Fix alignment, remove numbers, remove left shadows

Update reasoning traces, emojis, and styling

Redesign reasoning footers from developer traces to XAI causal explanations

Update relaxed match model cards to use standard white background with a subtle blue border for contrast

Replace all unicode checklist symbols with sleek inline Lucide SVGs for consistent UI

Set risk label tag background to explicit white so it stays white even on colored relaxed result cards

Change risk label to match the top-bar-tag styling with a small monospace font and transparent border

Fix badge overflow by using flex wrap and allowing text to break words inside the risk label badge

Replace OKF confidence % with highlighted risk label in model card headers

Revert model card redesign to the original stable HTML box format

Fix OKF and RAG prompt layouts to completely remove confidence percentage and inject SVGs; fix Agentic CSV returning Unknowns by explicitly requesting SELECT * in SQL prompt

Remove confidence percentage score and add dynamic provider logos (Meta, Google, Microsoft, Anthropic, OpenAI, etc) to model cards

Fix model card overflow issues by improving flex wrapping and word breaking

Redesign model cards to exactly match the provided visual mockup using the app's native color palette

Upgrade model cards UI to premium research aesthetic using Lucide icons

Update reasoning pipelines to 4-phase user-facing logs and add reasoning summaries

Unify fallback OKF text formatting with normal OKF boxes

Add semantic fallback logic for 0-result queries

Format RAG and Agentic RAG outputs into model cards

Use native title attributes for tooltips to avoid clipping

Generate rich HTML semantic cards for OKF output

Fix OKF hallucination by wiring new filters to LLM prompt and traversal loop

Format OKF results output cleanly

Update OKF LLM prompt to include detailed risk fields

Backend updates: replace em dashes in logs with hyphens and colons

Genuine OKF: ontology-driven graph traversal with scoring and ranking

Add few-shot examples to filter prompt + log active filters in UI

Bulletproof JSON parsing: brace-counting extractor + graceful fallback

Rewrite OKF lane: deterministic server-side filtering eliminates hallucination

Refine codebase: fix encoding bugs

Strip trailing hallucinations

Add strict anti-hallucination formatting rules to OKF prompt

Strip out thinking text from OKF output

Add Chain of Thought to OKF prompt for better accuracy

Upload folder using huggingface_hub

Upload folder using huggingface_hub

Upload folder using huggingface_hub

Upload folder using huggingface_hub

Add HF_TOKEN explicit check

Upload folder using huggingface_hub