Platform Triage — render screenshots¶

UI render screenshots of the Platform Triage page (/static/triage.html in duecare-llm-chat), captured 2026-06-20 with Playwright driving system Edge, after the one-model simplification (commit ddad750d).

Desktop (1440) — with a screened batch¶

Platform Triage page rendered at 1440px, showing a screened sample batch

Tablet (768)¶

Platform Triage page rendered at 768px

Mobile (375)¶

Platform Triage page rendered at 375px, single column with a collapsed nav

What they show: the simplified two-stage design — Stage 1 GREP rules → Stage 2 the one loaded Gemma 4 (no separate fast model, no endpoint, no model switch), the trust boundary, the input/results panes, the activity log, and the responsive layout.

Honest caveat (real, not faked): these verify the page — layout, JavaScript, routing flow, and chrome are the real production page. No GPU model was loaded locally when they were captured, so the screening model and GREP layer were stubbed deterministically for the render. The per-item verdicts, categories, and "flagged by model" labels are therefore illustrative, not real Gemma 4 inference. Real screening runs on the loaded Gemma in the kernel; see packages/duecare-llm-chat/tests/test_triage.py for the tested behaviour.