diff --git a/evaluation/web/src/api.test.ts b/evaluation/web/src/api.test.ts
index 0a2e8dfd6..c281c3a45 100644
--- a/evaluation/web/src/api.test.ts
+++ b/evaluation/web/src/api.test.ts
@@ -389,6 +389,7 @@ describe("EvaluationApi HTTP", () => {
const report = {
task_id: "task-1",
acceptance_valid: true,
+ treatment_mode: "off_on",
off: {
arm: "off",
state: "treatment_validated",
diff --git a/evaluation/web/src/api.ts b/evaluation/web/src/api.ts
index ca7830a11..d59ce8ac2 100644
--- a/evaluation/web/src/api.ts
+++ b/evaluation/web/src/api.ts
@@ -16,6 +16,11 @@
import type {
AccountUsage,
+ Arm,
+ BaselineCandidate,
+ BaselineComparisonResponse,
+ BaselineRecord,
+ BaselineSelection,
BatchCreate,
BatchControlEvent,
BatchEventSubscription,
@@ -42,6 +47,7 @@ import type {
TaskRecord,
TaskStatus,
TaskSummary,
+ TreatmentMode,
} from "./types";
import { z } from "zod";
@@ -117,6 +123,7 @@ const failureCategorySchema = z.enum([
]);
const codexModelSchema = z.string().regex(/^[A-Za-z0-9][A-Za-z0-9._-]{0,127}$/);
const batchTaskSetSchema = z.enum(["swebench-pro-public-v2", "swebench-pro-stability-v1"]);
+const treatmentModeSchema = z.enum(["off_on", "on_only", "off_only"]);
const taskCreateSchema = z.strictObject({
powercontext_ref: z.union([z.literal("latest"), z.string().regex(/^commit:[0-9a-fA-F]{40}$/)]),
@@ -124,15 +131,15 @@ const taskCreateSchema = z.strictObject({
instance_id: z.literal(INSTANCE_ID),
model: codexModelSchema,
reasoning_effort: z.literal("medium"),
- treatment_mode: z.literal("off_on"),
+ treatment_mode: treatmentModeSchema,
idempotency_key: z.string().min(8).max(128).regex(/^[A-Za-z0-9._-]+$/),
});
const taskResultSchema = z.strictObject({
artifact_dir: z.string(),
report_path: z.string(),
- off_resolved: z.boolean(),
- on_resolved: z.boolean(),
+ off_resolved: z.boolean().nullable(),
+ on_resolved: z.boolean().nullable(),
});
const taskRecordBaseShape = {
@@ -246,7 +253,7 @@ const capabilitiesSchema = z.strictObject({
instances: z.array(z.literal(INSTANCE_ID)),
models: z.array(codexModelSchema),
reasoning_efforts: z.array(z.literal("medium")),
- treatment_modes: z.array(z.literal("off_on")),
+ treatment_modes: z.array(treatmentModeSchema),
});
const healthSchema = z.strictObject({
@@ -334,12 +341,13 @@ const goldValidationAuditSchema = z.strictObject({
const reportSchema = z.strictObject({
task_id: z.string(),
acceptance_valid: z.boolean(),
- off: armSchema("off"),
- on: armSchema("on"),
- comparison: comparisonSchema,
+ treatment_mode: treatmentModeSchema,
+ off: armSchema("off").nullable(),
+ on: armSchema("on").nullable(),
+ comparison: comparisonSchema.nullable(),
evidence: z.strictObject({
- off: treatmentEvidenceSchema,
- on: treatmentEvidenceSchema,
+ off: treatmentEvidenceSchema.nullable(),
+ on: treatmentEvidenceSchema.nullable(),
}),
gold_validation: goldValidationAuditSchema.nullable().optional(),
revisions: z.record(z.string(), z.string()),
@@ -352,7 +360,7 @@ const batchCreateSchema = z.strictObject({
task_set: batchTaskSetSchema,
model: codexModelSchema,
reasoning_effort: z.literal("medium"),
- treatment_mode: z.literal("off_on"),
+ treatment_mode: treatmentModeSchema,
idempotency_key: z.string().min(8).max(128).regex(/^[A-Za-z0-9._-]+$/),
usage_pause_percent: z.number().int().min(1).max(100),
initial_control_intent: z.enum(["run", "pause"]),
@@ -428,7 +436,7 @@ const batchPreviewSchema = z.strictObject({
task_set: batchTaskSetSchema,
model: codexModelSchema,
reasoning_effort: z.literal("medium"),
- treatment_mode: z.literal("off_on"),
