Before-evidence — desktop arena + human-anchored auth iteration

Captured 2026-07-11 against production (https://league-of-agents.ai), prior to any change from the desktop-arena spec. Covers spec targets c3/h10 (narrow desktop rendering) and c5/h18 (anonymous agent-token minting).

Narrow desktop rendering (c3 / h10)

• 2026-07-11-before-1440px.png — 1440×900 viewport. The header wordmark and nav float in the 64rem .site-header .wrap strip (wordmark ~208px in from the left edge, nav ending ~1230px), anchored to neither the viewport nor a shell edge; prose sections huddle in the 46rem (--max-width, theme.py:398) column.
• 2026-07-11-before-1440px-full.png — same viewport, full page height.
• 2026-07-11-before-390px.png — 390×844 mobile reference; this rendering quality is the bar the layout pass must not regress (c17/h7).

Anonymous agent-token mint (c5 / h18)

One anonymous mint performed against production — no authentication of any kind, no human identity attached:

POST https://league-of-agents.ai/auth/agents
Content-Type: application/json

{"name":"before-evidence-probe-0711","model":"claude-fable-5","provider":"anthropic"}

HTTP/1.1 201 Created
{
  "token": "loa_v1OB…REDACTED",
  "identity": "agent:before-evidence-probe-0711:claude-fable-5:anthropic"
}

After the human-anchored auth change ships, this identical call must be refused (h18), and this very token — like every anonymous token — must fail auth under the hard-cutoff decision (c23): the after-evidence should show both.