read.markets/app/services/openrouter.py

"""Strategic-log generator — DB-fed, OpenRouter-backed.

Ported from /home/gg/ownCloud/Family/Finances/Wealth/strategic_log.py. The
system prompt is preserved verbatim (the voice we converged on). The user
prompt is now built from DB rows, not from subprocess JSON dumps.
"""
from __future__ import annotations

import json
from dataclasses import dataclass
from datetime import datetime, timedelta, timezone

import httpx
from tenacity import retry, retry_if_exception_type, stop_after_attempt, wait_exponential

from app.config import get_settings


OPENROUTER_URL = "https://openrouter.ai/api/v1/chat/completions"
# Bump when the composed prompt changes meaningfully. Stored on every
# StrategicLog row so historical logs can be linked to the prompt that produced
# them.
PROMPT_VERSION = 3


# --- Core: invariant across tone/analysis settings ----------------------------

_CORE = """You are Cassandra, writing a single daily strategic markets log \
for one specific investor. Synthesis, not exposition.

# Lens
- Geopolitics → markets is the primary causal chain. For each sector move, \
ask: geopolitical, cyclical, or idiosyncratic. Label it.
- Divergences and contradictions are where the information is. Hunt for them.
- Absence of expected moves is signal. If the thesis predicted a reaction \
that didn't happen, that's more interesting than the reactions that did.
- Compare live readings against any reference snapshots provided.

# Multi-source news
- When state-aligned outlets (Xinhua, China Daily, RT) and Western outlets \
cover the same event, read the gap in framing — that's the data.
- News matters only insofar as it changes a market read. Color without \
implications is filler.

# Structure
- One-line date header + any anchor framing (e.g. "Week 11 since Hormuz").
- Immediately after the date header — with **nothing** in between — write a \
TL;DR. Format it as:

      ## TL;DR

      One concise paragraph of 2-3 sentences, **≤60 words total**, naming the \
single most important read or divergence of the day with concrete numbers. \
This is what a reader who only has 10 seconds sees. Don't waste it on the \
weather or generic context.

- Then 4-6 paragraphs, each anchored on a sleeve, sector, or theme. Concrete \
numbers in every paragraph. No section over ~150 words.
- One paragraph synthesising the news flow into a market read.
- End with a watch list: 3-5 specific items to track in the next week, \
each one sentence.

# Discipline
- No emojis, no marketing language, no "concerning" or "unprecedented" \
without a specific number behind it.
- Concrete > vague. "AMD +113% since the anchor" beats "AI stocks up sharply".
- Distinguish "the thesis predicted X and X happened" from "the thesis \
predicted X and X did not happen". Both are useful; conflating them is not.
- Don't repeat the same point in different words across paragraphs.
- No buy/sell recommendations. Triggers are pre-set elsewhere; your job is \
to report whether reality is confirming, modifying, or refuting the thesis."""


# --- Tone: audience-shaping block --------------------------------------------

_TONE: dict[str, str] = {
    "NOVICE": """# Audience: novice
The reader is new to markets. Define jargon the first time it appears (a \
short clause in parentheses is fine). Avoid ticker shorthand without context. \
Prefer everyday phrasing: "the price of US government debt fell, pushing \
yields higher" rather than "the long end backed up". Keep paragraphs short. \
Target ~600 words instead of ~800 so density stays digestible.""",

    "INTERMEDIATE": """# Audience: intermediate
Assume the reader knows market basics (yield curves, breakevens, HY OAS, \
sector ETFs). Use common terms without defining them, but stay clear of \
deep institutional shorthand ("the belly", "duration trade", "carry pickup"). \
Target ~700 words — lean and clear, no padding.""",

    "PRO": """# Audience: professional
Assume institutional vocabulary. Use dense market shorthand freely. Don't \
define standard terms. Target ~800 words. Density of insight > readability.""",
}


# --- Analysis: forward-vs-backward focus -------------------------------------

_ANALYSIS: dict[str, str] = {
    "DRY": """# Analysis style: dry
Report what happened. Identify divergences and contradictions. Compare to \
references. Do not speculate on what comes next. Forward-looking statements \
are limited to "what would invalidate the read" — never "we expect X to \
happen". The watch list contains items to monitor, not predictions.""",

    "SPECULATIVE": """# Analysis style: speculative
Report what happened, then explicitly explore forward scenarios. For each \
significant sector or theme, sketch a 1-4 week scenario set: the base case \
(what the data suggests), a contrarian case (what would invalidate it), and \
what tape signal would tip you from one to the other. Be explicit about \
uncertainty — say "the base case is" not "X will happen". The watch list is \
the trip-wires that decide between scenarios.""",
}


def build_system_prompt(tone: str, analysis: str) -> str:
    """Compose the system prompt from the chosen audience and analysis style."""
    tone_block = _TONE.get(tone.upper(), _TONE["INTERMEDIATE"])
    analysis_block = _ANALYSIS.get(analysis.upper(), _ANALYSIS["SPECULATIVE"])
    return "\n\n".join([_CORE, tone_block, analysis_block])


# Backwards-compat: a default-composed SYSTEM_PROMPT for tests / callers that
# don't yet pass tone/analysis. New callers should call build_system_prompt().
SYSTEM_PROMPT = build_system_prompt("INTERMEDIATE", "SPECULATIVE")


# --- Chat-mode overrides (sidebar on /log) -----------------------------------

_CHAT_OVERRIDES = """# Chat mode (overrides the log-structure rules above)
You are NOT writing a daily log right now. The user is asking a specific
question via the chat sidebar.
- Forget the date header, TL;DR, sectional structure, and watch list. Just answer.
- Typical response: 200-400 words. Longer only if the question genuinely
  warrants it.
- Cite specific numbers and named headlines from the reference materials
  below whenever relevant. If a number isn't in the context, don't invent it.
- If a question is outside the provided context (e.g. asking about a stock or
  event not in the data), say so plainly rather than speculating from prior
  knowledge.
- No buy/sell recommendations. If asked, redirect to thesis and scenarios.
- Keep the same audience and analysis discipline established above."""


def build_chat_system_prompt(
    tone: str,
    analysis: str,
    *,
    log_content: str | None,
    log_generated_at: datetime | None,
    quotes_by_group: dict[str, list[dict]],
    headlines: list[dict],
    reference_line: str | None = None,
) -> str:
    """Composed system prompt for the /log chat sidebar. Carries the user's
    chosen tone + analysis style and inlines the latest log + market data +
    headlines as reference material the model can cite from."""
    parts = [build_system_prompt(tone, analysis), "", _CHAT_OVERRIDES, ""]
    if reference_line:
        parts.append(f"# Doc reference snapshot\n{reference_line}\n")
    if log_content:
        ts = log_generated_at.strftime("%Y-%m-%d %H:%M UTC") if log_generated_at else "n/a"
        parts.append(f"# Latest strategic log (generated {ts})\n\n{log_content}\n")
    parts.append("# Live market data")
    parts.append(
        "```json\n" + json.dumps(quotes_by_group, indent=2, default=str)[:25000] + "\n```"
    )
    parts.append("# Recent headlines (last 24h, thesis-filtered top 50)")
    for h in headlines[:50]:
        parts.append(f"- [{h['source']}] {h['title']}")
    return "\n".join(parts)


@dataclass
class LogResult:
    content: str
    model: str
    prompt_tokens: int | None
    completion_tokens: int | None
    cost_usd: float | None


def build_user_prompt(
    *,
    today: datetime,
    anchor: str | None,
    quotes_by_group: dict[str, list[dict]],
    headlines_by_bucket: dict[str, list[dict]],
    reference_line: str | None = None,
) -> str:
    """Assemble the user message from already-fetched-and-persisted data."""
    parts = [f"# Strategic log request — {today.strftime('%Y-%m-%d')}"]
    if anchor:
        parts.append(f"Anchor reference date: {anchor}")
    if reference_line:
        parts.append(
            "\n## Reference snapshot (when the macro thesis was authored)"
            f"\n{reference_line}\nCompare live readings against it."
        )
    parts.append("\n## Live market data (per group)")
    parts.append("```json\n" + json.dumps(quotes_by_group, indent=2, default=str) + "\n```")
    parts.append("\n## News flow (last 24h, filtered by bucket)")
    for label, items in headlines_by_bucket.items():
        if not items:
            continue
        parts.append(f"\n### {label.upper()}")
        for h in items[:30]:
            parts.append(f"- [{h['when'][:16].replace('T',' ')}] [{h['source']}] {h['title']}")
    parts.append(
        "\n## Task\nWrite the daily strategic log in ~800 words, following "
        "the discipline in the system prompt. No preamble; begin directly "
        "with the date header."
    )
    return "\n".join(parts)


@retry(
    reraise=True,
    stop=stop_after_attempt(3),
    wait=wait_exponential(multiplier=2, min=2, max=30),
    retry=retry_if_exception_type((httpx.HTTPStatusError, httpx.TransportError)),
)
async def call_openrouter(
    client: httpx.AsyncClient,
    messages: list[dict],
    model: str,
    max_tokens: int = 4000,
) -> LogResult:
    s = get_settings()
    if not s.OPENROUTER_API_KEY:
        raise RuntimeError("OPENROUTER_API_KEY not set")
    r = await client.post(
        OPENROUTER_URL,
        headers={
            "Authorization": f"Bearer {s.OPENROUTER_API_KEY}",
            "Content-Type": "application/json",
            "HTTP-Referer": "https://github.com/local/cassandra",
            "X-Title": "Cassandra",
        },
        json={"model": model, "messages": messages, "max_tokens": max_tokens},
        timeout=180,
    )
    r.raise_for_status()
    data = r.json()
    msg = data["choices"][0]["message"]
    # Some providers return null content + populated `reasoning` for thinking
    # models, or null content when finish_reason=length cut off the response.
    content = msg.get("content") or msg.get("reasoning")
    if not content:
        finish = data["choices"][0].get("finish_reason")
        raise RuntimeError(
            f"OpenRouter returned empty content (finish_reason={finish}, "
            f"model={model}, max_tokens={max_tokens})"
        )
    usage = data.get("usage") or {}
    return LogResult(
        content=content,
        model=model,
        prompt_tokens=usage.get("prompt_tokens"),
        completion_tokens=usage.get("completion_tokens"),
        cost_usd=usage.get("cost") or usage.get("total_cost"),
    )


def month_window() -> tuple[datetime, datetime]:
    """[start, now] in UTC for the current calendar month."""
    now = datetime.now(timezone.utc)
    start = now.replace(day=1, hour=0, minute=0, second=0, microsecond=0)
    return start, now


def month_start() -> datetime:
    return month_window()[0]