Fix backend defects, gate CI on lint/tests, tidy packaging + docs

.github/workflows/deploy-pages.yml

@@ -24,6 +24,17 @@ jobs:
         with:
           python-version: "3.12"
 
+      # Gate the deploy on a green suite: `deploy` needs `build`, so a failed
+      # lint or test here blocks publishing to Pages.
+      - name: Install package + dev deps
+        run: pip install -e ".[dev]"
+
+      - name: Lint (ruff)
+        run: ruff check .
+
+      - name: Run unit tests
+        run: pytest co_scientist/tests/unit -q
+
       - name: Export static demo data
         run: python scripts/export_static_demo.py
 

.github/workflows/test.yml

@@ -16,5 +16,7 @@ jobs:
           python-version: "3.12"
       - name: Install package + dev deps
         run: pip install -e ".[dev]"
+      - name: Lint (ruff)
+        run: ruff check .
       - name: Run unit tests
         run: pytest co_scientist/tests/unit -q

co_scientist/agents/generation.py

@@ -47,8 +47,8 @@ async def execute(self, task: Task) -> TaskResult:
         plan = session.research_plan
 
         if strategy != "literature":
-            # M3 ships only the literature strategy.
-            raise NotImplementedError(f"strategy {strategy!r} lands in a later milestone")
+            # Only the literature strategy is scheduled by the supervisor.
+            raise ValueError(f"unsupported strategy: {strategy!r}")
 
         # 1. Render the prompt and run the tool loop with `record_hypothesis` available.
         articles_block = (

co_scientist/agents/reflection.py

@@ -41,7 +41,7 @@ async def execute(self, task: Task) -> TaskResult:
             raise RuntimeError(f"hypothesis {hypothesis_id} missing")
 
         if kind != "full":
-            raise NotImplementedError(f"reflection kind {kind!r} lands in a later milestone")
+            raise ValueError(f"unsupported reflection kind: {kind!r}")
 
         prompt = render(
             "reflection.full",

co_scientist/web/app.py

@@ -9,7 +9,6 @@
 import asyncio
 import logging as stdlib_logging
 import os
-from datetime import UTC, datetime
 from pathlib import Path
 from typing import Any
 
@@ -22,6 +21,7 @@
 from ..config import Config, load_config
 from ..logging import get_logger
 from ..storage import db as db_mod
+from ..storage.repos import hypotheses as hyp_repo
 from ..storage.repos import reviews as rev_repo
 from ..storage.repos import sessions as sess_repo
 from ..storage.repos import transcripts as tx_repo

co_scientist/web/react_api.py

@@ -33,6 +33,10 @@
 
 log = get_logger("react_api")
 
+# Keep strong refs to background session tasks; the event loop only holds a weak
+# ref, so without this a live run can be garbage-collected mid-flight (RUF006).
+_bg_tasks: set[asyncio.Task] = set()
+
 PROVIDERS = [
     {"id": "anthropic", "label": "Anthropic", "models": ["claude-opus-4-7", "claude-sonnet-4-6", "claude-haiku-4-5-20251001"]},
     {"id": "openai", "label": "OpenAI", "models": ["gpt-5", "gpt-4o", "o3-mini"]},
@@ -344,7 +348,9 @@ async def _bg() -> None:
             finally:
                 live_sessions.discard("_pending")
 
-        asyncio.create_task(_bg())
+        _task = asyncio.create_task(_bg())
+        _bg_tasks.add(_task)
+        _task.add_done_callback(_bg_tasks.discard)
 
         # Session row is created within the first seconds of run_session.
         deadline = time.time() + 45.0

co_scientist/web/spa.py

@@ -2,8 +2,6 @@
 
 from __future__ import annotations
 
-from pathlib import Path
-
 from fastapi import APIRouter, HTTPException
 from fastapi.responses import FileResponse
 

pyproject.toml

@@ -4,13 +4,21 @@ build-backend = "hatchling.build"
 
 [project]
 name = "co-scientist"
-version = "0.1.0"
+version = "1.0.0"
 description = "Multi-agent AI Co-Scientist for tournament-style hypothesis generation"
 readme = "README.md"
 requires-python = ">=3.11,<3.14"
 license = { text = "Apache-2.0" }
-authors = [{ name = "Co-Scientist contributors" }]
+authors = [{ name = "Quang Bui" }]
 keywords = ["ai", "agents", "research", "anthropic", "claude"]
+classifiers = [
+    "Development Status :: 5 - Production/Stable",
+    "License :: OSI Approved :: Apache Software License",
+    "Intended Audience :: Science/Research",
+    "Programming Language :: Python :: 3.11",
+    "Programming Language :: Python :: 3.12",
+    "Topic :: Scientific/Engineering :: Artificial Intelligence",
+]
 
 dependencies = [
     # LLM clients (provider is selectable via [llm] provider = ...)
@@ -62,6 +70,11 @@ openai = ["openai>=1.40.0"]
 [project.scripts]
 co-scientist = "co_scientist.cli:app"
 
