Skip to content

AI-powered virtual executive team — a single coherent executive persona backed by 8 specialist agents (FastAPI + Next.js).

fix(onboarding): finish the ARR regex fix, cover burn rate, make the final step retryable

Claude147962ffdc97
diff --git a/packages/core/openexecutive/api/routes/onboarding.py b/packages/core/openexecutive/api/routes/onboarding.py
index 3921e451..797b960a 100644
--- a/packages/core/openexecutive/api/routes/onboarding.py
+++ b/packages/core/openexecutive/api/routes/onboarding.py
@@ -1,6 +1,7 @@
from __future__ import annotations
import asyncio
+import copy
import logging
import uuid
@@ -62,13 +63,28 @@ async def submit_answer(body: OnboardAnswerRequest) -> OnboardStatusResponse:
if state.completed:
raise HTTPException(status_code=400, detail="Onboarding already completed")
+ # process_answer mutates the stored state in place. Keep a snapshot so
+ # a failed profile build on the final answer can be rolled back —
+ # otherwise the session is stuck at completed=True and every retry
+ # hits the 400 above, forcing the user to restart onboarding.
+ snapshot = copy.deepcopy(state)
state = process_answer(state, body.answer)
- _wizard_sessions[body.session_id] = state
if state.completed:
from openexecutive.onboarding.profile_builder import build_and_save_profile
- build_and_save_profile(state)
+ try:
+ build_and_save_profile(state)
+ except Exception as exc:
+ logger.exception("onboarding: profile build failed on the final answer")
+ _wizard_sessions[body.session_id] = snapshot
+ raise HTTPException(
+ status_code=422,
+ detail=(
+ "Could not build the company profile from your answers. "
+ "Please rephrase your last answer and try again."
+ ),
+ ) from exc
# Fire the watchlist-research workflow once at onboarding
# completion so the principal's first /today after install
@@ -87,6 +103,8 @@ async def submit_answer(body: OnboardAnswerRequest) -> OnboardStatusResponse:
# the process lifetime.
task.add_done_callback(_background_research_tasks.discard)
+ _wizard_sessions[body.session_id] = state
+
question = get_current_question(state) if not state.completed else None
progress = state.get_progress()
diff --git a/packages/core/openexecutive/onboarding/wizard.py b/packages/core/openexecutive/onboarding/wizard.py
index 90787028..33abb0ac 100644
--- a/packages/core/openexecutive/onboarding/wizard.py
+++ b/packages/core/openexecutive/onboarding/wizard.py
@@ -198,13 +198,15 @@ def build_profile_from_answers(answers: dict[str, Any]) -> dict[str, Any]:
profile["target_customer"] = {"profile": text, "pain_points": []}
import re
- # The capture must START with a digit and the M must end a word.
- # `[\d,]+` alone matched a bare comma, so any answer with a comma
- # before an m-word — "Consultoría estratégica, marketing y
- # operaciones" — captured "," and crashed the whole wizard on
- # float(""). Requiring \b also stops "300 clientes, marketing"
- # from being read as $300M.
- arr_match = re.search(r"\$?(\d[\d,]*)\s*[Mm]\b", text)
+ # The capture must START with a digit and the magnitude must end a
+ # word. `[\d,]+` alone matched a bare comma, so any answer with a
+ # comma before an m-word — "IT, marketing and video agency" —
+ # captured "," and crashed the whole wizard on float(""). The \b
+ # also stops "300 clients, marketing" from being read as $300M and
+ # "$5 minimum" from becoming $5M. The alternation keeps "M", "MM"
+ # and "million" all parsing, since a bare `[Mm]\b` silently dropped
+ # "roughly 50 million" and "$50MM".
+ arr_match = re.search(r"\$?(\d[\d,]*)\s*(?:[Mm]{1,2}|[Mm]illion)\b", text)
if arr_match:
val = arr_match.group(1).replace(",", "")
# Defensive: this is a best-effort parse of free text, so a
@@ -262,14 +264,22 @@ def build_profile_from_answers(answers: dict[str, Any]) -> dict[str, Any]:
import re
text = answers["financials"]
- burn_match = re.search(r"\$?([\d,]+)\s*[Kk]?\s*(?:monthly|/month|per month|burn)", text)
+ # Same digit-first rule as the ARR parse above: `[\d,]+` matched a
+ # bare comma, so "runway is fine, monthly costs are low" captured
+ # "," and crashed on float("").
