skills/stock-sentiment-analysis/references/experience.md

# Stock Sentiment Analysis Experience

## Active Playbook

- Use `confirmed catalyst x expectation gap x sentiment cycle x price confirmation`; do not rely on a single headline or a single hot forum post.
- Treat forums/掲示板/股吧 as expectation and crowding data, not as facts. They are most useful when they show what the market feared or expected before the catalyst.
- A durable theme narrative must be accepted by price. If a stock makes lower highs, loses the catalyst-day support, or underperforms a stable sector after good news, treat forum confidence and long-term story strength as crowding data, not as proof the pullback is healthy.
- For cross-market sentiment, separate `mainline`, `funds`, `game`, and `cycle`: a real story still needs liquidity support and market-structure confirmation; a crowded late-cycle story can reverse even while the long-term narrative remains valid.
- Separate `情绪票`, `趋势票`, and `hybrid`. Emotion stocks top fast through failed boards, long upper shadows, and crowding; trend stocks usually deteriorate through guidance quality, sector breadth, and failed reclaim of higher-timeframe support.
- A healthy early-cycle disagreement can be a turnover test; a late-cycle disagreement after broad euphoria is usually a warning.
- For collective sector surges and stock-sector-market resonance, use the six-factor mnemonic `强大盘 + 高成交 + 上涨家数多 + 完整产业链扩散 + 基本面持续验证 + 非末端加速`. It is an interacting evidence matrix, not an equal-weight score: terminal acceleration can override otherwise strong tape. Always separate logic durability, tape continuity, and new-entry quality, then finish with `market -> sector/theme -> stock` relative-strength confirmation.
- For 1h+ or swing analysis, sentiment must be checked against news and chart structure. A narrative is not accepted until price holds/reclaims the relevant level with volume.
- In global risk-off moves, first check rates, FX, volatility, and crowded theme positioning before attributing everything to one local stock headline.
- For sector/theme tape, classify the market regime before picking stocks: `attack`, `defense`, `rotation`, or `de-risking`. In attack regimes, prioritize the main offensive line; in defense regimes, prioritize cash-flow, resource, healthcare, policy-hedge, or other defensive alternatives. Do not treat a defensive bounce during an offensive tape as a confirmed new main line.
- Identify the main line by repeated fund return, breadth, and sector-level money-making effect, not by one-day gains or the highest short-term runner. A few hot high-beta leaders without broad sector follow-through are usually emotion markers; a quieter sector with frequent institutional return can be the real profit center.
- Separate holder logic from new-entry logic. A main line in acceleration can be a `hold/enjoy profits` setup for early holders while being a poor chase for new money. Missed pullbacks are not automatically chase signals; wait for the next healthy divergence/pullback or accept that the trade becomes a fast in/out momentum trade.
- When a leading theme pauses, classify substitute strength carefully: it may be an assist, a defensive hedge, an old-leader rebound, or only a liquidity placeholder. Do not promote it to new leadership until it shows repeated return, breadth, and acceptance after the original leader weakens.
- Do not pre-commit to the next imagined theme. Candidate themes are equal before the tape selects one; follow the one that first shows continuity and expansion, otherwise prior bias can cause missed leadership and poor holding decisions.
- For A-share broad-market tape questions, use live headlines/快讯 first to identify policy, macro, liquidity, overseas, or sudden risk events; then use Sohu index and industry/concept board涨跌幅 as auxiliary confirmation of fund direction, breadth, and theme concentration.
- For Japan broad-market tape questions, use live headlines/快讯 first to identify JGB-yield, USD/JPY, BOJ/MOF, overseas tech, China/Korea/Taiwan spillover, commodity, or geopolitical drivers; then use JPX real-time indexes, TOPIX 33 sectors, TOPIX-17, size, style, and market-type indexes as auxiliary confirmation of which sectors or styles are dragging or supporting the tape.
- Use mature answer structures for recurring problem types because they make analysis easier to accept and compare, but choose the structure deliberately. Keep required single-stock frameworks when they fit; for genuinely different or new questions, think through the best structure instead of forcing every answer into the same template.
- For macro or broad-market tape, do not sound like a source log: avoid repetitive "`快讯显示` / `搜狐确认` / `JPX数据显示`" phrasing and answer in terms of drivers, drag/support, fund direction, breadth, expectation gap, and trading implication. Do not apply this too broadly to single-stock workflows; stock-specific news, quote data, and forum evidence can still be cited or presented in the skill's required structure when useful.