upstream-smoke #115

	name: upstream-smoke

	# Live-parse lane: re-runs docling against the fixture PDFs to catch upstream
	# regressions (docling, OCR engines, model updates) without blocking PRs.
	#
	# Comparisons run in lenient mode — only structural fields (members, pages,
	# shape) are checked, since cell text varies by OCR engine. The job is
	# allowed to fail; treat persistent red here as a signal to investigate, not
	# a merge blocker.

	on:
	schedule:
	# Daily at 06:17 UTC. Off-hour to avoid contention with weekday CI bursts.
	- cron: "17 6 * * *"
	workflow_dispatch:

	jobs:
	live-parse:
	name: live-parse (${{ matrix.os }})
	# macos-latest matches the OCR engine (ocrmac / Apple Vision) used to
	# generate the committed snapshots, so structural drift here points at a
	# genuine docling change rather than at OCR-engine divergence.
	runs-on: ${{ matrix.os }}
	strategy:
	fail-fast: false
	matrix:
	os: [macos-latest]
	steps:
	- uses: actions/checkout@v4
	- uses: actions/setup-python@v5
	with:
	python-version: "3.11"
	cache: pip
	- name: cache docling models
	uses: actions/cache@v4
	with:
	path: ~/.cache/huggingface
	key: docling-models-${{ runner.os }}-v1
	- name: install
	run: \|
	python -m pip install --upgrade pip
	pip install -e ".[dev]"
	- name: live-parse integration tests (lenient)
	# Allowed to fail — we never want this to block a PR.
	continue-on-error: true
	run: pytest -m integration --live-parse tests/ -v

Navigation Menu