feat(focus): review protocols as YAML label sets

This commit is contained in:
kert
2026-11-14 10:35:49 -05:00
parent b29bdc2682
commit e56a9c9a97
9 changed files with 550 additions and 1 deletions

View File

@@ -279,6 +279,6 @@ link-mode = "copy"
python-preference = "managed"
[tool.uv.build-backend]
module-name = ["aco", "api", "bcda", "bib", "bls", "ccw", "cli", "cms", "conf", "lead", "llm", "mail", "nlm", "opps", "perf", "pfs", "prisma", "qpp", "rec", "rex", "sem"]
module-name = ["aco", "api", "bcda", "bib", "bls", "ccw", "cli", "cms", "conf", "focus", "lead", "llm", "mail", "nlm", "opps", "perf", "pfs", "prisma", "qpp", "rec", "rex", "sem"]
namespace = true
source-exclude = ["compose.yml","infra/**","data/**","notebooks/**","tuva/**","assets/**","docs/**","dev/**","bundle/**","cloud/**"]

5
src/focus/__init__.py Normal file
View File

@@ -0,0 +1,5 @@
"""Deep-focus review: protocols now, annotations and tags beside them."""
from focus.protocols import Label, Protocol, available, load
__all__ = ["Label", "Protocol", "available", "load"]

View File

@@ -0,0 +1,171 @@
"""Review protocols are YAML label sets. ``protocols/*.yaml`` is the source.
``prisma-screen.yaml`` carries ``from: lead.review.codes`` instead of a
label list. :func:`load` expands that marker from the live LEAD codebooks
so a codebook edit shows up here without a copied list.
A ``protocols.py`` module cannot sit next to this directory, so the loader
lives in the package. ``from focus.protocols import load`` is unchanged.
"""
from __future__ import annotations
import re
from dataclasses import dataclass
from pathlib import Path
from typing import Any
import yaml
_ID = re.compile(r"^[a-z0-9-]+$")
_KEYS = "1234567890abcdefghijklmnopqrstuvwxyz"
_FROM = "lead.review.codes"
ROOT = Path(__file__).parent
@dataclass(frozen=True)
class Label:
"""One keyboard-selectable label. ``tag`` drives the bib ``hl:`` tag."""
id: str
text: str
key: str
tag: str
@dataclass(frozen=True)
class Protocol:
"""A versioned label set. ``version`` is the number stored on an annotation."""
name: str
version: int
title: str
scope: str
labels: tuple[Label, ...]
def label(self, label_id: str) -> Label | None:
"""The label with this id, or ``None``."""
for item in self.labels:
if item.id == label_id:
return item
return None
def available() -> list[str]:
"""Protocol names, sorted, from the YAML files next to this module."""
return sorted(path.stem for path in ROOT.glob("*.yaml"))
def load(name: str) -> Protocol:
"""Load and validate one protocol. A bad file raises ``ValueError``
whose message contains the filename."""
path = ROOT / f"{name}.yaml"
if not path.is_file():
raise ValueError(f"{path.name}: no such protocol")
try:
data = yaml.safe_load(path.read_text(encoding="utf-8"))
except yaml.YAMLError as exc:
raise ValueError(f"{path.name}: invalid YAML: {exc}") from exc
try:
return _protocol(name, data)
except ValueError as exc:
raise ValueError(f"{path.name}: {exc}") from exc
def _protocol(name: str, data: Any) -> Protocol:
if not isinstance(data, dict):
raise ValueError("protocol is not a mapping")
got = data.get("name")
if not isinstance(got, str) or not got:
raise ValueError("name is missing")
if got != name:
raise ValueError(f"name {got!r} does not match the file")
version = data.get("version")
if type(version) is not int:
raise ValueError("version must be an int")
title = _text(data, "title")
scope = _text(data, "scope")
marker = data.get("from")
if marker is not None and "labels" in data:
raise ValueError("from and labels are both set")
if marker is not None:
if marker != _FROM:
raise ValueError(f"from must be {_FROM}")
raw_labels = _reason_labels()
else:
raw_labels = data.get("labels")
if not isinstance(raw_labels, list) or not raw_labels:
raise ValueError("labels must be a non-empty list")
return Protocol(
name=name,
version=version,
title=title,
scope=scope,
labels=_labels(name, raw_labels),
)
def _text(data: dict, field: str) -> str:
value = data.get(field)
if not isinstance(value, str) or not value:
raise ValueError(f"{field} is missing")
return value
def _reason_labels() -> list[dict[str, str]]:
"""Union of ``reason_codes`` across the LEAD projects, codebook order,
first project's sentence kept when a code repeats. Keys are assigned
here because the codebook does not carry shortcuts."""
from lead.review.codes import PROJECTS, reason_choices, reason_codes
sentences: dict[str, str] = {}
ordered: list[str] = []
seen: set[str] = set()
for project in PROJECTS:
for code, sentence in reason_choices(project):
sentences.setdefault(code, sentence)
for code in reason_codes(project):
if code not in seen:
seen.add(code)
ordered.append(code)
if len(ordered) > len(_KEYS):
raise ValueError("more labels than single-character keys")
missing = [code for code in ordered if code not in sentences]
if missing:
raise ValueError(f"reason code has no sentence: {missing[0]}")
return [
{"id": code, "text": sentences[code], "key": _KEYS[i]}
for i, code in enumerate(ordered)
]
def _labels(protocol: str, raw: list[Any]) -> tuple[Label, ...]:
found: list[Label] = []
ids: set[str] = set()
keys: set[str] = set()
for item in raw:
if not isinstance(item, dict):
raise ValueError("label is not a mapping")
label_id = item.get("id")
text = item.get("text")
key = item.get("key")
if not isinstance(label_id, str) or not _ID.fullmatch(label_id):
raise ValueError(f"id {label_id!r} must match {_ID.pattern}")
if not isinstance(text, str) or not text:
raise ValueError(f"label {label_id} is missing text")
if not isinstance(key, str) or len(key) != 1:
raise ValueError(f"key for {label_id} must be one character")
if label_id in ids:
raise ValueError(f"duplicate label id {label_id!r}")
if key in keys:
raise ValueError(f"duplicate key {key!r}")
ids.add(label_id)
keys.add(key)
tag = item.get("tag", None)
if tag is None:
tag = f"hl:{protocol}:{label_id}"
elif not isinstance(tag, str) or not tag:
raise ValueError(f"tag for {label_id} must be a non-empty string")
found.append(Label(id=label_id, text=text, key=key, tag=tag))
return tuple(found)