+ treatment_mode: treatmentModeSchema,
total_tasks: z.number().int().positive(),
usage_pause_percent: z.number().int().min(1).max(100),
usage: usageSnapshotSchema.nullable(),
@@ -449,24 +457,25 @@ const resolutionAggregateSchema = z.strictObject({
rate_percent: z.number().min(0).max(100),
});
const tokenMetricAggregateSchema = z.strictObject({
- off: nonnegativeIntegerSchema,
- on: nonnegativeIntegerSchema,
- delta: z.number().int(),
- off_measured_tasks: nonnegativeIntegerSchema,
- on_measured_tasks: nonnegativeIntegerSchema,
+ off: nonnegativeIntegerSchema.nullable(),
+ on: nonnegativeIntegerSchema.nullable(),
+ delta: z.number().int().nullable(),
+ off_measured_tasks: nonnegativeIntegerSchema.nullable(),
+ on_measured_tasks: nonnegativeIntegerSchema.nullable(),
});
const batchReportSchema = z.strictObject({
batch_id: z.string(),
+ treatment_mode: treatmentModeSchema,
report_revision: nonnegativeIntegerSchema,
total_tasks: z.number().int().positive(),
terminal_tasks: nonnegativeIntegerSchema,
- comparable_pairs: nonnegativeIntegerSchema,
+ comparable_pairs: nonnegativeIntegerSchema.nullable(),
execution_failures: nonnegativeIntegerSchema,
cancelled_tasks: nonnegativeIntegerSchema,
- off: resolutionAggregateSchema,
- on: resolutionAggregateSchema,
- resolution_rate_delta_points: z.number(),
- pair_categories: z.record(pairCategorySchema, nonnegativeIntegerSchema),
+ off: resolutionAggregateSchema.nullable(),
+ on: resolutionAggregateSchema.nullable(),
+ resolution_rate_delta_points: z.number().nullable(),
+ pair_categories: z.record(pairCategorySchema, nonnegativeIntegerSchema).nullable(),
task_statuses: z.record(taskStatusSchema, nonnegativeIntegerSchema),
tokens: z.strictObject({
input: tokenMetricAggregateSchema,
@@ -624,6 +633,76 @@ const batchTaskDetailSchema = z.strictObject({
on: tokensflowFinalizationSchema.nullable(),
}),
});
+const baselineRecordSchema = z.strictObject({
+ baseline_id: z.string(),
+ name: z.string().min(1).max(120),
+ source_batch_id: z.string(),
+ source_arm: z.enum(["off", "on"]),
+ source_report_revision: nonnegativeIntegerSchema,
+ benchmark: z.literal("swebench-pro"),
+ task_set: batchTaskSetSchema,
+ instance_set_digest: z.string().regex(/^[0-9a-f]{64}$/),
+ total_tasks: z.number().int().positive(),
+ resolved_tasks: nonnegativeIntegerSchema,
+ execution_failures: nonnegativeIntegerSchema,
+ model: codexModelSchema,
+ reasoning_effort: z.literal("medium"),
+ dataset_revision: z.string(),
+ harness_revision: z.string(),
+ powercontext_sha: z.string().regex(/^[0-9a-f]{40}$/).nullable(),
+ codex_version: z.string().nullable(),
+ created_at: timestampSchema,
+});
+const baselineCompatibilitySchema = z.strictObject({
+ status: z.enum(["compatible", "warning", "incompatible"]),
+ reasons: z.array(z.string()),
+});
+const baselineCandidateSchema = z.strictObject({
+ baseline: baselineRecordSchema,
+ compatibility: baselineCompatibilitySchema,
+});
+const baselineSelectionSchema = z.strictObject({
+ baseline_id: z.string(),
+ current_arm: z.enum(["off", "on"]),
+});
+const historicalTokenComparisonSchema = z.strictObject({
+ baseline: nonnegativeIntegerSchema,
+ current: nonnegativeIntegerSchema,
+ delta: z.number().int(),
+ baseline_measured_tasks: nonnegativeIntegerSchema,
+ current_measured_tasks: nonnegativeIntegerSchema,
+});
+const baselineComparisonSchema = z.strictObject({
+ baseline: baselineRecordSchema,
+ current_arm: z.enum(["off", "on"]),
+ compatibility: baselineCompatibilitySchema,
+ coverage: z.strictObject({
