Agent 基准与可执行环境：定义成功和失败

import java.util.*;
public class Main {
    static void near(double expected,double actual) {
        if (!Double.isFinite(actual) || Math.abs(expected-actual)>1e-9*Math.max(1,Math.abs(expected)))
            throw new AssertionError(expected+" != "+actual);
    }
    static void check(boolean ok) { if(!ok) throw new AssertionError(); }
    static void bad(Runnable r) {
        try { r.run(); } catch(IllegalArgumentException e) { return; }
        throw new AssertionError("invalid input accepted");
    }
enum Status {SUCCESS,FAILURE,TIMEOUT,INFRA}
    record Task(String id,String group) {}
    record Report(double operational,double macro,OptionalDouble validRate,int infrastructure) {}
    static Report report(List<Task> tasks,Map<String,Status> results) {
        if(tasks==null||tasks.isEmpty()||results==null) throw new IllegalArgumentException("empty");
        Set<String> ids=new HashSet<>();Map<String,int[]> groups=new HashMap<>();int success=0,infra=0;
        for(Task t:tasks) {
            if(t==null||t.id()==null||t.group()==null||t.id().isBlank()||t.group().isBlank()||!ids.add(t.id())) throw new IllegalArgumentException("task");
            Status s=results.get(t.id());if(s==null) throw new IllegalArgumentException("missing"); // 缺失日志不静默当作有效完成。
            int[] counts=groups.computeIfAbsent(t.group(),x->new int[2]);counts[1]++;
            if(s==Status.SUCCESS) {success++;counts[0]++;}
            if(s==Status.INFRA) infra++; // 基础设施异常单列，同时保留预声明任务总分母。
        }
        if(results.size()!=ids.size()) throw new IllegalArgumentException("unknown task");
        double macro=0;for(int[] c:groups.values()) macro+=(double)c[0]/c[1]; // 先求每组比率再等权平均，与全体任务比例不同。
        int valid=tasks.size()-infra;
        return new Report((double)success/tasks.size(),macro/groups.size(),valid==0?OptionalDouble.empty():OptionalDouble.of((double)success/valid),infra);
    }
    public static void main(String[] args) {
List<Task> ts=List.of(new Task("a1","A"),new Task("a2","A"),new Task("a3","A"),new Task("b1","B"));
        Report r=report(ts,Map.of("a1",Status.SUCCESS,"a2",Status.FAILURE,"a3",Status.TIMEOUT,"b1",Status.SUCCESS));
        near(.5,r.operational());near(2.0/3,r.macro());near(.5,r.validRate().orElseThrow());
        Report i=report(List.of(new Task("a","G"),new Task("b","G")),Map.of("a",Status.SUCCESS,"b",Status.INFRA));
        near(.5,i.operational());near(1,i.validRate().orElseThrow());check(i.infrastructure()==1);
        check(report(List.of(new Task("a","G")),Map.of("a",Status.INFRA)).validRate().isEmpty());
        near(0,report(List.of(new Task("a","G")),Map.of("a",Status.TIMEOUT)).operational());
        bad(()->report(ts,Map.of()));
        bad(()->report(List.of(new Task("a","G")),Map.of("a",Status.SUCCESS,"x",Status.SUCCESS)));
        bad(()->report(List.of(new Task("a","G"),new Task("a","H")),Map.of("a",Status.SUCCESS)));
        bad(()->report(List.of(),Map.of()));
        System.out.println("PASS 8 cases");
    }
}