Files
jobboard/tests/test_enrich.py
T
WolfandClaude Opus 5.5 d04ceae7e0 Prep stage: host words stored raw, mapped centrally in app/prep after every pull
upsert_role stores the host's own words (roles.*_host), the place texts
unresolved when they change, and the company key unresolved on insert, and
marks the role prep_due. app.prep maps seniority, employment, remote and
country through the host tables in app/prep/hosts, resolves places and
company aliases, lists unknown words in `unmapped`, remaps every open role
when the tables' fingerprint changes, and runs rebuild_dupes, which moves
out of the pull's settle. Migration 22 adds the columns and the table and
fills the host words from raw. Cascade: pull, prep, embed, score, enrich,
prep, embed, score, extract, score, summarise. `app prep [unmapped]
[--remap]`, /api/prep/unmapped and a fold on the Sources page.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-09-30 11:19:22 +02:00

371 lines
16 KiB
Python

"""The `enrich` stage: the best roles without text, read on their own pages
after scoring. Which roles and in what order, the budgets, what is stored,
the cascade around it, and what triggers it. No network: a fake host stands
in, answering from a dict."""
from __future__ import annotations
import json
import sys
import types
from collections.abc import Iterator
from datetime import UTC, datetime
from pathlib import Path
from typing import Any
import httpx
import numpy as np
import pytest
from fastapi import FastAPI
from fastapi.testclient import TestClient
ROOT = Path(__file__).resolve().parents[1]
if str(ROOT) not in sys.path:
sys.path.insert(0, str(ROOT))
from app.config import Config, ServerConfig # noqa: E402
from app.jobs import cascade, embed, enrich, stages # noqa: E402
from app.main import create_app # noqa: E402
from app.storage import Store # noqa: E402
from desk.models import Item # noqa: E402
from pipeline import FINAL, JobQueue, Jobs, Tally # noqa: E402
from plugins import ats, readers # noqa: E402
BASE = "https://jobs.fake.example"
def _fake_host(name: str = "fake", *, detail: bool = True) -> types.ModuleType:
"""A host whose role pages are `BASE/<slug>/<id>`, answered by `web`."""
module: Any = types.ModuleType(name)
module.NAME, module.LABEL, module.PATTERNS = name, name.title(), ()
module.slug_ok = lambda slug: slug.isalnum()
module.allowed = lambda slug: lambda h: h == "jobs.fake.example"
module.careers_url = lambda slug: f"{BASE}/{slug}"
module.feed_url = lambda slug: f"{BASE}/{slug}/list"
module.parse = lambda body, slug: []
if detail:
module.detail = lambda get, slug, row: {
"description": get(f"{BASE}/{slug}/{row['source_id']}").decode(),
"location": "Copenhagen"}
return module
@pytest.fixture
def web(monkeypatch: pytest.MonkeyPatch) -> dict[str, Any]:
"""`ats._get` answering from `answers` (URL -> body), 404 otherwise,
remembering what it was asked; hosts `fake` and `plain` (no detail)."""
answers: dict[str, Any] = {}
asked: list[str] = []
def fake_get(client, url, host, slug, *, method="GET", payload=None, headers=None):
asked.append(url)
answer = answers.get(url)
if answer is None:
request = httpx.Request("GET", url)
raise httpx.HTTPStatusError("404", request=request,
response=httpx.Response(404, request=request))
return answer
monkeypatch.setattr(ats, "_get", fake_get)
monkeypatch.setattr(ats, "PAGE_DELAY", 0)
monkeypatch.setitem(readers.HOSTS, "fake", _fake_host())
monkeypatch.setitem(readers.HOSTS, "plain", _fake_host("plain", detail=False))
return {"answers": answers, "asked": asked}
@pytest.fixture
def store(tmp_path: Path) -> Store:
return Store(tmp_path / "roledesk.db")
def config(tmp_path: Path, **enrich_settings: Any) -> Config:
return Config(database=str(tmp_path / "roledesk.db"), embed={"model": "fake"},
enrich=enrich_settings)
def board(store: Store, target: str) -> Any:
host = target.partition(":")[0]
return store.add_subscription("ats", target, label=target, company=target, board=host)
def role(store: Store, target: str, native: str, *, body: str = "",
title: str = "Backend Engineer") -> str:
"""A role of board `target` as a pull stores it: without text on a host
with `detail`."""
