Update public model card

README.md

@@ -0,0 +1,172 @@
+---
+language: []
+license: other
+license_name: aion-edge-engine-license
+pipeline_tag: tabular-classification
+tags:
+- finance
+- trading
+- market-intelligence
+- indian-stock-market
+- risk-management
+- nifty
+- banknifty
+- sensex
+- indian-index-options
+- market-microstructure
+- trade-filter
+- evidence-based-trading
+- market-regime
+- pre-trade-gate
+- noise-filter
+- institutional-trading
+- nse
+- fno
+- commodity-trading
+- crude-oil
+- calibrated-confidence
+datasets: []
+widget: []
+---
+
+# AION Edge Engine — Market Intelligence Layer
+
+A pre-trade evidence filter for Indian index options, commodities, and crude oil.
+Edge Engine does **not** tell you what to trade. It answers a narrower question:
+**does the current market state carry meaningful evidence, or should you stand
+down?**
+
+## 1. Why This Exists
+
+Indian index options move fast. A short-horizon market state can look like a
+breakout, a reversal, or a trap, and mistaking noise for signal is one of the
+most expensive mistakes a trader can make.
+
+Edge Engine exists to reduce that mistake. It is designed for calibrated trust:
+it can stay silent when evidence is insufficient, and when it does produce a
+tradability judgment, it exposes audit fields instead of hiding everything
+behind one opaque score.
+
+## 2. What Edge Engine Does
+
+Edge Engine converts intraday market state for NIFTY, BANKNIFTY, SENSEX,
+commodities, and crude oil into a structured pre-trade judgment:
+
+- **Tradable or not**, based on the current evidence profile
+- **Expected value context** and **confidence-bound telemetry**, exposed
+  separately
+- **Underlying-specific routing**, because NIFTY, BANKNIFTY, SENSEX, and crude
+  oil are different markets
+- **Audit-ready output**, preserving raw and enforced tradability fields
+
+It does not output a brokerage instruction. It outputs context. The difference
+matters.
+
+## 3. Output Contract
+
+Every public field is intended to be auditable. No model weights, training data,
+or proprietary model internals are distributed through this repository.
+
+```json
+{
+  "edge_underlying": "nifty",
+  "edge_ev": 0.1421,
+  "edge_conf_low": 0.0314,
+  "edge_conf_high": 0.0471,
+  "edge_tradable": true,
+  "edge_tradable_raw": true,
+  "edge_motive": "model",
+  "edge_model_key": "model:edge:nifty",
+  "edge_shadow_mode": false,
+  "edge_expiry_days": 7.0
+}
+```
+
+| Field | Meaning |
+|---|---|
+| `edge_underlying` | Instrument family evaluated by the engine |
+| `edge_ev` | Expected-value context surfaced by the model |
+| `edge_conf_low` | Lower confidence-bound telemetry |
+| `edge_conf_high` | Upper confidence-bound telemetry |
+| `edge_tradable` | Enforced tradability after runtime policy gates |
+| `edge_tradable_raw` | Raw model tradability preserved for audit |
+| `edge_motive` | Why the judgment was produced, such as model or policy context |
+| `edge_model_key` | Underlying-specific model key used for inference |
+| `edge_shadow_mode` | Whether the output is live-enforced or shadow/paper-validation |
+| `edge_expiry_days` | Days-to-expiry context at inference time, when applicable |
+
+Raw tradability and enforced tradability are both preserved so integrators can
+audit when runtime policy differs from the raw model judgment.
+
+## 4. Why Market-State Filtering Matters
+
+In Indian index options and commodity derivatives, very short-horizon market
+states can be distorted by spread changes, expiry behavior, auction spillover,
+and reaction loops. Edge Engine is designed around market-state evidence rather
+than urgency. When evidence is weak, the correct output may be to stand down.
+
+That restraint is the product: it helps suppress low-quality entries before they
+reach execution logic.
+
+## 5. Intended Uses
+
+- **Pre-trade filtering:** run Edge Engine before existing entry logic to
+  suppress signals during low-evidence market states
+- **Regime awareness:** monitor expected-value and confidence-bound telemetry to
+  understand when conditions are improving or degrading
+- **Audit and post-trade analysis:** compare raw and enforced tradability
+  decisions against later outcomes
+- **Paper validation:** use shadow-mode outputs to evaluate behavior before live
+  enforcement
+
+## 6. Out-of-Scope Uses
+
+- Blind execution as a standalone trade signal
+- Exact price prediction or directional forecasting
+- Compliance, suitability, or investment-advice decisions
+- Markets outside the supported Indian-market scope
+- Sub-market-state scalping or ultra-low-latency execution
+
+## 7. Limitations
+
+- Edge Engine scores contextual tradability; it does not predict exact direction
+  or magnitude.
+- Sudden structural breaks may not be fully reflected until the current market
+  state is observed by the system.
+- Historical relationships can weaken during regime shifts, regulatory changes,
+  or liquidity shocks.
+- Non-index coverage can carry wider confidence bounds than the most liquid
+  index markets.
+- Shadow-mode outputs are for paper validation and audit, not live decision
+  authority.
+
+## 8. Deployment, Access, and Citation
+
+AION Edge Engine is an API-first system. Model weights are not publicly
+distributed. Production access is through the managed AION API.
+
+- API endpoint: https://api.aiondashboard.site/v1/edge
+- Dashboard: https://dashboard.aiondashboard.site/models/edge-engine
+- Registration: https://dashboard.aiondashboard.site/access/register
+- API gateway: https://dashboard.aiondashboard.site/systems/api-gateway
+- Organization: https://huggingface.co/AION-Analytics
+
+Freemium access includes selected AION market-intelligence surfaces. Edge Engine
+and broader multi-index access are available through API subscription. Requests
+are metered; over-quota requests are rejected by the managed API.
+
+If you integrate Edge Engine, monitor:
+
+- Whether stand-down decisions avoided chop, reversals, or false breakouts
+- Whether tradable decisions align with your own entry criteria
+- Differences between `edge_tradable_raw` and `edge_tradable`
+- Confidence-bound behavior over time
+
+```bibtex
+@software{aion_edge_engine_2026,
+  author = {AION Analytics},
+  title = {AION Edge Engine — Market Intelligence Layer for Indian Index Options},
+  year = {2026},
+  url = {https://huggingface.co/AION-Analytics/aion-edge-engine}
+}
+```