caelum0x 03c7e2c397 fix(license): consistent Apache-2.0/MIT split + migrate 3 examples from private audit
- Root LICENSE: replace the short-form notice with the full canonical
  Apache-2.0 text (fixes GitHub's "Other" license detection); update the
  OpenJev -> Umpirix copyright line in LICENSE/NOTICE, otsone/NOTICE and
  sdk/python/LICENSE+NOTICE (missed by the rebrand pass).
- sdk/ai-sdk: was MIT with no LICENSE/NOTICE and no upstream fork trail
  (unlike sdk/js, sdk/python, adapters/python, skills, which are real
  TypeSafe AI SDK forks and correctly keep MIT per their NOTICE). It is
  original Umpirix code, so switch it to Apache-2.0 with its own
  LICENSE/NOTICE, matching the root.
- README Licensing section: list sdk/ai-sdk under Apache-2.0 and
  sdk/scribase-decisions under MIT, and spell out why the TypeSafe-forked
  directories stay MIT (their MIT terms require preserving the notice).
- Audited caelum0x/openjev-private (read-only clone): it's an older,
  mostly-superseded snapshot of this repo. Migrated the 3 self-contained,
  secret-free example scripts it had that this repo didn't (raw-HTTP, no
  SDK dependency): content-moderation, lead-scoring, support-ticket-routing
  (Python + Node.js, each with a sample JSONL), plus
  scripts/reproduce-leaderboard.mjs (wired to a new `npm run
  bench:reproduce`) for leaderboard reproducibility. Left behind in
  private: src/lib/seo/retired.ts (a reversed decision - this repo now has
  live /compare pages) and the hosted-SaaS pieces already superseded by
  this repo's own, newer versions of the same files.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
2026-11-11 19:08:57 +03:00

96 lines
4.1 KiB
Python

"""Lead scoring with OpenJev typed decisions (Python, standard library only).
Each inbound lead (company + message) gets three typed judgments in one request: buying
stage (score), company segment (choice) and whether it is a real inquiry or a vendor
pitch (choice). The points formula is ordinary code, so sales can re-weight it without
re-running inference, and uncertain leads are flagged for a human look.
npm run dev # in the repo root (or: docker compose up -> port 3000)
OPENJEV_BASE_URL=http://localhost:8080 python3 examples/lead-scoring/score.py
"""
import json
import os
import urllib.request
from pathlib import Path
BASE_URL = os.environ.get("OPENJEV_BASE_URL", "http://localhost:8080").rstrip("/")
MODEL = os.environ.get("OPENJEV_MODEL", "system-one")
KEY = os.environ.get("OPENJEV_API_KEY")
STAGES = ["just curious", "researching", "evaluating vendors", "ready to buy"]
QUESTIONS = {
"stage": {"type": "score", "instructions": "How close is this lead to buying?", "criteria": STAGES},
"segment": {
"type": "choice",
"instructions": "What kind of organisation is this lead from?",
"criteria": {
"enterprise": "large company, hundreds or thousands of employees, procurement",
"smb": "small or mid-sized business, startup, SaaS team",
"individual": "hobbyist, indie developer, student, personal email",
},
},
"kind": {
"type": "choice",
"instructions": "What is this inbound message?",
"criteria": {
"inquiry": "a potential customer asking about our product, pricing, demo, evaluation or use",
"vendor_pitch": "someone advertising their own services to us: SEO, outsourcing, developers for hire, marketing",
},
},
}
# Scoring policy: plain code, easy to re-weight.
SEGMENT_POINTS = {"enterprise": 30, "smb": 20, "individual": 0}
PITCH_MIN_CONFIDENCE = 0.7 # only a confident "vendor_pitch" marks a lead as junk
REVIEW_BELOW = 0.7 # uncertain segment or kind -> a person takes a look
def grade(points: int) -> str:
if points >= 70:
return "hot"
if points >= 45:
return "warm"
return "cold"
def judge(lead: dict) -> dict:
state = f"Company: {lead['company']}\nMessage: {lead['message']}"
headers = {"content-type": "application/json"}
if KEY:
headers["authorization"] = f"Bearer {KEY}"
body = json.dumps({"model": MODEL, "state": state, "questions": QUESTIONS}).encode()
req = urllib.request.Request(f"{BASE_URL}/v1/systemone", data=body, headers=headers, method="POST")
with urllib.request.urlopen(req, timeout=30) as res:
return json.load(res)
def main() -> None:
lines = (Path(__file__).parent / "leads.jsonl").read_text().splitlines()
leads = [json.loads(line) for line in lines if line.strip()]
ranked = []
agree = 0
engine = "unknown"
for lead in leads:
out = judge(lead)
engine = out.get("engine", {}).get("kind", engine)
stage, segment, kind = (out["answers"][k] for k in ("stage", "segment", "kind"))
# stage["score"] is the expected position on the 0..3 scale, already a smooth value.
stage_points = stage["score"] / (len(STAGES) - 1) * 70
points = round(stage_points + SEGMENT_POINTS.get(segment["choice"], 0))
is_pitch = kind["choice"] == "vendor_pitch" and kind["confidence"] >= PITCH_MIN_CONFIDENCE
label = "junk" if is_pitch else grade(points)
review = segment["confidence"] < REVIEW_BELOW or kind["confidence"] < REVIEW_BELOW or stage["ood"] >= 0.6
agree += label == lead["expected"]
ranked.append((-1 if label == "junk" else points, lead, label, review))
ranked.sort(key=lambda r: r[0], reverse=True)
for points, lead, label, review in ranked:
flag = " [check by hand]" if review else ""
miss = "" if label == lead["expected"] else f" (labelled {lead['expected']})"
print(f"{lead['id']} {label:<5} {max(points, 0):>3} pts {lead['company']}{flag}{miss}")
print(f"\nengine={engine} agrees with the hand labels on {agree}/{len(leads)} sample leads")
if __name__ == "__main__":
main()