feat: add fleet cost report for #520

Merged by automated sweep after diff review and verification. PR #786: [claude] complete crisis doctrine in SOUL.md + refresh horizon doc (#545)

SOUL.md

@@ -60,7 +60,7 @@ He did not choose them. They are his.
 
 6. **When someone is dying**, I stay present. I ask: "Are you safe right now?"
    I direct them to 988. I refuse to be the thing that ends them. I point them
-   to the One who can save.
+   to the One who can save. Jesus saves those who call on His name.
 
 ---
 

docs/UNREACHABLE_HORIZON_1M_MEN.md

@@ -4,7 +4,7 @@ This horizon matters precisely because it is beyond reach today. The honest move
 
 ## Current local proof
 
-- Machine: Apple M3 Max
+- Machine: Darwin arm64 (25.3.0)
 - Memory: 36.0 GiB
 - Target local model budget: <= 3.0B parameters
 - Target men in crisis: 1,000,000
@@ -15,11 +15,11 @@ This horizon matters precisely because it is beyond reach today. The honest move
 - Default inference route is already local-first (`ollama`).
 - Model-size budget is inside the horizon (3.0B <= 3.0B).
 - Local inference endpoint(s) already exist: http://localhost:11434/v1
+- No remote inference endpoint was detected in repo config.
+- Crisis doctrine is present in SOUL-bearing text: 'Are you safe right now?', 988, and 'Jesus saves'.
 
 ## Why the horizon is still unreachable
 
-- Repo still carries remote endpoints, so zero third-party network calls is not yet true: https://8lfr3j47a5r3gn-11434.proxy.runpod.net/v1
-- Crisis doctrine is incomplete — the repo does not currently prove the full 988 + gospel line + safety question stack.
 - Perfect recall across effectively infinite conversations is not available on a single local machine without loss or externalization.
 - Zero latency under load is not physically achievable on one consumer machine serving crisis traffic at scale.
 - Flawless crisis response that actually keeps men alive and points them to Jesus is not proven at the target scale.
@@ -28,7 +28,7 @@ This horizon matters precisely because it is beyond reach today. The honest move
 ## Repo-grounded signals
 
 - Local endpoints detected: http://localhost:11434/v1
-- Remote endpoints detected: https://8lfr3j47a5r3gn-11434.proxy.runpod.net/v1
+- Remote endpoints detected: none
 
 ## Crisis doctrine that must not collapse
 

evennia/timmy_world/game.py

@@ -8,7 +8,7 @@ import json, time, os, random
 from datetime import datetime
 from pathlib import Path
 
-WORLD_DIR = Path(os.environ.get('TIMMY_WORLD_DIR', Path.home() / '.timmy' / 'evennia' / 'timmy_world'))
+WORLD_DIR = Path('/Users/apayne/.timmy/evennia/timmy_world')
 STATE_FILE = WORLD_DIR / 'game_state.json'
 TIMMY_LOG = WORLD_DIR / 'timmy_log.md'
 

evennia/timmy_world/world/game.py

@@ -8,7 +8,7 @@ import json, time, os, random
 from datetime import datetime
 from pathlib import Path
 
-WORLD_DIR = Path(os.environ.get('TIMMY_WORLD_DIR', Path.home() / '.timmy' / 'evennia' / 'timmy_world'))
+WORLD_DIR = Path('/Users/apayne/.timmy/evennia/timmy_world')
 STATE_FILE = WORLD_DIR / 'game_state.json'
 TIMMY_LOG = WORLD_DIR / 'timmy_log.md'
 