+ matched_tasks: nonnegativeIntegerSchema,
+ comparable_tasks: nonnegativeIntegerSchema,
+ current_execution_failures: nonnegativeIntegerSchema,
+ baseline_execution_failures: nonnegativeIntegerSchema,
+ }),
+ resolution: z.strictObject({
+ baseline_resolved: nonnegativeIntegerSchema,
+ current_resolved: nonnegativeIntegerSchema,
+ total: nonnegativeIntegerSchema,
+ baseline_rate_percent: z.number().min(0).max(100),
+ current_rate_percent: z.number().min(0).max(100),
+ delta_points: z.number(),
+ }),
+ outcome_categories: z.record(
+ z.enum(["baseline_fail_current_pass", "baseline_pass_current_fail", "both_pass", "both_fail"]),
+ nonnegativeIntegerSchema,
+ ),
+ input_tokens: historicalTokenComparisonSchema.nullable(),
+ output_tokens: historicalTokenComparisonSchema.nullable(),
+ total_tokens: historicalTokenComparisonSchema.nullable(),
+});
+const baselineComparisonResponseSchema = z.strictObject({
+ batch_id: z.string(),
+ report_revision: nonnegativeIntegerSchema,
+ comparisons: z.array(baselineComparisonSchema),
+});
const contextEventSchema = z.strictObject({
sequence: z.number().int().positive(),
observed_at: timestampSchema,
@@ -693,6 +772,22 @@ function validateBatchReport(value: unknown): BatchReport {
return validateWithSchema(batchReportSchema, value);
}
+function validateBaseline(value: unknown): BaselineRecord {
+ return validateWithSchema(baselineRecordSchema, value);
+}
+
+function validateBaselineCandidate(value: unknown): BaselineCandidate {
+ return validateWithSchema(baselineCandidateSchema, value);
+}
+
+function validateBaselineSelection(value: unknown): BaselineSelection {
+ return validateWithSchema(baselineSelectionSchema, value);
+}
+
+function validateBaselineComparisonResponse(value: unknown): BaselineComparisonResponse {
+ return validateWithSchema(baselineComparisonResponseSchema, value);
+}
+
function validateBatchTaskPage(value: unknown): BatchTaskPage {
return validateWithSchema(batchTaskPageSchema, value);
}
@@ -766,12 +861,101 @@ export class EvaluationApi {
);
}
+ listBaselines(signal?: AbortSignal): Promise
{
+ return this.#json(
+ apiPath("/baselines"),
+ (value) => {
+ if (!Array.isArray(value)) throw new ApiError(null, "invalid_response", GENERIC_ERROR_MESSAGE);
+ return value.map(validateBaseline);
+ },
+ withSignal(signal),
+ );
+ }
+
+ createBaseline(
+ request: {
+ name: string;
+ source_batch_id: string;
+ source_arm: Arm;
+ expected_report_revision: number;
+ idempotency_key: string;
+ },
+ signal?: AbortSignal,
+ ): Promise {
+ return this.#json(apiPath("/baselines"), validateBaseline, {
+ method: "POST",
+ headers: { "Content-Type": "application/json" },
+ body: JSON.stringify(request),
+ ...withSignal(signal),
+ });
+ }
+
+ listBaselineCandidates(
+ batchId: string,
+ currentArm: Arm,
+ signal?: AbortSignal,
+ ): Promise {
+ return this.#json(
+ batchPath(batchId, `/baseline-candidates?current_arm=${encodeURIComponent(currentArm)}`),
+ (value) => {
+ if (!Array.isArray(value)) throw new ApiError(null, "invalid_response", GENERIC_ERROR_MESSAGE);
+ return value.map(validateBaselineCandidate);
+ },
+ withSignal(signal),
+ );
+ }
+
+ listBaselineSelections(batchId: string, signal?: AbortSignal): Promise {
+ return this.#json(
+ batchPath(batchId, "/baseline-selections"),
+ (value) => {
+ if (!Array.isArray(value)) throw new ApiError(null, "invalid_response", GENERIC_ERROR_MESSAGE);
+ return value.map(validateBaselineSelection);
+ },
+ withSignal(signal),
+ );
+ }
+
+ updateBaselineSelections(