+ burn_match = re.search(
+ r"\$?(\d[\d,]*)\s*[Kk]?\s*(?:monthly|/month|per month|burn)", text
+ )
runway_match = re.search(r"(\d+)\s*month", text)
fin: dict[str, Any] = {}
if burn_match:
val = burn_match.group(1).replace(",", "")
multiplier = 1000 if "k" in text[burn_match.start():burn_match.end()].lower() else 1
- fin["burn_rate_monthly"] = float(val) * multiplier
+ try:
+ fin["burn_rate_monthly"] = float(val) * multiplier
+ except ValueError:
+ logger.warning("onboarding: could not parse burn rate from %r", text[:80])
if runway_match:
fin["runway_months"] = float(runway_match.group(1))
profile["financials"] = fin
diff --git a/packages/core/tests/unit/test_onboarding_wizard_arr_parse.py b/packages/core/tests/unit/test_onboarding_wizard_arr_parse.py
index 676a6d47..15affdcc 100644
--- a/packages/core/tests/unit/test_onboarding_wizard_arr_parse.py
+++ b/packages/core/tests/unit/test_onboarding_wizard_arr_parse.py
@@ -26,6 +26,8 @@ from openexecutive.onboarding.wizard import build_profile_from_answers
"Vendemos software B2B, modelo de suscripción",
"We sell software, mostly to mid-market teams",
"Design and build, maintenance included",
+ # The exact answer from issue #84.
+ "IT, marketing and video agency",
],
)
def test_comma_before_m_word_does_not_crash(text: str) -> None:
@@ -42,6 +44,11 @@ def test_comma_before_m_word_does_not_crash(text: str) -> None:
("We do $2M in ARR", 2_000_000.0),
("Roughly 12M annually", 12_000_000.0),
("ARR is $1,500M across all lines", 1_500_000_000.0),
+ # Phrasings the bare `[Mm]\b` narrowing silently dropped.
+ ("roughly 50 million in revenue", 50_000_000.0),
+ ("About 3 Million ARR", 3_000_000.0),
+ ("$50MM ARR last year", 50_000_000.0),
+ ("we closed the year at 7mm", 7_000_000.0),
],
)
def test_magnitude_still_parses(text: str, expected: float) -> None:
@@ -64,6 +71,21 @@ def test_digits_before_an_m_word_are_not_a_magnitude() -> None:
assert "annual_revenue_arr" not in profile
+ "text",
+ [
+ "$5 minimum order, mostly SMBs",
+ "We have 300 clients, marketing is word of mouth",
+ "12 major accounts and growing",
+ ],
+)
+def test_digits_before_a_non_magnitude_m_word_are_ignored(text: str) -> None:
+ """The word boundary must hold across the whole alternation."""
+ profile = build_profile_from_answers({"business_model": text})
+
+ assert "annual_revenue_arr" not in profile
+
+
def test_no_magnitude_leaves_the_field_unset() -> None:
profile = build_profile_from_answers({"business_model": "A boutique consultancy"})
diff --git a/packages/core/tests/unit/test_onboarding_wizard_financials_parse.py b/packages/core/tests/unit/test_onboarding_wizard_financials_parse.py
new file mode 100644
index 00000000..a29f5b67
--- /dev/null
+++ b/packages/core/tests/unit/test_onboarding_wizard_financials_parse.py
@@ -0,0 +1,49 @@
+"""Burn-rate parsing in the onboarding wizard's financials step.
+
+Sibling of the ARR regression in ``test_onboarding_wizard_arr_parse.py``:
+the burn-rate heuristic used the same ``[\\d,]+`` character class, so a bare
+comma before "monthly", "/month", "per month" or "burn" captured ``","`` and
+crashed ``build_profile_from_answers`` on ``float("")``.
+"""
+from __future__ import annotations
+
+import pytest
+
+from openexecutive.onboarding.wizard import build_profile_from_answers
+
+
+ "text",
+ [
+ "Runway is fine, monthly costs are low",
+ "We are profitable, burn is zero",
+ "12 months runway, monthly burn 100k",
+ ],
+)
+def test_comma_before_burn_keyword_does_not_crash(text: str) -> None:
+ profile = build_profile_from_answers({"financials": text})
+
+ assert "burn_rate_monthly" not in profile["financials"]
+
+
+def test_comma_case_still_records_runway() -> None:
+ profile = build_profile_from_answers(
+ {"financials": "12 months runway, monthly burn 100k"}
+ )
+
+ assert profile["financials"]["runway_months"] == 12.0
+
+
+ ("text", "expected"),
+ [
+ ("burn 50k monthly", 50_000.0),
+ ("$120,000 per month, 18 months runway", 120_000.0),
+ ("We spend 30K/month", 30_000.0),
+ ("about 2,500 monthly burn", 2_500.0),
+ ],
+)
+def test_burn_rate_still_parses(text: str, expected: float) -> None:
+ profile = build_profile_from_answers({"financials": text})
+
+ assert profile["financials"]["burn_rate_monthly"] == expected
diff --git a/packages/core/tests/unit/test_onboarding_wizard_route.py b/packages/core/tests/unit/test_onboarding_wizard_route.py
new file mode 100644
index 00000000..177b85f2
--- /dev/null
+++ b/packages/core/tests/unit/test_onboarding_wizard_route.py
@@ -0,0 +1,101 @@
+"""HTTP-level tests for the company-onboarding wizard's final step.