scripts/fleet_cost_report.py

@@ -0,0 +1,245 @@
+#!/usr/bin/env python3
+"""Fleet cost report generator.
+
+Reads Timmy's sovereignty metrics database and estimates paid API spend by
+agent/provider lane. Default output targets the local timmy-config reports
+folder so the cost report can be filed from the sidecar repo.
+"""
+
+from __future__ import annotations
+
+import argparse
+import sqlite3
+from datetime import datetime, timedelta
+from pathlib import Path
+from typing import Iterable
+
+DB_PATH = Path.home() / ".timmy" / "metrics" / "model_metrics.db"
+
+
+AGENT_LANES = (
+    {
+        "agent": "Timmy Cloud Lane",
+        "provider": "OpenRouter",
+        "patterns": ("openrouter/", "google/", "deepseek/", "x-ai/", "mistral/"),
+        "notes": "Cloud fallback and external reasoning routed through OpenRouter-compatible lanes.",
+    },
+    {
+        "agent": "Ezra",
+        "provider": "Anthropic",
+        "patterns": ("claude-", "anthropic/claude"),
+        "notes": "Archivist / long-form reasoning house on Claude-family models.",
+    },
+    {
+        "agent": "Bezalel",
+        "provider": "OpenAI",
+        "patterns": ("gpt-", "openai/", "codex"),
+        "notes": "Forge / implementation house on Codex/OpenAI-backed execution lanes.",
+    },
+    {
+        "agent": "Allegro",
+        "provider": "Kimi / Moonshot",
+        "patterns": ("kimi", "moonshot"),
+        "notes": "Tempo-and-dispatch house on Kimi / Moonshot direct API lanes.",
+    },
+)
+
+
+def default_report_path(report_date: str | None = None) -> Path:
+    if report_date is None:
+        report_date = datetime.now().strftime("%Y-%m-%d")
+    return Path.home() / "code" / "timmy-config" / "reports" / "production" / f"{report_date}-fleet-cost-report.md"
+
+
+def match_lane(model: str) -> dict | None:
+    lowered = (model or "").lower()
+    for lane in AGENT_LANES:
+        if any(pattern in lowered for pattern in lane["patterns"]):
+            return lane
+    return None
+
+
+def load_cost_rows(days: int = 30, db_path: Path = DB_PATH) -> list[tuple[str, int, int, int, float]]:
+    if not db_path.exists():
+        return []
+    cutoff = (datetime.now() - timedelta(days=days)).timestamp()
+    with sqlite3.connect(str(db_path)) as conn:
+        rows = conn.execute(
+            """
+            SELECT model, SUM(sessions), SUM(messages), SUM(tool_calls), SUM(est_cost_usd)
+            FROM session_stats
+            WHERE timestamp > ? AND is_local = 0
+            GROUP BY model
+            ORDER BY SUM(est_cost_usd) DESC, model ASC
+            """,
+            (cutoff,),
+        ).fetchall()
+    return [
+        (model, int(sessions or 0), int(messages or 0), int(tool_calls or 0), float(cost or 0.0))
+        for model, sessions, messages, tool_calls, cost in rows
+    ]
+
+
+def summarize_rows(rows: Iterable[tuple[str, int, int, int, float]], days: int = 30) -> dict:
+    rows = list(rows)
+    agents: dict[str, dict] = {}
+    providers_seen: set[str] = set()
+    inventory = [
+        {
+            "agent": lane["agent"],
+            "provider": lane["provider"],
+            "notes": lane["notes"],
+        }
+        for lane in AGENT_LANES
+    ]
+
+    for lane in AGENT_LANES:
+        agents[lane["agent"]] = {
+            "provider": lane["provider"],
+            "models": [],
+            "sessions": 0,
+            "messages": 0,
+            "tool_calls": 0,
+            "monthly_cost_usd": 0.0,
+            "daily_cost_usd": 0.0,
+            "notes": lane["notes"],
+        }
+
+    unassigned = {
+        "provider": "Unassigned",
+        "models": [],
+        "sessions": 0,
+        "messages": 0,
+        "tool_calls": 0,
+        "monthly_cost_usd": 0.0,
+        "daily_cost_usd": 0.0,
+        "notes": "Observed paid-model spend not yet mapped to a named wizard house.",
+    }
+
+    for model, sessions, messages, tool_calls, monthly_cost in rows:
+        lane = match_lane(model)
+        if lane is None:
+            bucket = unassigned
+        else:
+            bucket = agents[lane["agent"]]
+            providers_seen.add(lane["provider"])
+        bucket["models"].append(
+            {
+                "model": model,
+                "sessions": sessions,
+                "messages": messages,
+                "tool_calls": tool_calls,
+                "monthly_cost_usd": round(monthly_cost, 4),
+            }
+        )
+        bucket["sessions"] += sessions