+ batchId: string,
+ selections: BaselineSelection[],
+ signal?: AbortSignal,
+ ): Promise {
+ return this.#json(
+ batchPath(batchId, "/baseline-selections"),
+ (value) => {
+ if (!Array.isArray(value)) throw new ApiError(null, "invalid_response", GENERIC_ERROR_MESSAGE);
+ return value.map(validateBaselineSelection);
+ },
+ {
+ method: "PUT",
+ headers: { "Content-Type": "application/json" },
+ body: JSON.stringify({ selections }),
+ ...withSignal(signal),
+ },
+ );
+ }
+
+ getBaselineComparisons(batchId: string, signal?: AbortSignal): Promise {
+ return this.#json(
+ batchPath(batchId, "/baseline-comparisons"),
+ validateBaselineComparisonResponse,
+ withSignal(signal),
+ );
+ }
+
getBatch(batchId: string, signal?: AbortSignal): Promise {
return this.#json(batchPath(batchId), validateBatch, withSignal(signal));
}
previewBatch(
- request: { powercontext_ref: string; task_set: BatchTaskSet; model: string; usage_pause_percent: number },
+ request: {
+ powercontext_ref: string;
+ task_set: BatchTaskSet;
+ model: string;
+ treatment_mode: TreatmentMode;
+ usage_pause_percent: number;
+ },
signal?: AbortSignal,
): Promise {
return this.#json(apiPath("/batches/preview"), validateBatchPreview, {
diff --git a/evaluation/web/src/components/AppShell.tsx b/evaluation/web/src/components/AppShell.tsx
index e63b90fe2..f3a1f9399 100644
--- a/evaluation/web/src/components/AppShell.tsx
+++ b/evaluation/web/src/components/AppShell.tsx
@@ -52,7 +52,7 @@ export function AppShell({ api, path, batchId, navigate, children }: AppShellPro
{
href: encodedBatchId === null ? "/" : `/report/${encodedBatchId}`,
label: "总体报告",
- current: !taskReport && !runtimeReport,
+ current: path !== "/baselines" && !taskReport && !runtimeReport,
disabled: false,
},
{
@@ -67,6 +67,12 @@ export function AppShell({ api, path, batchId, navigate, children }: AppShellPro
current: taskReport,
disabled: encodedBatchId === null,
},
+ {
+ href: "/baselines",
+ label: "基线库",
+ current: path === "/baselines",
+ disabled: false,
+ },
];
const onLink = (event: MouseEvent, href: string, disabled = false) => {
if (disabled) {
diff --git a/evaluation/web/src/components/BaselineLibrary.tsx b/evaluation/web/src/components/BaselineLibrary.tsx
new file mode 100644
index 000000000..c4133bae6
--- /dev/null
+++ b/evaluation/web/src/components/BaselineLibrary.tsx
@@ -0,0 +1,95 @@
+/*
+ * Copyright (c) 2026 OceanBase.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+import { useCallback, useEffect, useRef, useState, type MouseEvent } from "react";
+
+import type { EvaluationApi } from "../api";
+import type { BaselineRecord } from "../types";
+
+export function BaselineLibrary({ api, navigate }: { api: EvaluationApi; navigate(path: string): void }) {
+ const [baselines, setBaselines] = useState(null);
+ const [error, setError] = useState(false);
+ const controller = useRef(null);
+
+ const load = useCallback(() => {
+ controller.current?.abort();
+ const next = new AbortController();
+ controller.current = next;
+ setError(false);
+ api.listBaselines(next.signal)
+ .then((items) => { if (!next.signal.aborted) setBaselines(items); })
+ .catch(() => { if (!next.signal.aborted) setError(true); });
+ }, [api]);
+
+ useEffect(() => {
+ load();
+ return () => controller.current?.abort();
+ }, [load]);
+
+ const onLink = (event: MouseEvent, path: string) => {
+ if (event.button !== 0 || event.metaKey || event.ctrlKey || event.shiftKey || event.altKey) return;
+ event.preventDefault();
+ navigate(path);
+ };
+
+ if (error) {
+ return ;