sub = next(s for s in store.subscriptions(source="ats") if s.target == target)
host, _, slug = target.partition(":")
item = Item(source_id=f"{target}:{native}", location=f"{BASE}/{slug}/{native}",
title=title, body=body, published_at=datetime(2026, 9, 20, tzinfo=UTC),
raw={"company": target, "place": "2 Locations", "board": host, "slug": slug})
item.source = "ats"
store.upsert_role(item, sub)
return item.uid
def scored(store: Store, name: str, scores: dict[str, float], *,
gated: tuple[str, ...] = ()) -> Any:
agent = store.save_agent(name, f"{name} python backend")
store.replace_scores(agent.id, [(uid, value, "gated" if uid in gated else "")
for uid, value in scores.items()])
return agent
# -------------------------------------------------------------- which roles
def test_roles_go_best_first_by_their_best_agent_score(
store: Store, tmp_path: Path, web: dict[str, Any]) -> None:
board(store, "fake:acme")
board(store, "fake:beta")
a1, a2 = role(store, "fake:acme", "1"), role(store, "fake:acme", "2")
b1 = role(store, "fake:beta", "1")
scored(store, "One", {a1: 0.6, a2: 0.9, b1: 0.55})
scored(store, "Two", {a1: 0.95, b1: 0.7})
assert [(r["uid"], r["score"]) for r in store.enrich_waiting(500, 0.5)] == [
(a1, 0.95), (a2, 0.9), (b1, 0.7)]
assert enrich.plan(config(tmp_path), store) == [("fake:acme", [a1, a2]),
("fake:beta", [b1])]
def test_a_role_no_agent_ranks_high_enough_gets_no_call(
store: Store, tmp_path: Path, web: dict[str, Any]) -> None:
board(store, "fake:acme")
low, gated, deep, kept = (role(store, "fake:acme", n) for n in ("1", "2", "3", "4"))
scored(store, "One", {low: 0.4, gated: 0.0, deep: 0.8, kept: 0.9}, gated=(gated,))
# `top` 1 per agent: only the best ungated role of each agent is in reach.
assert enrich.plan(config(tmp_path, top=1), store) == [("fake:acme", [kept])]
assert enrich.plan(config(tmp_path), store) == [("fake:acme", [kept, deep])]
# A disabled agent ranks nothing.
agent = store.agents()[0]
store.save_agent(agent.name, gates=agent.gates, weights=agent.weights, enabled=False,
agent_id=agent.id, limits=agent.limits, prompts=agent.prompts)
assert enrich.plan(config(tmp_path), store) == []
def test_roles_with_text_hosts_without_pages_and_failing_boards_are_left_out(
store: Store, tmp_path: Path, web: dict[str, Any]) -> None:
for target in ("fake:acme", "plain:acme", "fake:down"):
board(store, target)
wanted = role(store, "fake:acme", "1")
has_text = role(store, "fake:acme", "2", body="Python in full")
plain = role(store, "plain:acme", "1")
down = role(store, "fake:down", "1")
store.state["ats:failing:fake:down"] = json.dumps({"since": "2026-09-29", "reason": "500"})
scored(store, "One", {wanted: 0.9, has_text: 0.95, plain: 0.9, down: 0.9})
assert enrich.plan(config(tmp_path), store) == [("fake:acme", [wanted])]
def test_a_role_read_already_waits_for_the_retry_and_a_failed_one_goes_last(
store: Store, tmp_path: Path, web: dict[str, Any]) -> None:
board(store, "fake:acme")
read, failed, fresh = (role(store, "fake:acme", n) for n in ("1", "2", "3"))
now = datetime.now(UTC).isoformat(timespec="seconds")
store.state["ats:detailed:fake:acme"] = json.dumps(["fake:acme:1"])
store.state["ats:detailempty:fake:acme"] = json.dumps({"fake:acme:1": now})
store.state["ats:detailfailed:fake:acme"] = json.dumps(["fake:acme:2"])
scored(store, "One", {read: 0.9, failed: 0.8, fresh: 0.6})
assert enrich.plan(config(tmp_path), store) == [("fake:acme", [fresh, failed])]
# ------------------------------------------------------------------ budgets
def test_the_budgets_per_board_and_per_run(
store: Store, tmp_path: Path, web: dict[str, Any]) -> None:
board(store, "fake:acme")
board(store, "fake:beta")
acme = [role(store, "fake:acme", str(n)) for n in range(4)]
beta = [role(store, "fake:beta", str(n)) for n in range(4)]
