Makefile

CHALLENGE ?= ../mib-doc-challenge
PY        := ./.venv/bin/python
IMAGE     := mib-submission
OUT       := /tmp/mib-output

.DEFAULT_GOAL := help
.PHONY: help build policy score validate clean

help:
	@echo "Targets:"
	@echo "  build     build the linux/amd64 submission image"
	@echo "  policy    cross-validate the adjudication cascade on ground-truth fields"
	@echo "  score     run the image over the training set and score it"
	@echo "  validate  check predictions.jsonl against the submission format"
	@echo "  clean     remove build and output artifacts"

# --platform is load-bearing: this builds on arm64 but the graders run x86_64,
# and an arm64-only image is a zero.
build:
	docker build --platform linux/amd64 -t $(IMAGE) .

policy:
	$(PY) scripts/fit_policy.py --labels $(CHALLENGE)/data/train_labels.csv

# Mirrors the scoring invocation in DOCKER_SUBMISSION.md, limits included.
score: build
	mkdir -p $(OUT)
	docker run --rm --network none \
	  --cpus 4 --memory 8g --pids-limit 512 \
	  --read-only --tmpfs /tmp:rw,nosuid,nodev,size=2g \
	  --mount type=bind,src="$(abspath $(CHALLENGE))/data/train",dst=/input,readonly \
	  --mount type=bind,src="$(OUT)",dst=/output \
	  $(IMAGE) /input /output/predictions.jsonl
	$(MAKE) validate
	python3 $(CHALLENGE)/scripts/evaluate.py \
	  --truth $(CHALLENGE)/data/train_labels.csv \
	  --submission $(OUT)/predictions.jsonl \
	  --output-json $(OUT)/evaluation.json \
	  --case-scores-jsonl $(OUT)/case_scores.jsonl

# Hard gate: a format violation scores zero, so this runs on every scored build,
# not once at the end.
validate:
	python3 $(CHALLENGE)/scripts/validate_submission.py \
	  --submission $(OUT)/predictions.jsonl \
	  --manifest $(CHALLENGE)/data/train_labels.csv

clean:
	rm -rf out
	find . -name __pycache__ -type d -prune -exec rm -rf {} +