+[project.urls]
+Homepage = "https://github.com/duckyquang/Co-Scientist"
+Repository = "https://github.com/duckyquang/Co-Scientist"
+Issues = "https://github.com/duckyquang/Co-Scientist/issues"
+
 [tool.hatch.build.targets.wheel]
 packages = ["co_scientist"]
 

scripts/build_bench_report.py

@@ -412,7 +412,7 @@ def _headline_findings_section() -> str:
         "",
         "| model | run 1 Δ Elo | run 2 Δ Elo |",
         "| --- | --- | --- |",
-        "| claude-haiku-4.5 | **+180** (1-9 → 10-0) | **−28** (10-2 → 8-4) |",
+        "| claude-haiku-4.5 | **+180** (1-9 → 10-0) | **-28** (10-2 → 8-4) |",
         "| openai-o1 | +43 | +29 |",
         "",
         "Haiku's *raw* record alone flipped 1-9 → 10-2 across the two runs — "
@@ -427,12 +427,12 @@ def _headline_findings_section() -> str:
         "| gemini-2.5-flash | +172 |",
         "| gemini-2.5-pro | +47 |",
         "| gpt-5 | +26 |",
-        "| gemini-2.0-flash | −48 |",
-        "| gemini-3-flash | −36 |",
-        "| gemini-3-pro | −89 |",
+        "| gemini-2.0-flash | -48 |",
+        "| gemini-3-flash | -36 |",
+        "| gemini-3-pro | -89 |",
         "",
         "Within Google alone the 2.5 models gain (+172, +47) and the 3.x "
-        "models lose (−36, −89) — so there is no clean \"provider\" or "
+        "models lose (-36, -89) — so there is no clean \"provider\" or "
         "\"stronger-model\" story; an earlier draft that claimed one was "
         "reading noise. **The only repeatable signal is openai-o1** (pipeline "
         "modestly ahead in both runs). A real per-model verdict needs many "
@@ -454,9 +454,9 @@ def _headline_findings_section() -> str:
         "| ferroptosis | 3 |",
         "",
         "At the **drug** level it's a long tail of one-offs. The only "
-        "compounds proposed more than once are **Itraconazole** (×5, as an "
-        "OXPHOS inhibitor) and **Auranofin** (×2, thioredoxin-reductase). "
-        "**Venetoclax** appears ×6 but as the resistance/combo context, not "
+        "compounds proposed more than once are **Itraconazole** (x5, as an "
+        "OXPHOS inhibitor) and **Auranofin** (x2, thioredoxin-reductase). "
+        "**Venetoclax** appears x6 but as the resistance/combo context, not "
         "the novel candidate. Tellingly, all three recurrent names already "
         "have prior AML evidence — models default to the familiar, which is "
         "exactly what the strict no-prior-evidence prompt forbids (no "

webapp/seed.py

@@ -161,7 +161,7 @@ def build_session(conn: sqlite3.Connection, *, goal: str, status: str,
     match_t = start + timedelta(minutes=n_hyps * 3 + 5)
     for _round in range(n_rounds):
         random.Random(f"{sid}{_round}").shuffle(in_tournament)
-        for a, b in zip(in_tournament[::2], in_tournament[1::2]):
+        for a, b in zip(in_tournament[::2], in_tournament[1::2], strict=False):
             if a is b:
                 continue
             mode = "debate" if r.random() < 0.3 else "pairwise"
@@ -264,7 +264,6 @@ def build_session(conn: sqlite3.Connection, *, goal: str, status: str,
           {"goal": goal[:200], "n_initial": 3, "budget_usd": budget}, start)
     _transcript(conn, sid, "supervisor", "parse_goal", content.MODELS["supervisor"],
                 start, 0.01)
-    ev_t = start + timedelta(minutes=1)
     for h in hyps:
         agent = "evolution" if h["created_by"] == "evolution" else "generation"
         cost = round(r.uniform(0.04, 0.22), 4)
@@ -307,6 +306,7 @@ def build_session(conn: sqlite3.Connection, *, goal: str, status: str,
         top_hyps = sorted(hyps, key=lambda h: -h["elo"])[:5]
         ov_md = content.make_overview(goal, top_hyps)
         import os
+
         from .store import REPO_ROOT
         ov_dir = REPO_ROOT / "data" / "artifacts" / sid / "final"
         ov_dir.mkdir(parents=True, exist_ok=True)

webapp/server.py

@@ -37,7 +37,7 @@
 ]
 
 
-def _json(handler: "Handler", obj, status=200):
+def _json(handler: Handler, obj, status=200):
     body = json.dumps(obj, default=str).encode()
     handler.send_response(status)
     handler.send_header("Content-Type", "application/json")

webapp/simulator.py

@@ -14,13 +14,13 @@
 import random
 import threading
 import time
-from datetime import UTC, datetime, timedelta
+from datetime import UTC, datetime
 
 from . import content
 from .seed import EXTRA_TABLES, _elo_update, _emit, _transcript
 from .store import REPO_ROOT, connect
 
-_RUNNING: dict[str, "Sim"] = {}
+_RUNNING: dict[str, Sim] = {}
 _LOCK = threading.Lock()
 
 TOKENS_PER_USD = 220_000
@@ -277,7 +277,7 @@ def _ranking(self, conn, rounds: int):
         for _r in range(rounds):
             pool = list(self.hyps)
             self.r.shuffle(pool)
-            for a, b in zip(pool[::2], pool[1::2]):
+            for a, b in zip(pool[::2], pool[1::2], strict=False):
                 if a is b or not self._wait(conn, 1.2):
                     return
                 mode = "debate" if self.r.random() < 0.35 else "pairwise"

webapp/store.py

@@ -58,7 +58,7 @@ def connect(db_path: Path | str = DEFAULT_DB) -> sqlite3.Connection:
 def _rows(conn: sqlite3.Connection, sql: str, params: tuple = ()) -> list[dict]:
     cur = conn.execute(sql, params)
     cols = [c[0] for c in cur.description]
-    return [dict(zip(cols, r)) for r in cur.fetchall()]
+    return [dict(zip(cols, r, strict=True)) for r in cur.fetchall()]
 
 
 def _row(conn: sqlite3.Connection, sql: str, params: tuple = ()) -> dict | None: