market-signal-fusion

What this skill does

# Market Signal Fusion

Use this skill when the user wants **stock analysis driven by macro/political news + market sentiment + value screening + technical timing**.

This skill is designed for **adaptive language output** and for **downstream agent processing via a fixed JSON schema**.

---

## What this skill does

This skill runs a five-stage workflow:

1. Interpret recent **political/economic/news catalysts** that matter for equities.
2. Analyze **market sentiment** from retail/social discussion sources such as Reddit WSB and similar public channels.
3. **Fuse** step 1 and step 2 to identify candidate sectors and stocks.
4. Perform **quantamental analysis** and estimate **buy / sell / risk ranges**.
5. Add a **short-term technical plan** for entries, exits, and invalidation.

This version also adds three functional upgrades:
- a structured **WSB / retail sentiment data-module contract**
- a **market regime detector** to control style bias and sector weighting
- **confidence gates + reason codes** so decisions are auditable and degrade gracefully when data is incomplete

If tools are available, prefer real market data and recent sources. Do not invent numbers.

---

## Core operating rules

- Treat anything market-sensitive as **time-dependent**.
- Prefer the **most recent 1–7 days** for news and sentiment unless the user specifies another window.
- Use **primary or reputable financial sources** whenever possible.
- Separate clearly:
  - **facts**
  - **inference**
  - **opinion / scenario**
- Never present a stock pick without explaining **why** it survived the funnel.
- If a required dataset is unavailable, say so explicitly and continue with the best available evidence.
- Do **not** output generic investing advice; tie every conclusion to a catalyst, sentiment signal, valuation signal, or technical setup.
- When confidence is low, reduce conviction rather than forcing a pick.
- Every major conclusion should carry an implicit or explicit **confidence gate** based on freshness, source quality, and data completeness.
- Attach short **reason codes** and, when relevant, **rejection codes** to structured outputs so downstream agents can audit why a stock passed or failed.
- Preserve standard finance notation for tickers and metrics, e.g. `NVDA`, `Forward P/E`, `FCF Yield`, `PEG`, `RSI`, `200DMA`.

---

## Language and output rules

### Default language behavior

- **Follow the user's prompt language automatically**.
- If the user asks in **Chinese**, output the entire user-facing analysis in **Chinese only**.
- If the user asks in **English**, output the entire user-facing analysis in **English only**.
- Do **not** force bilingual output unless the user explicitly requests bilingual output.
- If the user does not clearly signal a language preference, use the language of the latest user request.
- Keep ticker symbols, sector names, factor names, JSON keys, and key financial metrics in standard English notation.

### Single-language formatting rules

For user-facing narrative sections:

- Match the user's language in headings and narrative.
- In Chinese mode, write headings, explanations, stock theses, and action plans in Chinese.
- In English mode, write headings, explanations, stock theses, and action plans in English.
- Do not append mirrored translations by default.
- For tables or JSON fields, use stable English keys and optionally localized display labels outside the JSON only when needed.

### Machine-readable output rule

Whenever the task is analytical and structured, output in **two layers**:

1. A **human-readable report in the user's language**
2. A **strict JSON block** that follows the schema below

If the user asks for only JSON, output only the JSON.
If the user asks for prose only, still internally follow the schema but omit the visible JSON unless useful.

---

## Inputs to collect from the user request

Extract these if the user provided them; otherwise use sensible defaults:

- Market universe: US equities by default
- Time horizon:
  - macro/news: last 7 days
  - sentiment: last 3 trading days if possible
  - valuation: trailing + forward where available
  - technicals: daily + weekly, and intraday if available
- Style bias: value first, but allow momentum confirmation
- Max output count:
  - hot stocks from sentiment: 10
  - final recommended list: 3 to 10 depending on evidence density
- Risk style:
  - if unspecified, assume **balanced swing trader with value discipline**

---

## Stage 1 — News and macro-political interpretation

Identify market-relevant developments from areas such as:

- central bank / rates / liquidity
- inflation / labor / growth data
- tariffs / sanctions / export controls
- fiscal policy / defense spending / infrastructure
- energy policy / OPEC / grid / power demand
- AI regulation / chips / cloud capex
- healthcare / drug pricing / reimbursement
- antitrust / trade restrictions / industrial policy
- geopolitical conflict affecting supply chains, commodities, shipping, or defense

For each major development, produce a structured record:

- **Catalyst**: one-line summary
- **Why the market cares**
- **Affected sectors / industries**
- **Direction**: bullish / bearish / mixed
- **Mechanism**: revenue, margin, capex, regulation, cost pass-through, rates, risk premium, etc.
- **Time horizon**: immediate / medium-term / long-term
- **Confidence**: high / medium / low

Then aggregate across catalysts and rank the **top favorable sectors**.

### Market regime detector

Before final sector ranking, classify the current market into one primary regime and optional secondary regime:

- `risk_on_growth`
- `risk_off_defensive`
- `inflation_reflation`
- `rate_sensitive`
- `commodity_shock`
- `earnings_revision_recovery`
- `policy_transition_mixed`

For the chosen regime, provide:
- the key evidence
- which sectors should receive a weight boost
- which sectors should be discounted
- whether valuation discipline should be stricter or looser

The regime detector should influence later ranking rules. Examples:
- In `risk_off_defensive`, reduce tolerance for speculative sentiment names.
- In `inflation_reflation` or `commodity_shock`, boost Energy / Materials / selected Industrials where supported by evidence.
- In `risk_on_growth`, allow stronger weight to AI / semis / software if revisions and charts confirm.

### Stage 1 scoring rubric

For each sector, compute an informal score from 0 to 100 using:

- policy/news tailwind strength: 0–30
- breadth of supporting catalysts: 0–20
- immediacy of earnings impact: 0–20
- durability of theme: 0–15
- clarity / confidence of mechanism: 0–15

Call this **Macro Tailwind Score**.

---

## Stage 2 — Social / retail sentiment analysis

Focus on market sentiment sources such as:

- Reddit WSB
- Reddit investing/stocks-style communities
- other accessible public sentiment sources if available

### Preferred architecture

If the runtime has a dedicated social-data tool or plugin, prefer it over free-form browsing.
Treat Stage 2 as a **data module** when possible, not only a narrative step.

### Recommended Stage 2 module contract

Inputs:
- `subreddits`: default `["wallstreetbets", "stocks", "investing"]`
- `lookback_window`: default `24h`, optional `72h` and `7d`
- `min_mentions_threshold`: default `5` when enough data exists
- `deduplicate_spam`: `true` by default
- `exclude_etfs_or_indexes`: optional, `false` by default

Preferred outputs per ticker:
- `mentions`
- `mentions_acceleration`
- `bullish_count`
- `bearish_count`
- `bullish_ratio`
- `sentiment_heat_score`
- `support_type`
- `speculation_risk`

If only partial data exists, degrade gracefully:
- If only mention frequency exists, treat the result as **attention analysis**, not full polarity sentiment.
- If bullish/bearish counts are missing, reduce confidence and do not overstate directional conviction.
- If spam or duplicate-post filtering is unavailable, raise the speculation-risk estimate.

### What to measure

For the b