View File

@@ -0,0 +1,17 @@
name: cms-disposition
version: 1
title: CMS disposition
scope: How CMS disposed of the point in the final rule.
labels:
- id: finalized-as-proposed
text: Finalized as proposed
key: "1"
- id: finalized-with-modifications
text: Finalized with modifications
key: "2"
- id: considered-for-future-rulemaking
text: Considered for future rulemaking
key: "3"
- id: other
text: Other
key: "4"

View File

@@ -0,0 +1,24 @@
# Placeholder. pfs.elements.ElementType (actor, time, period, ...) is a
# descriptor vocabulary, not the span labels of a family biography, and
# the dossier design does not name a closed set. These five stand in
# until that set is settled.
name: code-family
version: 1
title: Code family
scope: A span in a source for a code family. Placeholder labels.
labels:
- id: code
text: Code
key: "1"
- id: descriptor-change
text: Descriptor change
key: "2"
- id: payment-change
text: Payment change
key: "3"
- id: rationale
text: Rationale
key: "4"
- id: other
text: Other
key: "5"

View File

@@ -0,0 +1,14 @@
name: comment-stance
version: 1
title: Comment stance
scope: A commenter's position on the proposal being read.
labels:
- id: support
text: Support
key: "1"
- id: support-with-modifications
text: Support with modifications
key: "2"
- id: do-not-support
text: Do not support
key: "3"

View File

@@ -0,0 +1,7 @@
name: prisma-screen
version: 1
title: PRISMA screen
scope: >
Evidence for an include or exclude on a LEAD screening item.
Labels expand at load from lead.review.codes, one per reason code.
from: lead.review.codes

