How mrphilo420/burnwatch works
Burnwatch screens submitted text against human-calibrated conformal rules and returns an alert, non-alert, or referral for human review. The web path accepts text or uploaded documents, prepares text, obtains calibration/scoring results, and records privacy-preserving job summaries; the CLI exposes the same screening core. Separate from serving, the empirical study evaluates constructions across corpora. The graph also includes the legacy perturbation experiment pipeline. Some source excerpts are partial, so edges are limited to relationships shown in excerpts, the SOURCE INDEX, or README.
Main entry points: app.py Flask interface · conformal.py Conformal screening · data.py Corpus access · study.py Empirical study
Components 13
Web application
- Flask interface
app.py - Screening UI
templates/index.html - File extraction
docutils.py - Decision audit
audit.py
Screening core
- Conformal screening
conformal.py - Calibration bundles
conformal.py - Style markers
slop.py
Data and evaluation
- Corpus access
data.py - Custom datasets
custom_datasets.py - Empirical study
study.py
Research pipeline
- Perturbation experiments
run.py
Other components
- Reviewer
- Dataset sources
Connections 13
- Reviewer → Screening UI: submits text
README.md - Screening UI → Flask interface: uses
README.md - Flask interface → File extraction: extracts uploads
app.py - Flask interface → Conformal screening: screens text
app.py - Flask interface → Calibration bundles: loads bundle
app.py - Flask interface → Decision audit: records jobs
app.py - Conformal screening → Calibration bundles: uses bundle
conformal.py - Conformal screening → Style markers: uses detector
conformal.py - Conformal screening → Corpus access: loads data
conformal.py - Empirical study → Conformal screening: evaluates constructions
study.py - Empirical study → Corpus access: loads records
study.py - Corpus access → Dataset sources: fetches datasets
data.py - Perturbation experiments → Custom datasets: loads datasets
run.py