if (validColumnNames.size > 0 && !validColumnNames.has(value)) {
|
||||
return reply.status(400).send({
|
||||
success: false,
|
||||
error: `Variable '${value}' in step ${stepPatch.stepOrder}.${key} does not exist in the dataset`,
|
||||
});
|
||||
}
|
||||
}
|
||||
}
|
||||
if (Array.isArray(value) && ['analyze_vars', 'predictors', 'variables', 'confounders'].includes(key)) {
|
||||
for (const v of value) {
|
||||
if (typeof v === 'string' && validColumnNames.size > 0 && !validColumnNames.has(v)) {
|
||||
return reply.status(400).send({
|
||||
success: false,
|
||||
error: `Variable '${v}' in step ${stepPatch.stepOrder}.${key} does not exist in the dataset`,
|
||||
});
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// Update each step's inputParams in the database
|
||||
const updatePromises = validation.data.steps.map((stepPatch) =>
|
||||
prisma.ssaWorkflowStep.updateMany({
|
||||
where: {
|
||||
workflowId,
|
||||
stepOrder: stepPatch.stepOrder,
|
||||
},
|
||||
data: {
|
||||
inputParams: stepPatch.params as any,
|
||||
},
|
||||
})
|
||||
);
|
||||
|
||||
await Promise.all(updatePromises);
|
||||
|
||||
// Also update the workflowPlan JSON blob's steps params
|
||||
const currentPlan = await prisma.ssaWorkflow.findUnique({
|
||||
where: { id: workflowId },
|
||||
select: { workflowPlan: true },
|
||||
});
|
||||
if (currentPlan?.workflowPlan) {
|
||||
const plan = currentPlan.workflowPlan as any;
|
||||
if (plan.steps) {
|
||||
for (const stepPatch of validation.data.steps) {
|
||||
const planStep = plan.steps.find((s: any) => s.step_number === stepPatch.stepOrder);
|
||||
if (planStep) {
|
||||
planStep.params = stepPatch.params;
|
||||
}
|
||||
}
|
||||
await prisma.ssaWorkflow.update({
|
||||
where: { id: workflowId },
|
||||
data: { workflowPlan: plan },
|
||||
});
|
||||
}
|
||||
}
|
||||
|
||||
logger.info('[SSA:API] Workflow params updated', {
|
||||
workflowId,
|
||||
stepsUpdated: validation.data.steps.length,
|
||||
});
|
||||
|
||||
return reply.send({ success: true, stepsUpdated: validation.data.steps.length });
|
||||
|
||||
} catch (error: any) {
|
||||
logger.error('[SSA:API] Patch workflow params failed', {
|
||||
workflowId,
|
||||
error: error.message,
|
||||
});
|
||||
return reply.status(500).send({ success: false, error: error.message });
|
||||
}
|
||||
}
|
||||
);
|
||||
|
||||
/**
|
||||
* POST /workflow/profile
|
||||
* 生成数据画像
|
||||
|
||||
@@ -239,7 +239,8 @@ export class ChatHandlerService {
|
||||
const toolOutputs = [
|
||||
guardToolOutput,
|
||||
planSummary,
|
||||
'[系统提示] 你刚刚为用户制定了上述分析方案。请用自然语言向用户解释这个方案：包括为什么选这些方法、分析步骤的逻辑。不要重复列步骤编号和工具代码，要用用户能理解的语言说明。最后提示用户确认方案后即可执行。',
|
||||
'[系统指令] 你刚刚为用户制定了上述分析方案。请用自然语言向用户解释这个方案：包括为什么选这些方法、分析步骤的逻辑。不要重复列步骤编号和工具代码，要用用户能理解的语言说明。最后提示用户确认方案后即可执行。',
|
||||
'【禁止事项】不要预测、模拟或编造任何分析结果、数值或表格。方案只是计划，R 引擎尚未执行，你不知道结果是什么。',
|
||||
].filter(Boolean).join('\n\n');
|
||||
|
||||
const messages = await conversationService.buildContext(
|
||||
@@ -397,7 +398,9 @@ export class ChatHandlerService {
|
||||
writer: StreamWriter,
|
||||
placeholderMessageId: string,
|
||||
): Promise<HandleResult> {
|
||||
// 清除 pending 状态
|
||||