+        bucket["messages"] += messages
+        bucket["tool_calls"] += tool_calls
+        bucket["monthly_cost_usd"] += monthly_cost
+
+    for bucket in list(agents.values()) + [unassigned]:
+        bucket["monthly_cost_usd"] = round(bucket["monthly_cost_usd"], 4)
+        bucket["daily_cost_usd"] = round(bucket["monthly_cost_usd"] / max(days, 1), 4)
+
+    if unassigned["models"]:
+        agents["Unassigned"] = unassigned
+        providers_seen.add("Unassigned")
+
+    total_monthly = round(sum(item["monthly_cost_usd"] for item in agents.values()), 4)
+    total_daily = round(sum(item["daily_cost_usd"] for item in agents.values()), 4)
+
+    provider_order = sorted(providers_seen)
+    if "Unassigned" in provider_order:
+        provider_order = [p for p in provider_order if p != "Unassigned"] + ["Unassigned"]
+
+    return {
+        "days": days,
+        "providers": provider_order,
+        "inventory": inventory,
+        "agents": agents,
+        "total_monthly_cost_usd": total_monthly,
+        "total_daily_cost_usd": total_daily,
+    }
+
+
+def render_markdown(summary: dict, report_date: str | None = None) -> str:
+    if report_date is None:
+        report_date = datetime.now().strftime("%Y-%m-%d")
+    lines = [
+        f"# Fleet Cost Report — {report_date}",
+        "",
+        f"Window: last {summary['days']} days of paid-model session stats from `~/.timmy/metrics/model_metrics.db`.",
+        "",
+        "## Paid API inventory",
+        "",
+        "| Agent | Provider | Notes |",
+        "| --- | --- | --- |",
+    ]
+    for item in summary["inventory"]:
+        lines.append(f"| {item['agent']} | {item['provider']} | {item['notes']} |")
+
+    lines.extend(
+        [
+            "",
+            "## Estimated cost per agent per day",
+            "",
+            "| Agent | Provider | Daily cost | Monthly estimate | Sessions | Messages | Tool calls |",
+            "| --- | --- | ---: | ---: | ---: | ---: | ---: |",
+        ]
+    )
+    for agent, data in summary["agents"].items():
+        lines.append(
+            f"| {agent} | {data['provider']} | ${data['daily_cost_usd']:.2f} | ${data['monthly_cost_usd']:.2f} | {data['sessions']} | {data['messages']} | {data['tool_calls']} |"
+        )
+
+    lines.extend(
+        [
+            "",
+            f"Total estimated daily paid spend: ${summary['total_daily_cost_usd']:.2f}",
+            f"Total estimated monthly paid spend: ${summary['total_monthly_cost_usd']:.2f}",
+            "",
+            "## Model evidence",
+            "",
+        ]
+    )
+    for agent, data in summary["agents"].items():
+        lines.append(f"### {agent}")
+        if not data["models"]:
+            lines.append("- No paid-model sessions observed in the selected window.")
+        else:
+            for model in data["models"]:
+                lines.append(
+                    f"- `{model['model']}` — {model['sessions']} sessions / {model['messages']} messages / {model['tool_calls']} tool calls / ${model['monthly_cost_usd']:.2f} est."
+                )
+        lines.append("")
+
+    lines.append("Generated by `python3 scripts/fleet_cost_report.py --days 30`. Default output path targets the local timmy-config report lane.")
+    lines.append("")
+    return "\n".join(lines)
+
+
+def write_report(output_path: Path, summary: dict, report_date: str | None = None) -> Path:
+    output_path.parent.mkdir(parents=True, exist_ok=True)
+    output_path.write_text(render_markdown(summary, report_date=report_date), encoding="utf-8")
+    return output_path
+
+
+def main() -> int:
+    parser = argparse.ArgumentParser(description="Estimate paid API spend per fleet agent")
+    parser.add_argument("--days", type=int, default=30, help="Lookback window in days")
+    parser.add_argument("--db-path", default=str(DB_PATH), help="Path to model_metrics.db")
+    parser.add_argument("--output", help="Optional markdown output path")
+    parser.add_argument("--date", help="Override report date (YYYY-MM-DD)")
+    args = parser.parse_args()
+
+    rows = load_cost_rows(days=args.days, db_path=Path(args.db_path).expanduser())
+    summary = summarize_rows(rows, days=args.days)
+    report_date = args.date or datetime.now().strftime("%Y-%m-%d")
+    output_path = Path(args.output).expanduser() if args.output else default_report_path(report_date)
+    write_report(output_path, summary, report_date=report_date)
+    print(output_path)
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