# Interleaved: acme 0.99, beta 0.98, acme 0.97, ...
scored(store, "One", {uid: 0.99 - 0.01 * n
for n, uid in enumerate(u for pair in zip(acme, beta, strict=True)
for u in pair)})
assert enrich.plan(config(tmp_path, per_board=2), store) == [
("fake:acme", acme[:2]), ("fake:beta", beta[:2])]
assert enrich.plan(config(tmp_path, per_run=3), store) == [
("fake:acme", acme[:2]), ("fake:beta", beta[:1])]
assert enrich.plan(config(tmp_path, per_board=1, per_run=5), store) == [
("fake:acme", acme[:1]), ("fake:beta", beta[:1])]
def test_the_stage_off_finds_nothing(
store: Store, tmp_path: Path, web: dict[str, Any]) -> None:
board(store, "fake:acme")
scored(store, "One", {role(store, "fake:acme", "1"): 0.9})
assert len(enrich.batches(config(tmp_path), store)) == 1
assert enrich.batches(config(tmp_path, enabled=False), store) == []
stages.set_stage(store, "enrich", True)
assert len(enrich.batches(config(tmp_path, enabled=False), store)) == 1
stages.set_stage(store, "enrich", False)
assert enrich.batches(config(tmp_path), store) == []
# ------------------------------------------------------------ what is stored
def test_a_batch_stores_what_the_pages_filled_and_drops_those_vectors(
store: Store, tmp_path: Path, web: dict[str, Any]) -> None:
board(store, "fake:acme")
one, two, gone = (role(store, "fake:acme", n) for n in ("1", "2", "3"))
store.put_vectors("fake", [(uid, np.ones(3, dtype=np.float32)) for uid in (one, two, gone)])
web["answers"].update({f"{BASE}/acme/1": b"Python in full", f"{BASE}/acme/2": b"Go too"})
out = enrich.run_batch(config(tmp_path), store, target="fake:acme", uids=[two, gone, one])
assert out == {"target": "fake:acme", "read": 3, "filled": 2, "texts": 2}
assert web["asked"] == [f"{BASE}/acme/2", f"{BASE}/acme/3", f"{BASE}/acme/1"]
first = store.role(one)
assert first is not None and first.description == "Python in full"
assert first.place == "Copenhagen" and first.title == "Backend Engineer"
assert set(store.vectors("fake")) == {gone}
assert json.loads(store.state["ats:detailed:fake:acme"]) == [
"fake:acme:1", "fake:acme:2", "fake:acme:3"]
# A later pull of the same listing keeps the text and the page's place.
item = ats.item_of({"source_id": "1", "title": "Backend Engineer", "url": f"{BASE}/acme/1",
"description": "", "location": "", "places": [], "remote": None},
"fake:acme", "fake:acme")
item.source = "ats"
store.upsert_role(item, store.subscriptions(source="ats")[0])
again = store.role(one)
assert again is not None and again.description == "Python in full"
assert again.place == "Copenhagen"
def test_a_board_switched_off_since_the_plan_reads_nothing(
store: Store, tmp_path: Path, web: dict[str, Any]) -> None:
sub = board(store, "fake:acme")
uid = role(store, "fake:acme", "1")
store.update_subscription(sub.id, fetch=0)
assert enrich.run_batch(config(tmp_path), store, target="fake:acme", uids=[uid])["filled"] == 0
assert web["asked"] == []
# ------------------------------------------------------------------ cascade
class _Context:
def __init__(self) -> None:
self.made: list[Any] = []
def task(self, t: Any) -> Any:
self.made.append(t)
return t
def test_nothing_filled_passes_over_the_prep_embed_and_score_behind_enrich() -> None:
context = _Context()
empty = Tally(results=({"filled": 0},))
out = cascade.step(None, None, empty, {"cascade": "pull", "task": "enrich", "at": 4},
context=context)
assert out["next"] == "extract" and context.made[0].carries["at"] == 8
out = cascade.step(None, None, Tally(results=({"filled": 2}, {"filled": 0})),
{"cascade": "pull", "task": "enrich", "at": 4}, context=context)
assert out["next"] == "prep" and context.made[1].carries["at"] == 5
out = cascade.step(None, None, empty, {"cascade": "enrich", "task": "enrich", "at": 0},
context=context)
assert out["next"] == "" and len(context.made) == 2
class _Embedder:
