# TFM aInventory - Scanner Technical Specification (v1.2)

This document defines the "Frozen" parameters for the scanner component. Do NOT modify these values without explicit user approval.

## 1. Hardware Access
- **API**: Direct `MediaStreamTrack` access for high-performance camera control.
- **Zoom**: Dynamic detection of `capabilities.zoom.max`.
- **Zoom Cycle**: 1x -> 2x -> Max/2 -> Max.

## 2. Image Pre-processing (The OCR "Secret Sauce")
Before being sent to Tesseract.js, every frame undergoes these transformations:
- **Rescaling**: Fixed canvas width of **1200px** (Optimal for Mobile RAM/Accuracy).
- **Cropping**: **60% Center Crop** of the video stream (forces focus & removes peripheral noise).
- **Filters**: `grayscale(100%) contrast(180%) brightness(105%)`.
- **Format**: High-quality JPEG (`0.85` quality) via `toDataURL`.

## 3. OCR Matching Engine (page.tsx)
- **Sanitization**: Text normalized to UPPERCASE, non-alphanumeric replaced with spaces.
- **Noise Filtering**:
    - Tokens < 3 chars ignored.
    - Decimals (e.g., `0.11`, `0.18`) ignored via regex: `/^\d+\.\d+$/`.
    - Dates (e.g., `2024-07-25`) ignored via regex: `/^\d{2,4}-\d{2}-\d{2}$/`.
- **Scoring System**:
    - **Exact Serial Number Match**: +500 points (Instant Match).
    - **Exact Part Number Match**: +200 points.
    - **Token match (e.g., "LC" in "LC/UPC")**: +50 points.
    - **Category Match**: +20 points.
- **Threshold**: Minimum **40 points** required for automatic match without user intervention.

## 4. UI/UX Flow
1. **Auto-Match**: Attempt to identify item immediately from raw text.
2. **Interactive Selection (Fallback)**: If match confidence is low, freeze frame and show clickable bounding boxes for manual word selection.

---
*Created: 2026-04-10*
*Status: FROZEN*