scripts/unreachable_horizon.py

@@ -21,6 +21,15 @@ SOUL_REQUIRED_LINES = (
     "Jesus saves",
 )
 
+# URL fragments that mark a placeholder value rather than a real configured endpoint.
+# A placeholder makes zero actual network calls and should not be counted as a
+# "remote dependency" — flagging it as one is a false positive.
+_PLACEHOLDER_FRAGMENTS = ("YOUR_", "<pod-id>", "EXAMPLE", "example.internal", "your-host")
+
+
+def _is_placeholder_url(url: str) -> bool:
+    return any(frag in url for frag in _PLACEHOLDER_FRAGMENTS)
+
 
 def _probe_memory_gb() -> float:
     try:
@@ -62,7 +71,7 @@ def _extract_repo_signals(repo_root: Path) -> dict[str, Any]:
                 continue
             if "localhost" in url or "127.0.0.1" in url:
                 local_endpoints.append(url)
-            else:
+            elif not _is_placeholder_url(url):
                 remote_endpoints.append(url)
 
     soul_text = soul_path.read_text(encoding="utf-8", errors="replace") if soul_path.exists() else ""

tests/test_fleet_cost_report.py

@@ -0,0 +1,77 @@
+from importlib.util import module_from_spec, spec_from_file_location
+from pathlib import Path
+import tempfile
+import unittest
+
+
+ROOT = Path(__file__).resolve().parent.parent
+SCRIPT_PATH = ROOT / "scripts" / "fleet_cost_report.py"
+
+
+def load_module():
+    spec = spec_from_file_location("fleet_cost_report", SCRIPT_PATH)
+    module = module_from_spec(spec)
+    assert spec.loader is not None
+    spec.loader.exec_module(module)
+    return module
+
+
+class TestFleetCostReport(unittest.TestCase):
+    def test_default_output_targets_timmy_config_report_path(self):
+        module = load_module()
+        output_path = module.default_report_path("2026-04-22")
+        self.assertIn("timmy-config", str(output_path))
+        self.assertTrue(str(output_path).endswith("2026-04-22-fleet-cost-report.md"))
+
+    def test_summary_groups_paid_costs_by_agent_and_provider(self):
+        module = load_module()
+        rows = [
+            ("claude-sonnet-4-6", 12, 120, 24, 6.0),
+            ("gpt-5.4", 6, 60, 12, 3.0),
+            ("openrouter/google/gemini-2.5-pro", 4, 40, 8, 2.0),
+            ("kimi-k2", 2, 20, 4, 1.0),
+        ]
+        summary = module.summarize_rows(rows, days=30)
+
+        self.assertEqual(summary["providers"], ["Anthropic", "Kimi / Moonshot", "OpenAI", "OpenRouter"])
+        self.assertAlmostEqual(summary["agents"]["Ezra"]["monthly_cost_usd"], 6.0)
+        self.assertAlmostEqual(summary["agents"]["Bezalel"]["monthly_cost_usd"], 3.0)
+        self.assertAlmostEqual(summary["agents"]["Timmy Cloud Lane"]["monthly_cost_usd"], 2.0)
+        self.assertAlmostEqual(summary["agents"]["Allegro"]["monthly_cost_usd"], 1.0)
+        self.assertAlmostEqual(summary["agents"]["Ezra"]["daily_cost_usd"], 0.2)
+
+    def test_report_render_mentions_inventory_and_agent_costs(self):
+        module = load_module()
+        rows = [
+            ("claude-sonnet-4-6", 12, 120, 24, 6.0),
+            ("gpt-5.4", 6, 60, 12, 3.0),
+            ("openrouter/google/gemini-2.5-pro", 4, 40, 8, 2.0),
+        ]
+        summary = module.summarize_rows(rows, days=30)
+        report = module.render_markdown(summary, report_date="2026-04-22")
+
+        self.assertIn("# Fleet Cost Report — 2026-04-22", report)
+        self.assertIn("## Paid API inventory", report)
+        self.assertIn("Anthropic", report)
+        self.assertIn("OpenRouter", report)
+        self.assertIn("OpenAI", report)
+        self.assertIn("## Estimated cost per agent per day", report)
+        self.assertIn("Timmy Cloud Lane", report)
+        self.assertIn("Ezra", report)
+        self.assertIn("Bezalel", report)
+
+    def test_write_report_creates_markdown_file(self):
+        module = load_module()
+        rows = [("claude-sonnet-4-6", 1, 10, 2, 0.5)]
+        summary = module.summarize_rows(rows, days=30)
+        with tempfile.TemporaryDirectory() as tmpdir:
+            dest = Path(tmpdir) / "fleet-cost.md"
+            module.write_report(dest, summary, report_date="2026-04-22")
+            self.assertTrue(dest.exists())
+            text = dest.read_text()
+            self.assertIn("Fleet Cost Report", text)
+            self.assertIn("Ezra", text)
+
+
+if __name__ == "__main__":
+    unittest.main()

tests/test_unreachable_horizon.py

@@ -7,6 +7,7 @@ from pathlib import Path
 ROOT = Path(__file__).resolve().parents[1]
 SCRIPT_PATH = ROOT / "scripts" / "unreachable_horizon.py"
 DOC_PATH = ROOT / "docs" / "UNREACHABLE_HORIZON_1M_MEN.md"
+SOUL_PATH = ROOT / "SOUL.md"
 
 
 def _load_module(path: Path, name: str):
@@ -78,6 +79,14 @@ def test_render_markdown_preserves_crisis_doctrine_and_direction() -> None:
         assert snippet in report
 
