# add-calc

> Adds new CNC machining calculation types (pocket, slot, chamfer, etc.) to the manufacturing time estimator.

- Author: ductrantrong
- Repository: ductrantrong/cnc-time-estimator
- Version: 20260126164838
- Stars: 0
- Forks: 0
- Last Updated: 2026-02-06
- Source: https://github.com/ductrantrong/cnc-time-estimator
- Web: https://mule.run/skillshub/@@ductrantrong/cnc-time-estimator~add-calc:20260126164838

---

---
name: add-calc
description: Adds new CNC machining calculation types (pocket, slot, chamfer, etc.) to the manufacturing time estimator.
---

# Adding a New Calculation Step

Guide for adding new machining calculation types to the CNC manufacturing time estimator.

## Decision Tree

1. **Parameter source?** Image (drawing) / Notes / Both
2. **Calculation type?** Simple math → Pure Python | Complex reasoning → LLM-based
3. **Tool types?** DRILL, ENDMILL, TAP, FACE, REAMER from `BANG_CHE_DO_CAT.csv`

## File Modification Checklist

Modify **10 file categories** (14+ actual files) in order:

| # | Category | Path | Type |
|:--|:---------|:-----|:-----|
| A | Extractor | `backend/extractors/{feature}.py` | NEW |
| B | Calculator | `backend/core/{feature}_calc.py` | NEW |
| C | Pipeline | `backend/analysis/_extractors/{feature}.py` | NEW |
| D | Facade Export | `backend/analysis/extractors.py` | EDIT |
| E | Registry | `backend/analysis/registry.py` | EDIT (1 line) |
| F | Tools Facade | `backend/tools.py` | EDIT |
| G | Formula Doc | `docs/{feature}.md` | NEW |
| H | Prompt Partial | `prompts/partials/_{feature}.jinja` | NEW |
| I | Documentation | 4× CLAUDE.md files | EDIT |
| J | Debug-Logs | `.claude/skills/debug-logs/scripts/*.py` | EDIT |

**Note**: streaming.py no longer needs modification - the Feature Registry handles it automatically.

## Quick Links

- [Extractor patterns](extractor.md) - LLM/regex extraction from images/notes
- [Calculator patterns](calculator.md) - Pure Python and LLM-based formulas
- [Pipeline integration](pipeline.md) - Connecting extractor + calculator + registry
- [Prompt templates](prompt-partial.md) - Jinja partials for LLM display
- [Complete checklist](checklist.md) - Full implementation checklist

## Feature Registry Pattern

Adding a new calculation now uses a **data-driven registry** instead of manual streaming.py edits.

### Adding a New Feature (Minimal Steps)

1. Create extractor, calculator, pipeline (steps A-C above)
2. Export from facade (step D)
3. Add ONE line to `backend/analysis/registry.py`:

```python
FeatureConfig(name="feature", extractor=extract_feature_and_calculate),
```

**No changes needed in streaming.py** - the registry loops handle submission, error collection, merging, and yielding automatically.

### FeatureConfig Options

```python
@dataclass(frozen=True)
class FeatureConfig:
    name: str                           # e.g., "tap", "drill"
    extractor: Callable                 # extract_X_and_calculate function
    requires_client: bool = True        # False for regex-only (no LLM)
    merge_into: Optional[str] = None    # Merge results into another feature
    skip_yield: bool = False            # Don't yield to frontend
    skip_use_flag: bool = False         # Don't set use_{name} in config
    skip_info: bool = False             # Don't set {name}_info in config
```

### Special Case Examples

| Feature | Behavior | Registry Config |
|:--------|:---------|:----------------|
| Standard | Default (yield, use_flag, info) | `FeatureConfig(name="tap", extractor=extract_tap_and_calculate)` |
| `face` | No use_flag or info stored | `skip_use_flag=True, skip_info=True` |
| `lathe_drill` | Regex-only, merges into drill | `requires_client=False, merge_into="drill", skip_yield=True, skip_use_flag=True, skip_info=True` |

## Config Key Naming Convention

```python
config["use_{feature}"] = True                # Boolean flag
config["{feature}_info"] = [...]              # Extracted data (List[Dict])
config["{feature}_calculated"] = {...}        # Calculation results
```

## Return Dict Structure (Required)

All calculators must return this exact structure:

```python
{
    "tool_info": {
        "tool_name": {"name": str, "S": int, "F": float, "AP": float}
    },
    "calculations": {
        "param1": value,
        "step1": {"formula": str, "substitution": str, "time_minutes": float}
    },
    "total_time_minutes": float,  # Rounded to 4 decimals
    "summary": "Vietnamese summary"
}
```

## Extraction Types

| Type | When to Use | Example |
|:-----|:------------|:--------|
| LLM-based | Complex patterns, context-dependent | TAP, precision holes |
| Regex-based | Simple patterns, fixed format | C via L (`C via L=50`), vát đĩa (`Gia công n vát đĩa`) |
| Notes-only | Explicit production notes | Drill, pocket |
| Image-only | Geometric features | Normal holes |

## Calculation Complexity

| Type | Tool Lookup | Material-Dependent | Example |
|:-----|:------------|:-------------------|:--------|
| Full | Yes (database) | Yes | Drill, pocket, counterbore |
| Partial | No | Yes (F constant varies) | C via L |
| Simple | No | No (all constants fixed) | Vát đĩa |

## Key Rules

- Always return `List[Dict]` from extractors
- Always validate NaN and positive values in calculators
- Get tool parameters from database **OR** use hardcoded constants (for simple formulas like C via L)
- Always add prompt.jinja section (or results won't appear in final analysis)
- Always update all 4 CLAUDE.md files