+
+Regression for issue #84's second half: ``/onboard/answer`` used to commit the
+final answer (``completed=True``) *before* building the profile. When the
+build raised, the client got a 500 and every retry hit "Onboarding already
+completed" — the only way out was to restart onboarding from scratch.
+"""
+from __future__ import annotations
+
+from pathlib import Path
+
+import pytest
+from fastapi import FastAPI
+from fastapi.testclient import TestClient
+
+from openexecutive.api.routes import onboarding as route
+from openexecutive.onboarding import profile_builder
+from openexecutive.onboarding.wizard import TOTAL_STEPS
+
+
+def client(monkeypatch: pytest.MonkeyPatch) -> TestClient:
+ async def _no_research(session_id: str) -> None:
+ return None
+
+ monkeypatch.setattr(route, "_fire_post_onboarding_research", _no_research)
+ monkeypatch.setattr(route, "_wizard_sessions", {})
+ monkeypatch.setattr(route, "_onboarding_research_fired", set())
+ app = FastAPI()
+ app.include_router(route.router)
+ return TestClient(app)
+
+
+def _answer_all_but_last(client: TestClient) -> str:
+ session_id = client.get("/onboard/start").json()["session_id"]
+ for _ in range(TOTAL_STEPS - 1):
+ resp = client.post("/onboard/answer", json={"session_id": session_id, "answer": "x"})
+ assert resp.status_code == 200
+ assert resp.json()["completed"] is False
+ return session_id
+
+
+def test_builder_failure_on_final_answer_is_retryable(
+ client: TestClient, monkeypatch: pytest.MonkeyPatch
+) -> None:
+ calls: list[str] = []
+
+ def _boom(state): # type: ignore[no-untyped-def]
+ calls.append("boom")
+ raise ValueError("could not convert string to float: ''")
+
+ monkeypatch.setattr(profile_builder, "build_and_save_profile", _boom)
+ session_id = _answer_all_but_last(client)
+
+ resp = client.post("/onboard/answer", json={"session_id": session_id, "answer": "final"})
+ assert resp.status_code == 422
+ assert "rephrase" in resp.json()["detail"]
+
+ # The session was rolled back to the last step, not stuck at completed.
+ status = client.get(f"/onboard/status/{session_id}").json()
+ assert status["completed"] is False
+ assert status["current_step"] == TOTAL_STEPS - 1
+
+ # A retry with a working builder completes normally...
+ def _ok(state): # type: ignore[no-untyped-def]
+ calls.append("ok")
+
+ monkeypatch.setattr(profile_builder, "build_and_save_profile", _ok)
+ resp = client.post("/onboard/answer", json={"session_id": session_id, "answer": "final"})
+ assert resp.status_code == 200
+ assert resp.json()["completed"] is True
+ assert calls == ["boom", "ok"]
+
+ # ...and only then does the session refuse further answers.
+ resp = client.post("/onboard/answer", json={"session_id": session_id, "answer": "again"})
+ assert resp.status_code == 400
+
+
+def test_issue_84_answer_completes_through_the_real_builder(
+ client: TestClient, monkeypatch: pytest.MonkeyPatch, tmp_path: Path
+) -> None:
+ """End-to-end: the reporter's exact business-model answer finishes onboarding."""
+ real_build = profile_builder.build_and_save_profile
+
+ def _build_to_tmp(state): # type: ignore[no-untyped-def]
+ return real_build(state, profile_path=tmp_path / "profile.yaml")
+
+ monkeypatch.setattr(profile_builder, "build_and_save_profile", _build_to_tmp)
+ monkeypatch.setattr(profile_builder, "_save_wizard_people", lambda answers: None)
+
+ session_id = client.get("/onboard/start").json()["session_id"]
+ answers = {"business_model": "IT, marketing and video agency"}
+ for step in range(TOTAL_STEPS):
+ from openexecutive.onboarding.wizard import WIZARD_STEPS
+
+ answer = answers.get(WIZARD_STEPS[step]["field"], "x")
+ resp = client.post("/onboard/answer", json={"session_id": session_id, "answer": answer})
+ assert resp.status_code == 200, resp.json()
+
+ assert resp.json()["completed"] is True
+ assert (tmp_path / "profile.yaml").exists()