0
tests/focus/__init__.py Normal file
View File

View File

@@ -0,0 +1,311 @@
"""Review protocols: YAML label sets, validation, and PRISMA expansion."""
from __future__ import annotations
from pathlib import Path
import pytest
import yaml
from focus import protocols
from lead.review.codes import PROJECTS, reason_choices, reason_codes
_KEYS = "1234567890abcdefghijklmnopqrstuvwxyz"
def _write(directory: Path, name: str, body: str) -> None:
(directory / f"{name}.yaml").write_text(body, encoding="utf-8")
def _base(**overrides: object) -> dict:
data: dict = {
"name": "sample",
"version": 1,
"title": "Sample",
"scope": "A test protocol.",
"labels": [
{"id": "alpha", "text": "Alpha", "key": "1"},
{"id": "beta", "text": "Beta", "key": "2", "tag": "custom:beta"},
],
}
data.update(overrides)
return data
def _dump(directory: Path, name: str, data: dict) -> None:
_write(directory, name, yaml.safe_dump(data))
def test_available_lists_shipped_names():
assert protocols.available() == [
"cms-disposition",
"code-family",
"comment-stance",
"prisma-screen",
]
def test_comment_stance_labels_and_default_tags():
proto = protocols.load("comment-stance")
assert proto.name == "comment-stance"
assert proto.version == 1
assert proto.title == "Comment stance"
assert proto.scope
assert [(lb.id, lb.key) for lb in proto.labels] == [
("support", "1"),
("support-with-modifications", "2"),
("do-not-support", "3"),
]
assert [lb.tag for lb in proto.labels] == [
"hl:comment-stance:support",
"hl:comment-stance:support-with-modifications",
"hl:comment-stance:do-not-support",
]
assert proto.label("support") == proto.labels[0]
assert proto.label("missing") is None
def test_cms_disposition_labels():
proto = protocols.load("cms-disposition")
assert [lb.id for lb in proto.labels] == [
"finalized-as-proposed",
"finalized-with-modifications",
"considered-for-future-rulemaking",
"other",
]
assert all(lb.tag == f"hl:cms-disposition:{lb.id}" for lb in proto.labels)
def test_code_family_placeholder_labels():
"""Element names for a family biography are not a closed label set.
These five are the plan's placeholder."""
proto = protocols.load("code-family")
assert [lb.id for lb in proto.labels] == [
"code",
"descriptor-change",
"payment-change",
"rationale",
"other",
]
def test_prisma_screen_expands_real_reason_codes():
proto = protocols.load("prisma-screen")
expected: list[str] = []
seen: set[str] = set()
sentences: dict[str, str] = {}
for project in PROJECTS:
for code, sentence in reason_choices(project):
sentences.setdefault(code, sentence)
for code in reason_codes(project):
if code not in seen:
seen.add(code)
expected.append(code)
assert [lb.id for lb in proto.labels] == expected
assert {lb.id for lb in proto.labels} == {
code for project in PROJECTS for code in reason_codes(project)
}
assert [lb.text for lb in proto.labels] == [sentences[i] for i in expected]
assert [lb.key for lb in proto.labels] == list(_KEYS[: len(expected)])
assert all(lb.tag == f"hl:prisma-screen:{lb.id}" for lb in proto.labels)
def test_explicit_tag_is_kept(tmp_path: Path, monkeypatch: pytest.MonkeyPatch):
monkeypatch.setattr(protocols, "ROOT", tmp_path)
_dump(tmp_path, "sample", _base())
proto = protocols.load("sample")
assert proto.labels[0].tag == "hl:sample:alpha"
assert proto.labels[1].tag == "custom:beta"
assert protocols.available() == ["sample"]
def test_missing_protocol_names_the_file():
with pytest.raises(ValueError, match=r"missing\.yaml"):
protocols.load("missing")
@pytest.mark.parametrize(
"mutate",
[
"duplicate-id",
"duplicate-key",
"long-key",
"empty-key",
"int-key",
"bad-id",
"empty-id",
"missing-text",
"label-not-mapping",
"missing-labels",
"empty-labels",
"name-mismatch",
"version-str",
"version-bool",
"missing-title",
"title-not-str",
"missing-scope",
"scope-not-str",
"not-a-mapping",
"bad-from",
"from-and-labels",
"empty-tag",
"tag-not-str",
"missing-name",
],
)
def test_bad_file_raises_with_filename(
tmp_path: Path, monkeypatch: pytest.MonkeyPatch, mutate: str