+ }
+ if (baselines === null) return ;
+
+ return (
+
+
+
+ {baselines.length === 0 ? 暂无基线，请从已完成批次的总体报告保存。
: (
+
+
+
+
+ | 名称 | Arm | 解决率 | 任务集 | 模型 |
+ PowerContext | 保存时间 | 来源 |
+
+
+
+ {baselines.map((baseline) => {
+ const path = `/report/${encodeURIComponent(baseline.source_batch_id)}`;
+ return (
+
+ | {baseline.name}{baseline.baseline_id} |
+ {baseline.source_arm.toUpperCase()} |
+ {baseline.resolved_tasks} / {baseline.total_tasks} |
+ {baseline.task_set} |
+ {baseline.model} · {baseline.reasoning_effort} |
+ {baseline.powercontext_sha?.slice(0, 12) ?? "—"} |
+ {new Date(baseline.created_at).toLocaleString("zh-CN", { hour12: false })} |
+ onLink(event, path)}>查看总体报告 |
+
+ );
+ })}
+
+
+
+
{taskSet === "swebench-pro-public-v2" ? "SWE-bench Pro public v2" : "稳定性回归 v1"}
- {taskSet === "swebench-pro-public-v2" ? "731" : "24"} 个任务，每个任务依次运行 OFF / ON
+
+ {taskSet === "swebench-pro-public-v2" ? "731" : "24"} 个任务 · {treatmentMode === "off_on"
+ ? "OFF + ON"
+ : treatmentMode === "on_only" ? "仅 ON" : "仅 OFF"}
+
{model} · medium
- Worker 按配置并行运行独立任务对
+ Worker 按配置并行运行独立任务
-
- {preview !== null && (
-
-
-
-
确认信息
-
{number(preview.total_tasks)} 个基准任务
-
-
- {preview.usage === null ? "API Key 计费" : `当前用量 ${preview.usage.used_percent}%`}
-
-
- - 任务集
- SWE-bench Pro public v2
- - 运行方式
- 每个任务 OFF / ON 配对执行
- - Codex 模型
- {preview.model} · {preview.reasoning_effort}
- - 暂停阈值
- {preview.usage === null ? "不适用" : `${preview.usage_pause_percent}%`}
- - 计量窗口
- {preview.usage === null ? "API Key" : formatUsageWindow(preview.usage.window_duration_minutes)}
- - 额度重置
- {preview.usage === null ? "由 Provider 管理" : dateTime(preview.usage.resets_at)}
- - 用量采样
- {preview.usage === null ? "不采集订阅用量" : dateTime(preview.usage.observed_at)}
-
-
- 剩余估算
- -
- {preview.estimate.quality === "unavailable"
- ? "暂无可靠估算"
- : `${preview.estimate.quality === "preliminary" ? "初步估算" : "已测量"} · ${preview.estimate.sample_size} 个样本`}
-
-
-
- {!preview.can_start && (
- 当前用量已达到暂停阈值
- )}
-
-
- )}
-
diff --git a/evaluation/web/src/components/BatchOverview.tsx b/evaluation/web/src/components/BatchOverview.tsx
index 4ac28b68d..b9c188742 100644
--- a/evaluation/web/src/components/BatchOverview.tsx
+++ b/evaluation/web/src/components/BatchOverview.tsx
@@ -18,7 +18,16 @@ import { useCallback, useEffect, useRef, useState, type MouseEvent } from "react
import type { EvaluationApi } from "../api";
import { batchStatusLabel } from "../batchStatus";
-import type { BatchRecord, BatchReport, PairCategory, TokenMetricAggregate } from "../types";
+import type {
+ Arm,
+ BaselineCandidate,
+ BaselineComparisonResponse,
+ BaselineSelection,
+ BatchRecord,
+ BatchReport,
+ PairCategory,
+ TokenMetricAggregate,
+} from "../types";
import { BatchControls } from "./BatchControls";
interface BatchOverviewProps {
@@ -49,6 +58,13 @@ function signed(value: number, suffix = ""): string {
export function BatchOverview({ api, batchId, navigate }: BatchOverviewProps) {
const [batch, setBatch] = useState(null);
const [report, setReport] = useState(null);
+ const [candidates, setCandidates] = useState([]);
+ const [selections, setSelections] = useState([]);
+ const [comparisons, setComparisons] = useState(null);
+ const [comparisonArm, setComparisonArm] = useState("on");
+ const [baselineName, setBaselineName] = useState("");
+ const [baselineMessage, setBaselineMessage] = useState("");
+ const [baselinePending, setBaselinePending] = useState(false);
const [error, setError] = useState(false);
const generation = useRef(0);