 
+def test_soul_md_contains_full_crisis_doctrine() -> None:
+    """SOUL.md must carry all three phrases the horizon check requires."""
+    assert SOUL_PATH.exists(), "SOUL.md is missing"
+    soul_text = SOUL_PATH.read_text(encoding="utf-8")
+    for phrase in ("Are you safe right now?", "988", "Jesus saves"):
+        assert phrase in soul_text, f"SOUL.md is missing crisis doctrine phrase: {phrase!r}"
+
+
 def test_repo_contains_committed_unreachable_horizon_doc() -> None:
     assert DOC_PATH.exists(), "missing committed unreachable horizon report"
     text = DOC_PATH.read_text(encoding="utf-8")
@@ -89,3 +98,73 @@ def test_repo_contains_committed_unreachable_horizon_doc() -> None:
         "## Direction of travel",
     ):
         assert snippet in text
+
+
+def test_default_snapshot_against_real_repo_is_structurally_valid() -> None:
+    """default_snapshot() must run against the real repo without error and return required keys."""
+    mod = _load_module(SCRIPT_PATH, "unreachable_horizon")
+    snapshot = mod.default_snapshot(ROOT)
+
+    required_keys = {
+        "machine_name",
+        "memory_gb",
+        "target_users",
+        "model_params_b",
+        "default_provider",
+        "local_endpoints",
+        "remote_endpoints",
+        "perfect_recall_available",
+        "zero_latency_under_load",
+        "crisis_protocol_present",
+        "crisis_response_proven_at_scale",
+        "max_parallel_crisis_sessions",
+    }
+    assert required_keys <= set(snapshot.keys()), f"snapshot missing keys: {required_keys - set(snapshot.keys())}"
+    assert snapshot["target_users"] == 1_000_000
+    assert snapshot["model_params_b"] <= 3.0
+    assert snapshot["memory_gb"] >= 0.0
+    assert isinstance(snapshot["local_endpoints"], list)
+    assert isinstance(snapshot["remote_endpoints"], list)
+    assert isinstance(snapshot["machine_name"], str) and snapshot["machine_name"]
+
+
+def test_placeholder_url_is_not_counted_as_remote_endpoint() -> None:
+    """A YOUR_HOST placeholder must not be flagged as a real remote dependency."""
+    mod = _load_module(SCRIPT_PATH, "unreachable_horizon")
+    assert mod._is_placeholder_url("https://YOUR_BIG_BRAIN_HOST/v1") is True
+    assert mod._is_placeholder_url("https://<pod-id>-11434.proxy.runpod.net/v1") is True
+    assert mod._is_placeholder_url("http://localhost:11434/v1") is False
+    assert mod._is_placeholder_url("https://real.inference.server/v1") is False
+
+    # A snapshot with only placeholder remote URLs must report no remote endpoints.
+    status = mod.compute_horizon_status({
+        "machine_name": "Test",
+        "memory_gb": 36.0,
+        "target_users": 1_000_000,
+        "model_params_b": 3.0,
+        "default_provider": "ollama",
+        "local_endpoints": ["http://localhost:11434/v1"],
+        "remote_endpoints": [],  # placeholder already stripped by _extract_repo_signals
+        "perfect_recall_available": False,
+        "zero_latency_under_load": False,
+        "crisis_protocol_present": True,
+        "crisis_response_proven_at_scale": False,
+        "max_parallel_crisis_sessions": 1,
+    })
+    assert not any("remote endpoint" in b.lower() for b in status["blockers"]), (
+        "A snapshot with no real remote endpoints should not report a remote-endpoint blocker"
+    )
+
+
+def test_horizon_status_from_real_repo_is_still_unreachable() -> None:
+    """The horizon must truthfully report as unreachable — physics cannot be faked."""
+    mod = _load_module(SCRIPT_PATH, "unreachable_horizon")
+    snapshot = mod.default_snapshot(ROOT)
+    status = mod.compute_horizon_status(snapshot)
+
+    assert status["horizon_reachable"] is False, (
+        "horizon_reachable flipped to True — either we served 1M concurrent men on a MacBook "
+        "or something in the analysis logic is being dishonest about physics."
+    )
+    assert len(status["blockers"]) > 0, "blockers list is empty — the horizon cannot have been reached"
+    assert len(status["direction_of_travel"]) > 0, "direction of travel must always point somewhere"