fix: add robust regex json extraction in LangGraph agents to prevent JSONDecodeError from trailing LLM output

feat: remove replay topology button, enforce strict DB event fetching to stop hallucination, bump chronos num_samples to 100

feat: hardcode 3-part semantic risk narrative in agent pipeline using dynamic dashboard values

feat: force LLM to output highly detailed, semantic 3-part operational briefing for summary rationale

fix: make forecast window text dynamic based on actual horizon selected

feat: overhaul seasonality analysis — ramp dynamics, load composition, demand heatmap, and LLM synthesis

fix: resolve undefined variable seasonal_risk in nodes.py

fix: explicit risk/reward definitions, correct seasonality prompt & parse, and PDF print UI enhancements

feat: reorder rationale paragraphs, remove infrastructure risk UI, and request longer LLM summary

feat: add individual chronos/sarima charts, math formula to severity, and punchier LLM summary

refactor: rewrite LLM prompts to be ultra-lean and hallucination-free

refactor: simplify LLM prompts to only receive quantitative rationale summary

feat: restructure rationale generation to use static math paragraphs and LLM summary

fix: resolve NameError for ANOMALY_THRESHOLD and pass threshold from UI state properly

fix: resolve prompt hallucination, incorrect math display, print text logic, and demo data sentences

fix: resolve PDF tabs, print background color, state MW drops, and prompt run-on sentences

fix: resolve UI output rendering and metrics (PDF buttons, NA metrics, prompt formatting, duplicated season analysis)

feat: enforce 4-part metric explanation rule (definition, significance, calculation, interpretation) in all LLM prompts + demo data

feat: upgrade LLM to Llama 4 Maverick, overhaul prompts with anti-slop rules, fix event card labels, enrich demo data

refactor: strip 6 wasteful LLM fields (reasoning_trace, key_signals, risk_factors, position_size, contract_type, stop_loss_trigger) from prompts and remove Contract Phase + Semantic Query from UI

fix: complete prompt rewrite to eliminate LLM slop — structural hints instead of embedded instructions, temperature 0.4, full data context for conservative node

feat: revamp anomaly severity formula and add MW metrics/formulas to UI

fix: resolve CSS rationale text wall and replace prompt structure to physically stop LLM looping

fix: enforce strict density constraint to prevent LLM looping and force explicit mathematical definition of Model Divergence

fix: JSONDecodeError by converting multi-paragraph AI generations to JSON arrays of strings to prevent unescaped newline termination

fix: total overhaul of agent prompts to enforce strict physics-grounded metric explanation and eliminate all AI filler

fix: forbid robotic glossary definitions in AI prompts and force natural narrative weaving

fix: radically expand rationale prompts to force deep metric explanation

fix: enforce strict negative prompt constraints to eliminate AI slop, and fix Dashboard severity threshold scaling mismatch

feat: AI-generated dynamic historical reasoning and clear Model Divergence definitions

perf: completely rewrite AI prompts with advanced personas and remove inline f-string logic

perf: force deep semantic generation using advanced grid terminologies

perf: enable Chain-of-Thought reasoning by forcing an internal monologue before json output, and quadruple max_tokens to leverage Groq limits

feat: prompt engineer the AI to speak in plain English control-room terminology instead of spitting raw numbers

feat: finetuned Chronos-T5-Base integrated and verified — WAPE improvement over SARIMA

Feat/Chronos-2 Integrated,Tests added,Finetuned

UI added and fixed

7 nodes implementation done