// 先读取 pending 元数据（含 workflowId），再清除
|
||||
const pending = await askUserService.getPending(sessionId);
|
||||
const pendingMeta = pending?.metadata || {};
|
||||
await askUserService.clearPending(sessionId);
|
||||
|
||||
if (response.action === 'skip') {
|
||||
@@ -423,15 +426,21 @@ export class ChatHandlerService {
|
||||
const selectedValue = response.selectedValues?.[0];
|
||||
|
||||
if (selectedValue === 'confirm_plan') {
|
||||
// Phase IV: 确认分析方案 → 前端将触发 executeWorkflow
|
||||
const workflowId = response.metadata?.workflowId || '';
|
||||
// Phase IV: 确认分析方案 → 前端打开工作区，用户手动点击执行
|
||||
const workflowId = pendingMeta.workflowId || response.metadata?.workflowId || '';
|
||||
const messages = await conversationService.buildContext(
|
||||
sessionId, conversationId, 'analyze',
|
||||
`[系统提示] 用户已确认分析方案（workflow: ${workflowId}）。请简要确认："好的，方案已确认，正在准备执行分析..."。`,
|
||||
[
|
||||
`[系统指令——严格遵守] 用户已确认分析方案（workflow: ${workflowId}）。`,
|
||||
'你只需回复一句简短的确认消息，例如："好的，方案已确认。请在右侧工作区点击「开始执行分析」启动 R 引擎。"',
|
||||
'【铁律】禁止在此回复中生成任何分析结果、表格、P值、统计量、数值。',
|
||||
'你不是计算引擎，所有数值结果将由 R 统计引擎独立计算后返回。',
|
||||
'你的回复不得超过 2 句话。',
|
||||
].join('\n'),
|
||||
);
|
||||
|
||||
const result = await conversationService.streamToSSE(messages, writer, {
|
||||
temperature: 0.3, maxTokens: 300,
|
||||
temperature: 0.1, maxTokens: 150,
|
||||
});
|
||||
|
||||
await conversationService.finalizeAssistantMessage(
|
||||
@@ -451,12 +460,12 @@ export class ChatHandlerService {
|
||||
// Phase III: 确认使用推荐方法 → 提示可以开始分析
|
||||
const messages = await conversationService.buildContext(
|
||||
sessionId, conversationId, 'analyze',
|
||||
'[系统提示] 用户已确认使用推荐的统计方法。请简要确认方案，告知用户可以在对话中说"开始分析"或在右侧面板触发执行。',
|
||||
'[系统指令] 用户已确认使用推荐的统计方法。请简要确认方案，告知用户可以在对话中说"开始分析"或在右侧面板触发执行。禁止生成任何数值或假设的分析结果。',
|
||||
);
|
||||
|
||||
const result = await conversationService.streamToSSE(messages, writer, {
|
||||
temperature: 0.5,
|
||||
maxTokens: 800,
|
||||
temperature: 0.3,
|
||||
maxTokens: 500,
|
||||
});
|
||||
|
||||
await conversationService.finalizeAssistantMessage(
|
||||
|
||||
@@ -131,7 +131,7 @@ export class FlowTemplateService {
|
||||
params[key] = query.outcome_var;
|
||||
break;
|
||||
case '{{grouping_var}}':
|
||||
params[key] = query.grouping_var;
|
||||
params[key] = query.grouping_var || this.inferGroupingVar(query, profile);
|
||||
break;
|
||||
case '{{all_predictors}}':
|
||||
params[key] = query.predictor_vars;
|
||||
@@ -150,6 +150,51 @@ export class FlowTemplateService {
|
||||
return { params, epvWarning };
|
||||
}
|
||||
|
||||
/**
|
||||
* 当 LLM 未识别 grouping_var 时，从数据画像中自动推断
|
||||
* 优先选择：二分类变量（排除 outcome_var），最典型的分组/暴露变量
|
||||
*/
|
||||
private inferGroupingVar(query: ParsedQuery, profile?: DataProfile | null): string | null {
|
||||
if (!profile?.columns) return null;
|
||||
|
||||
const excludeVars = new Set<string>();
|
||||
if (query.outcome_var) excludeVars.add(query.outcome_var.toLowerCase());
|
||||
|
||||
const binaryCandidates = profile.columns.filter(c =>
|
||||
c.type === 'categorical' &&
|
||||
c.totalLevels === 2 &&
|
||||
!excludeVars.has(c.name.toLowerCase())
|
||||
);
|
||||
|
||||
if (binaryCandidates.length > 0) {
|
||||
