Component 4 (C4)

Local-first, open-source extraction pipeline for TradingView-like chart screenshots.

What is implemented

• Frame/axis localization, axis OCR (optional Paddle), calibration
• Elements detection, candle alignment, scenario reconstruction
• Final signal build (direction, entry, stop, targets, confidence)
• JSON output schema (Pydantic), YAML config, abstain plumbing
• Debug artifacts (overlay.png, axis_debug.png, plot_crop.png, axis_crop.png, elements_debug.png)

Quickstart

python -m pip install -e .
pytest -q
c4 extract --help

Single image run

c4 extract --image tests/assets/synth.png --out_json out.json --debug_dir debug --config configs/default.yaml

Batch runner

Use the helper script to run one image or many images and write deterministic outputs:

python scripts/run_c4.py --image chart1.jpeg --debug-dir runs/_debug --out-dir runs/_out
python scripts/run_c4.py --dir runs --glob "*.jpeg" --debug-dir runs/_debug --out-dir runs/_out

Per-image outputs:

• runs/_out/<image_stem>/output.json
• runs/_debug/<image_stem>/... debug PNG artifacts

Evaluation suite rerun (Windows-friendly)

Run the formal suite over test-images with one command. Each execution creates a fresh timestamped folder:

python scripts/eval_c4_suite.py

Outputs are written under:

• runs_eval/run_<YYYYMMDD_HHMMSS>/suite_summary.{md,json,csv}
• runs_eval/run_<...>/flagged/by_issue_code.{md,json}
• runs_eval/run_<...>/determinism/comparisons.{md,json}
• runs_eval/run_<...>/images/<stem>/artifacts and eval

Latest run pointer:

• runs_eval/latest.txt

Custom run name:

python scripts/eval_c4_suite.py --run-name run_after_determinism_fix

Output schema (top-level)

• chart_frame
• axis_ticks
• elements
• scenario
• signal
• abstain
• abstain_reasons
• debug_artifacts