How It Works

📡 1. Data sources

Every analysis starts with real, live data — never invented numbers:

• Twelve Data — real-time and historical pricing for US stocks and major cryptocurrencies.
• IndianAPI — real-time pricing for NSE/BSE-listed Indian stocks.
• NewsAPI — recent headlines for the ticker, used for the sentiment signal below. When no news feed is configured, this signal is skipped entirely rather than faked.

📈 2. The three signals

Sentiment is scored with a keyword approach: headlines are scanned for finance-relevant positive words (e.g. "surge," "beat," "upgrade") and negative words (e.g. "downgrade," "lawsuit," "plunge"). The score is the balance between the two. It's intentionally simple — it does not understand context, sarcasm, or nuance the way a human analyst would.

🎯 3. From score to verdict

The three signals are combined into a single composite score between -1 and +1:

• Composite above +0.20 → BUY
• Composite below -0.20 → SELL
• Everything in between → HOLD

Confidence (shown as 45–95%) reflects how strong that composite score is — how far it sits from zero — not the statistical odds that the call turns out correct. A 90% confidence verdict means the signals pointed firmly one direction, not that the outcome is 90% likely.

🎲 4. The "chaos" element

By design, roughly one in three verdicts that would otherwise land as an uncertain HOLD get nudged into a directional BUY or SELL instead. This is deterministic per ticker (the same ticker gets the same treatment until its underlying data changes meaningfully) — it is not random per visit, and it's not hidden from you: it exists because real market conviction routinely outpaces the actual evidence, and that gap is a big part of what this tool is satirizing.

🎭 5. The commentary layer

Once the verdict and confidence are locked in, a written commentary is selected from a large bank of pre-written, pop-culture-flavored templates matched to the verdict, price direction, and news tone. This is chosen for entertainment — it does not influence the score in step 3, and it is regenerated daily per ticker so the same stock doesn't say the exact same thing every day.

⚠️ Limitations, honestly

• Keyword-based sentiment cannot detect sarcasm, nuance, or a headline that means the opposite of what its words suggest.
• The model uses no fundamentals — no earnings, revenue, valuation multiples, or balance sheet data.
• Past trend and momentum are not predictive of future price movement.
• The one-in-three "chaos" nudge described above is a deliberate design choice, not a bug — but it does mean some HOLD-strength situations get reported as directional calls.