fix(inference): default scenario to 8K/1K instead of Agentic Traces / 默认场景改为 8K/1K，不再是 Agentic Traces

[functionstackx]

Summary

• /inference (and any page without an explicit ?i_seq=) now defaults the Scenario selector to 8K / 1K instead of Agentic Traces: the selectedSequence default in GlobalFilterContext.tsx flips from Sequence.AgenticTraces to Sequence.EightK_OneK.
• Everything else is unchanged by design:
  • Explicit ?i_seq= share links still win (including agentic-traces).
  • Per-route initialSequence seeds (/compare pages) still win.
  • The availability fallback ordering in resolveEffectiveSequence (keep selection → prefer 8k/1k → first available) is untouched, so agentic-only models still resolve to their agentic scenario.
  • PARAM_DEFAULTS.i_seq stays '' (not a strippable '8k/1k'): per-route initialSequence seeds make the no-param resolution route-dependent, so stripping an explicit 8K/1K pick would revert it to the route's seeded scenario on reload.
• Updated all comments that justified behavior by the old agentic-preferred default (default-sequence.ts, InferenceContext.tsx, url-state.ts, chart-selectors.tsx, test comments). The scenario dropdown's group order (Agentic listed first) is unchanged — display order only.

Tests

• New e2e: bare /inference defaults to 8K / 1K even when the model has agentic data (ttft-x-axis-toggle.cy.ts) — availability contains both agentic and fixed-seq rows, and the selector must still land on 8K/1K.
• The agentic x-axis e2e suites previously relied on the bare /inference default resolving to agentic; they now pin ?i_seq=agentic-traces explicitly (including the overlay-path suite).
• Verified locally: pnpm typecheck, pnpm lint, pnpm fmt, pnpm test:unit (2560 passed), and e2e specs ttft-x-axis-toggle (13/13), inference-chart, gpu-compare-agentic-detail, agentic-point-time-series against an E2E_FIXTURES=1 server.

Overlay support

Works for both official runs and ?unofficialrun= overlays: the default applies before any data loads, and unofficial-run availability still participates in availableSequences/auto-switch exactly as before. The overlay-path e2e suite passes with the explicit i_seq=agentic-traces pin.

中文说明

• /inference（以及所有未携带 ?i_seq= 参数的页面）的 Scenario 选择器默认值从 Agentic Traces 改为 8K / 1K：GlobalFilterContext.tsx 中 selectedSequence 的默认值由 Sequence.AgenticTraces 改为 Sequence.EightK_OneK。
• 其余行为刻意保持不变：
  • 显式 ?i_seq= 分享链接仍然优先（包括 agentic-traces）。
  • /compare 页面的 initialSequence 预设仍然优先。
  • resolveEffectiveSequence 的可用性回退顺序（保留用户选择 → 优先 8k/1k → 第一个可用项）未改动，因此仅有 agentic 数据的模型仍会解析到 agentic 场景。
  • PARAM_DEFAULTS.i_seq 保持为 ''：由于各路由的 initialSequence 预设不同，若把 '8k/1k' 作为可剥离默认值，显式选择 8K/1K 后重新加载会回退到路由预设场景。
• 更新了所有以"agentic 优先默认"为依据的代码注释；场景下拉菜单的分组顺序（Agentic 组在前）不变，仅为显示顺序。

测试：新增 E2E 用例验证即使模型有 agentic 数据，裸 /inference 默认仍为 8K/1K；原依赖默认 agentic 的 E2E 套件改为显式指定 ?i_seq=agentic-traces（含 unofficial-run overlay 套件）。本地已通过 pnpm typecheck、pnpm lint、pnpm fmt、pnpm test:unit（2560 个用例）及相关 E2E 测试。

🤖 Generated with Claude Code

---

Note

Medium Risk
Default scenario affects first paint, benchmark fetches, and share-link behavior on /inference, though explicit URL/route seeds and availability resolution are preserved; regression risk is mainly UX and wrong data path on load.

Overview
Bare /inference (no ?i_seq=) now opens on 8K / 1K instead of Agentic Traces. The main behavioral change is in GlobalFilterContext: selectedSequence initializes to Sequence.EightK_OneK rather than Sequence.AgenticTraces. Explicit ?i_seq=, per-route initialSequence, and resolveEffectiveSequence fallback rules are unchanged.

Comments and docs across default-sequence.ts, InferenceContext, url-state, and chart-selectors now describe 8K/1K as the app default (Agentic still listed first in the dropdown for display only). PARAM_DEFAULTS.i_seq stays empty so share URLs keep explicit scenario picks.

E2E: Agentic x-axis and overlay suites visit with ?i_seq=agentic-traces instead of relying on the old default. A new test asserts bare /inference shows 8K / 1K even when availability includes both agentic and fixed-seq data.

^{Reviewed by Cursor Bugbot for commit 014bf30. Bugbot is set up for automated code reviews on this repo. Configure here.}

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
inferencemax-app	Ready	Preview, Comment	Jul 10, 2026 7:35pm