def encode(self, texts: list[str]) -> np.ndarray:
return np.vstack([np.asarray([t.lower().count("python"), 0.1], dtype=np.float32)
for t in texts])
def test_the_enrich_job_reads_the_pages_then_embeds_and_scores_those_roles(
store: Store, tmp_path: Path, web: dict[str, Any],
monkeypatch: pytest.MonkeyPatch) -> None:
monkeypatch.setattr(embed, "load_embedder", lambda model: _Embedder())
board(store, "fake:acme")
one, two = role(store, "fake:acme", "1"), role(store, "fake:acme", "2")
store.put_vectors("fake", [(uid, np.ones(2, dtype=np.float32)) for uid in (one, two)])
agent = scored(store, "One", {one: 0.9, two: 0.1})
web["answers"][f"{BASE}/acme/1"] = b"Python in full"
queue = JobQueue(Jobs(tmp_path / "roledesk.db"), config(tmp_path), store,
workers={"cpu": 1, "http": 1, "llm": 1})
started = queue.enqueue("enrich")
queue.drain(timeout=60)
assert started is not None
jobs = sorted(queue.jobs.of_cascade(queue.jobs.last(enrich.KIND).cascade_id), # type: ignore[union-attr]
key=lambda j: j.id)
assert [j.result.get("task") for j in jobs if j.kind == FINAL] == \
["enrich", "prep", "embed", "score"]
assert [j.kind for j in jobs if j.kind != FINAL] == \
["enrich", enrich.KIND, "prep", "embed", "score"]
assert web["asked"] == [f"{BASE}/acme/1"]
embedded = next(j for j in jobs if j.kind == "embed")
assert embedded.result["embedded"] == 1
assert set(store.vectors("fake")) == {one, two}
assert store.scores(agent.id)
# ----------------------------------------------------------------- triggers
@pytest.fixture
def app(tmp_path: Path) -> FastAPI:
return create_app(Config(server=ServerConfig(host="127.0.0.1", port=0),
database=tmp_path / "roledesk.db"))
@pytest.fixture
def client(app: FastAPI) -> Iterator[TestClient]:
with TestClient(app) as c:
yield c
ANY_STATE = ("queued", "running", "done", "failed", "cancelled")
def _enrich_behind_a_score(queue: Any) -> Any:
found = queue.jobs.last("enrich", states=ANY_STATE)
assert found is not None
assert [queue.jobs.get(i).kind for i in found.depends_on] == ["score"]
assert found.after == "done"
return found
def test_saving_an_agent_queues_a_score_then_enrich(client: TestClient, app: FastAPI) -> None:
res = client.post("/api/agents", json={"name": "Backend", "intent": "python backend"})
assert res.status_code == 200
_enrich_behind_a_score(app.state.queue)
res = client.post("/api/agents", json={"id": res.json()["id"], "name": "Backend",
"intent": "go backend"})
assert res.status_code == 200
def test_a_company_affinity_queues_a_score_then_enrich(
client: TestClient, app: FastAPI) -> None:
store: Store = app.state.store
store.save_agent("Backend", "python backend")
store.add_company("Acme")
res = client.post("/api/companies/acme/affinity", json={"affinity": 2})
assert res.status_code == 200
found = _enrich_behind_a_score(app.state.queue)
assert found.depends_on == [res.json()["score_job"]]
def test_one_score_is_one_enrich(app: FastAPI) -> None:
queue = app.state.queue
score = queue.enqueue("score")
first = cascade.enrich_after(queue, score)
assert cascade.enrich_after(queue, score).id == first.id
assert cascade.enrich_after(queue, None) is None
def test_a_vote_queues_neither_score_nor_enrich(client: TestClient, app: FastAPI) -> None:
store: Store = app.state.store
store.save_agent("Backend", "python backend")
item = Item(source_id="fake:acme:1", location=f"{BASE}/acme/1", title="Backend Engineer",
body="", published_at=datetime(2026, 9, 20, tzinfo=UTC), raw={"company": "Acme"})
item.source = "ats"
store.upsert_role(item)
assert client.post(f"/api/roles/{item.uid}/vote", json={"value": 1}).status_code == 200
assert client.delete(f"/api/roles/{item.uid}/vote").status_code == 200
queue = app.state.queue
assert queue.jobs.last("enrich", states=ANY_STATE) is None
assert queue.jobs.last("score", states=ANY_STATE) is None