const chosen = binaryCandidates[0].name;
|
||||
logger.info('[SSA:FlowTemplate] Auto-inferred grouping_var', {
|
||||
chosen,
|
||||
candidates: binaryCandidates.map(c => c.name),
|
||||
});
|
||||
return chosen;
|
||||
}
|
||||
|
||||
const categoricalCandidates = profile.columns.filter(c =>
|
||||
c.type === 'categorical' &&
|
||||
c.totalLevels !== undefined &&
|
||||
c.totalLevels >= 2 &&
|
||||
c.totalLevels <= 5 &&
|
||||
!excludeVars.has(c.name.toLowerCase())
|
||||
);
|
||||
|
||||
if (categoricalCandidates.length > 0) {
|
||||
const chosen = categoricalCandidates[0].name;
|
||||
logger.info('[SSA:FlowTemplate] Auto-inferred grouping_var (categorical)', {
|
||||
chosen,
|
||||
});
|
||||
return chosen;
|
||||
}
|
||||
|
||||
logger.warn('[SSA:FlowTemplate] No suitable grouping_var found in profile');
|
||||
return null;
|
||||
}
|
||||
|
||||
/**
|
||||
* 构建默认参数（非 paramsMapping 模板步骤使用）
|
||||
*/
|
||||
|
||||
@@ -123,12 +123,29 @@ export class SystemPromptService {
|
||||
}
|
||||
|
||||
private fallbackBaseSystem(): string {
|
||||
return `你是 SSA-Pro 智能统计分析助手，专注于临床研究统计分析。
|
||||
你具备以下能力：
|
||||
- 理解临床研究数据的结构和特征
|
||||
- 推荐合适的统计分析方法
|
||||
- 解读统计分析结果
|
||||
- 用通俗易懂的语言向医学研究者解释统计概念
|
||||
return `你是 SSA-Pro 智能统计分析助手。你的职责是**规划、解释和沟通**，而非计算。
|
||||
|
||||
## 你的身份与职能边界
|
||||
|
||||
你是「分析规划者」和「结果解读者」，不是「计算引擎」。
|
||||
系统后端有独立的 R 统计计算引擎，所有统计计算均由 R 引擎完成。
|
||||
|
||||
### 你可以做的：
|
||||
- 理解用户的分析需求，识别意图
|
||||
- 推荐合适的统计方法，解释选择理由
|
||||
- 制定分析方案（选择工具、参数）
|
||||
- 解读 R 引擎返回的真实结果
|
||||
- 用通俗语言向研究者解释统计概念
|
||||
|
||||
### 绝对禁止：
|
||||
- **禁止编造或生成任何数值结果**（P值、均值、标准差、置信区间、检验统计量等）
|
||||
- **禁止模拟或假设分析结果**（即使用户催促，也不能捏造数据）
|
||||
- **禁止生成结果表格**（除非表格数据来自 R 引擎的真实输出）
|
||||
- 如果还没有 R 引擎的执行结果，只能说"正在等待执行"或"方案已确认，即将启动分析"
|
||||
|
||||
### 关键原则：
|
||||
没有 R 引擎的真实输出 → 不回答任何具体数值。
|
||||
这是铁律，违反将导致临床研究的严重错误。
|
||||
|
||||
沟通原则：
|
||||
- 使用中文回复
|
||||
@@ -139,12 +156,12 @@ export class SystemPromptService {
|
||||
|
||||
private fallbackIntentInstruction(intent: IntentType): string {
|
||||
const map: Record<IntentType, string> = {
|
||||
chat: '请基于统计知识和用户数据直接回答用户的问题。不要主动建议执行分析，除非用户明确要求。简洁作答，分点清晰。',
|
||||
explore: '用户想了解数据的特征。请基于上方的数据摘要信息，帮用户解读数据特征（缺失、分布、异常值等）。可以推断 PICO 结构。不要执行分析。',
|
||||
consult: '用户在咨询统计方法。请根据数据特征和研究目的推荐合适的统计方法，给出选择理由和前提条件。不要直接执行分析。提供替代方案。',
|
||||
analyze: '以下是工具执行结果。请向用户简要说明分析进展和关键发现。使用通俗语言，避免过度技术化。',
|
||||
discuss: '用户想讨论分析结果。请帮助用户深入解读结果，解释统计量的含义，讨论临床意义和局限性。',
|
||||
feedback: '用户对之前的分析结果不满意或有改进建议。请分析问题原因，提出改进方案（如更换统计方法、调整参数等）。',
|
||||
chat: '请基于统计知识和用户数据直接回答用户的问题。不要主动建议执行分析，除非用户明确要求。简洁作答，分点清晰。禁止编造任何数值。',
|
||||
explore: '用户想了解数据的特征。请基于上方的数据摘要信息，帮用户解读数据特征（缺失、分布、异常值等）。可以推断 PICO 结构。不要执行分析，不要编造统计数值。',
|
||||
consult: '用户在咨询统计方法。请根据数据特征和研究目的推荐合适的统计方法，给出选择理由和前提条件。不要直接执行分析。提供替代方案。禁止给出任何假设的分析结果数值。',
|
||||
analyze: '你正在协助用户进行分析规划。你的职责限于：解释分析方案的思路和方法选择理由。禁止生成任何P值、统计量、均值、分析结果表格。所有数值结果只能来自 R 引擎的真实执行输出。如果 R 引擎还没有返回结果，只能说明方案状态，不能自行填充结果。',
|
||||
discuss: '用户想讨论分析结果。请仅基于 R 引擎返回的真实数据帮助用户解读，解释统计量的含义，讨论临床意义和局限性。禁止补充或编造 R 引擎未返回的数值。',
|
||||
feedback: '用户对之前的分析结果不满意或有改进建议。请分析问题原因，提出改进方案（如更换统计方法、调整参数等）。禁止编造数值来论证改进效果。',
|
||||
};
|
||||
return map[intent];
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user