const controller = useRef(null);
@@ -62,11 +78,23 @@ export function BatchOverview({ api, batchId, navigate }: BatchOverviewProps) {
Promise.all([
api.getBatch(batchId, nextController.signal),
api.getBatchReport(batchId, nextController.signal),
+ api.listBaselineSelections(batchId, nextController.signal),
+ api.getBaselineComparisons(batchId, nextController.signal),
])
- .then(([nextBatch, nextReport]) => {
+ .then(([nextBatch, nextReport, nextSelections, nextComparisons]) => {
if (nextController.signal.aborted || currentGeneration !== generation.current) return;
setBatch(nextBatch);
setReport(nextReport);
+ setSelections(nextSelections);
+ setComparisons(nextComparisons);
+ const nextArm = nextBatch.request.treatment_mode === "off_only" ? "off" : "on";
+ setComparisonArm(nextArm);
+ return api.listBaselineCandidates(batchId, nextArm, nextController.signal);
+ })
+ .then((nextCandidates) => {
+ if (nextCandidates !== undefined && !nextController.signal.aborted && currentGeneration === generation.current) {
+ setCandidates(nextCandidates);
+ }
})
.catch(() => {
if (!nextController.signal.aborted && currentGeneration === generation.current) setError(true);
@@ -87,6 +115,56 @@ export function BatchOverview({ api, batchId, navigate }: BatchOverviewProps) {
navigate(path);
};
+ const refreshCandidates = async (arm: Arm) => {
+ setComparisonArm(arm);
+ try {
+ setCandidates(await api.listBaselineCandidates(batchId, arm));
+ } catch {
+ setBaselineMessage("当前无法读取兼容基线。");
+ }
+ };
+
+ const saveBaseline = async (arm: Arm) => {
+ if (report === null || batch === null || baselinePending) return;
+ const name = baselineName.trim() || `${batch.batch_id} ${arm.toUpperCase()}`;
+ setBaselinePending(true);
+ setBaselineMessage("");
+ try {
+ await api.createBaseline({
+ name,
+ source_batch_id: batch.batch_id,
+ source_arm: arm,
+ expected_report_revision: report.report_revision,
+ idempotency_key: `web-baseline-${Date.now().toString(36)}-${Math.random().toString(36).slice(2)}`,
+ });
+ setBaselineName("");
+ setBaselineMessage(`${arm.toUpperCase()} 基线已保存。`);
+ await refreshCandidates(comparisonArm);
+ } catch {
+ setBaselineMessage("基线保存失败，请刷新报告后重试。");
+ } finally {
+ setBaselinePending(false);
+ }
+ };
+
+ const toggleBaseline = async (baselineId: string, checked: boolean) => {
+ const retained = selections.filter(
+ (selection) => !(selection.baseline_id === baselineId && selection.current_arm === comparisonArm),
+ );
+ const next = checked ? [...retained, { baseline_id: baselineId, current_arm: comparisonArm }] : retained;
+ setBaselinePending(true);
+ setBaselineMessage("");
+ try {
+ const saved = await api.updateBaselineSelections(batchId, next);
+ setSelections(saved);
+ setComparisons(await api.getBaselineComparisons(batchId));
+ } catch {
+ setBaselineMessage("基线选择更新失败。");
+ } finally {
+ setBaselinePending(false);
+ }
+ };
+
if (error) {
return (
@@ -104,6 +182,11 @@ export function BatchOverview({ api, batchId, navigate }: BatchOverviewProps) {
const taskListPath = `/report/${encodedBatchId}/tasks`;
const progress = `${report.terminal_tasks} / ${report.total_tasks}`;
const status = batchStatusLabel[batch.status];
+ const paired = report.treatment_mode === "off_on" && report.off !== null && report.on !== null;
+ const singleArm = report.treatment_mode === "off_only" ? "off" : "on";
+ const singleResolution = singleArm === "off" ? report.off : report.on;
+ const pairCategories = report.pair_categories;
+ const comparablePairs = report.comparable_pairs;
return (