):
monkeypatch.setattr(protocols, "ROOT", tmp_path)
data = _base()
raw: str | None = None
if mutate == "duplicate-id":
data["labels"][1]["id"] = "alpha"
elif mutate == "duplicate-key":
data["labels"][1]["key"] = "1"
elif mutate == "long-key":
data["labels"][0]["key"] = "ab"
elif mutate == "empty-key":
data["labels"][0]["key"] = ""
elif mutate == "int-key":
data["labels"][0]["key"] = 1
elif mutate == "bad-id":
data["labels"][0]["id"] = "Bad_Id"
elif mutate == "empty-id":
data["labels"][0]["id"] = ""
elif mutate == "missing-text":
del data["labels"][0]["text"]
elif mutate == "label-not-mapping":
data["labels"][0] = "alpha"
elif mutate == "missing-labels":
del data["labels"]
elif mutate == "empty-labels":
data["labels"] = []
elif mutate == "name-mismatch":
data["name"] = "other"
elif mutate == "version-str":
data["version"] = "1"
elif mutate == "version-bool":
data["version"] = True
elif mutate == "missing-title":
del data["title"]
elif mutate == "title-not-str":
data["title"] = 1
elif mutate == "missing-scope":
del data["scope"]
elif mutate == "scope-not-str":
data["scope"] = 1
elif mutate == "not-a-mapping":
raw = "- just\n- a\n- list\n"
elif mutate == "bad-from":
del data["labels"]
data["from"] = "somewhere.else"
elif mutate == "from-and-labels":
data["from"] = "lead.review.codes"
elif mutate == "empty-tag":
data["labels"][0]["tag"] = ""
elif mutate == "tag-not-str":
data["labels"][0]["tag"] = 3
elif mutate == "missing-name":
del data["name"]
else:
raise AssertionError(mutate)
if raw is None:
_dump(tmp_path, "sample", data)
else:
_write(tmp_path, "sample", raw)
with pytest.raises(ValueError, match=r"sample\.yaml"):
protocols.load("sample")
def test_invalid_yaml_names_the_file(tmp_path: Path, monkeypatch: pytest.MonkeyPatch):
monkeypatch.setattr(protocols, "ROOT", tmp_path)
_write(tmp_path, "sample", "name: [\n")
with pytest.raises(ValueError, match=r"sample\.yaml"):
protocols.load("sample")
def test_expansion_rejects_more_labels_than_keys(
tmp_path: Path, monkeypatch: pytest.MonkeyPatch
):
monkeypatch.setattr(protocols, "ROOT", tmp_path)
_dump(
tmp_path,
"wide",
{
"name": "wide",
"version": 1,
"title": "Wide",
"scope": "too many",
"from": "lead.review.codes",
},
)
codes = [f"c{i}" for i in range(len(_KEYS) + 1)]
monkeypatch.setattr(
"lead.review.codes.PROJECTS",
("only",),
)
monkeypatch.setattr(
"lead.review.codes.reason_codes",
lambda _project: tuple(codes),
)
monkeypatch.setattr(
"lead.review.codes.reason_choices",
lambda _project: tuple((code, f"{code}.") for code in codes),
)
with pytest.raises(ValueError, match=r"wide\.yaml"):
protocols.load("wide")
def test_expansion_rejects_a_code_with_no_sentence(
tmp_path: Path, monkeypatch: pytest.MonkeyPatch
):
monkeypatch.setattr(protocols, "ROOT", tmp_path)
_dump(
tmp_path,
"wide",
{
"name": "wide",
"version": 1,
"title": "Wide",
"scope": "no sentence",
"from": "lead.review.codes",
},
)
monkeypatch.setattr("lead.review.codes.PROJECTS", ("only",))
monkeypatch.setattr("lead.review.codes.reason_codes", lambda _p: ("orphan",))
monkeypatch.setattr("lead.review.codes.reason_choices", lambda _p: ())
with pytest.raises(ValueError, match=r"wide\.yaml"):
protocols.load("wide")
def test_expansion_failure_names_the_file(
tmp_path: Path, monkeypatch: pytest.MonkeyPatch
):
monkeypatch.setattr(protocols, "ROOT", tmp_path)
_dump(
tmp_path,
"wide",
{
"name": "wide",
"version": 1,
"title": "Wide",
"scope": "boom",
"from": "lead.review.codes",
},
)
def _boom(_project: str) -> tuple[str, ...]:
raise ValueError("codebook mismatch")
monkeypatch.setattr("lead.review.codes.reason_codes", _boom)
with pytest.raises(ValueError, match=r"wide\.yaml"):
protocols.load("wide")
def test_available_skips_non_yaml(tmp_path: Path, monkeypatch: pytest.MonkeyPatch):
monkeypatch.setattr(protocols, "ROOT", tmp_path)
_dump(tmp_path, "b", _base(name="b"))
_dump(tmp_path, "a", _base(name="a"))
(tmp_path / "notes.txt").write_text("nope", encoding="utf-8")
assert protocols.available() == ["a", "b"]