chore(e2e): refresh Cypress fixtures so MiniMax M3 appears in dropdown

[adibarra]

Note

Low Risk
Test-only URL and comment changes; no production code paths affected.

Overview
After Cypress fixtures made the default inference model (DeepSeek-V4-Pro) FP4-only, two e2e visits that relied on the default model no longer exercised the precision behavior under test.

line-labels.cy.ts now opens /inference with g_model=DeepSeek-R1-0528 alongside i_linelabel=1 and i_prec=fp4,fp8, so both precisions survive effectivePrecisions and the spec can assert dual-precision line labels.

url-params.cy.ts uses the same model for i_prec=fp8, so the precision multiselect shows FP8 instead of falling back to FP4. Comments document why the default model is insufficient.

^{Reviewed by Cursor Bugbot for commit ca7c284. 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	Jun 14, 2026 7:43am

[adibarra]

@claude keep fixing the cypress tests until they fully pass without issues

[github-actions]

Claude finished @adibarra's task in 6m 58s —— View job

---

Fixed failing Cypress tests

• Gather context on failing tests
• Investigate root causes
• Implement fixes
• Run E2E tests locally to verify (37/37 passing)
• Commit and push

Root cause

PR #435 changed the default inference model to DeepSeek-V4-Pro (dsv4). The refreshed availability fixture lists dsv4 with FP4 only at both 1K/1K and 8K/1K sequences. When the failing tests visited /inference?i_prec=fp8…, GlobalFilterContext's effectivePrecisions intersected the URL-seeded ['fp8'] against the available set ['fp4'], producing an empty result and falling back to FP4. The precision-multiselect consequently rendered "FP4" instead of "FP8", and no FP8 line label was drawn.

The sibling test …with multiple URL params seeds all of them… already worked around this by passing g_model=DeepSeek-R1-0528 alongside i_prec=fp8. I applied the same pattern — minus the fixture refresh — to the two failing specs.

Changes (commit 3c73c14)

• packages/app/cypress/e2e/url-params.cy.ts:188 — /inference?i_prec=fp8 seeds the precision… now visits with g_model=DeepSeek-R1-0528&i_prec=fp8.
• packages/app/cypress/e2e/line-labels.cy.ts:82 — appends the precision to each line label… now visits with g_model=DeepSeek-R1-0528&i_linelabel=1&i_prec=fp4,fp8.

A short code comment in each test explains why the model is being pinned.

Local verification

✔  url-params.cy.ts                         00:55       29       29
✔  line-labels.cy.ts                        00:08        8        8
✔  All specs passed!                        01:04       37       37

| Branch: claude/pr-456-20260614-0714

[adibarra]

@claude linter/formatter failing fix

[github-actions]

Claude finished @adibarra's task in 2m 5s —— View job

---

Linter/formatter fixed

• Run lint + format checks to identify failures
• Apply fixes
• Verify clean
• Commit and push

Result

• pnpm lint → already clean (exit 0)
• pnpm fmt → 3 fixture files had format drift:
  • packages/app/cypress/fixtures/api/benchmarks-history.json
  • packages/app/cypress/fixtures/api/benchmarks.json
  • packages/app/cypress/fixtures/api/workflow-info.json
• Ran pnpm fmt:fix; re-ran both checks → both now pass.

Committed as ca7c284 and pushed to claude/pr-456-20260614-0714.
| Branch: claude/pr-456-20260614-0714