@@ -124,34 +207,48 @@ export function BatchOverview({ api, batchId, navigate }: BatchOverviewProps) {
-
-
-
+ {paired && report.off !== null && report.on !== null && report.resolution_rate_delta_points !== null ? (
+ <>
+
+
+
+
+ ) : singleResolution !== null ? (
+ <>
+
+
+
+
+ ) : null}
-
+ {paired && pairCategories !== null && comparablePairs !== null &&
实验对比结果
-
可比较任务 {number(report.comparable_pairs)} / {number(report.total_tasks)}
+
可比较任务 {number(comparablePairs)} / {number(report.total_tasks)}
onLink(event, taskListPath)}>查看全部任务
)}
-
+ }
@@ -183,11 +280,88 @@ export function BatchOverview({ api, batchId, navigate }: BatchOverviewProps) {
-
-
-
+ {paired ? (
+ <>
+
+
+
+
+ ) : (
+ <>
+
+
+
+
+ )}
+
+
+
+
+
+
+
历史基线对比
+
可选择多个基线；增删对比不会重新运行任务。
+
+ {batch.request.treatment_mode === "off_on" && (
+
+ )}
+
+ {candidates.filter((candidate) => candidate.compatibility.status !== "incompatible").length === 0 ? (
+
没有兼容基线。可以在下方先保存当前结果。
+ ) : candidates.filter((candidate) => candidate.compatibility.status !== "incompatible").map((candidate) => (
+
+ ))}
{comparison.baseline.name}{comparison.baseline.source_arm.toUpperCase()} → 当前 {comparison.current_arm.toUpperCase()}
+
+ - 基线解决率
- {percent(comparison.resolution.baseline_rate_percent)}
+ - 当前解决率
- {percent(comparison.resolution.current_rate_percent)}
+ - 差值
- {signed(comparison.resolution.delta_points, " pp")}
+ - 可比较任务
- {number(comparison.coverage.comparable_tasks)}
+
+
+ ))}
+ {comparisons !== null && comparisons.comparisons.length === 0 && (
+ 尚未选择基线；当前报告只展示本次真实结果。
+ )}
+
+ {batch.status === "completed" && (
+
+ )}
);
}
@@ -203,6 +377,13 @@ function MetricCard({ label, value, detail }: { label: string; value: string; de
}
function TokenCard({ label, metric, total }: { label: string; metric: TokenMetricAggregate; total: number }) {
+ if (
+ metric.off === null
+ || metric.on === null
+ || metric.delta === null
+ || metric.off_measured_tasks === null
+ || metric.on_measured_tasks === null
+ ) return null;
const measured = Math.min(metric.off_measured_tasks, metric.on_measured_tasks);
const deltaPercent = metric.off === 0 ? null : metric.delta / metric.off * 100;
return (
@@ -220,3 +401,25 @@ function TokenCard({ label, metric, total }: { label: string; metric: TokenMetri
);
}
+
+function SingleTokenCard({
+ label,
+ metric,
+ arm,
+ total,
+}: {
+ label: string;
+ metric: TokenMetricAggregate;
+ arm: Arm;
+ total: number;
+}) {
+ const value = arm === "off" ? metric.off : metric.on;
+ const measured = arm === "off" ? metric.off_measured_tasks : metric.on_measured_tasks;
+ return (
+
+ );
+}
diff --git a/evaluation/web/src/components/ContextTimeline.tsx b/evaluation/web/src/components/ContextTimeline.tsx
index 6c96538a6..70dbd412f 100644
--- a/evaluation/web/src/components/ContextTimeline.tsx
+++ b/evaluation/web/src/components/ContextTimeline.tsx
@@ -17,19 +17,20 @@
import { useEffect, useRef, useState } from "react";
import type { EvaluationApi } from "../api";
-import type { ContextEvent } from "../types";
+import type { Arm, ContextEvent } from "../types";
interface ContextTimelineProps {
api: EvaluationApi;
batchId: string;
taskId: string;
attemptId?: string;
+ availableArms?: Arm[];
}
const PAGE_SIZE = 200;
-export function ContextTimeline({ api, batchId, taskId, attemptId }: ContextTimelineProps) {
- const [arm, setArm] = useState<"off" | "on">("on");
+export function ContextTimeline({ api, batchId, taskId, attemptId, availableArms = ["off", "on"] }: ContextTimelineProps) {
+ const [arm, setArm] = useState