How Kasiel's AI stock analyzer actually works

Kasiel ships three analysis depths because different decisions need different rigor. A quick read on whether to glance at a ticker is not the same as a full position-sizing decision.

Every analysis starts with the same data pull. The depth tier just determines how many sources we go to and how deep we go in each.

Market data

OHLCV (open, high, low, close, volume) at multiple timeframes — 1h, 4h, daily, weekly, monthly. We use Yahoo Finance for stocks and ETFs, Binance for crypto. The 6-month daily window is the default for indicator computation; shorter windows for intraday context, longer for secular-trend reads.

Fundamentals

Income statement, balance sheet, cash flow statement. Revenue, earnings, gross/operating/net margins, free cash flow, debt levels, share count. We pull the latest reported quarter plus the trailing twelve months and compare against sector medians.

Filings and transcripts

Recent 10-K (annual), 10-Q (quarterly), 8-K (material events). Earnings call transcripts when available. Insider Form 4 filings for the trailing 90 days. Institutional 13F filings for the most recent reporting period.

News and sentiment

News flow from financial wire services, industry trades, and major financial press for the trailing 30 days. Social sentiment indicators where they meaningfully move price (which is rare — most retail social signal is noise). For Deep Research, we pull options flow data including unusual activity, IV skew, and put/call ratios.

Macro context

Interest rate environment, yield curve shape, dollar strength, sector rotation indicators, breadth measures, VIX regime. The macro layer is what lets us tell you when a fundamentally-fine stock is fighting a tape it can't beat.

We compute a fixed set of indicators on every analysis. The exact formulas are standard — we don't use proprietary indicators because there's no good reason to:

• RSI — 14-period Wilder smoothing
• MACD — 12/26/9 EMA
• Bollinger Bands — 20-period SMA, 2 standard deviations
• ATR — 14-period
• VWAP — session-anchored
• OBV — cumulative
• SMA — 20, 50, 200 period
• EMA — 9, 21 period
• Pattern detection — Doji, Hammer, Engulfing, Morning/Evening Star
• Support/resistance — pivot-point clustering

These get computed across multiple timeframes simultaneously. The reasoning step then evaluates which timeframes are confluent (all pointing the same direction) versus conflicting (1h says one thing, daily says another). Confluence raises confidence; conflict caps it.

The qualitative layer is where AI stock analysis earns its keep. Anyone can compute RSI. The hard part is reading 12 news articles, 3 SEC filings, an earnings transcript, and 2 analyst notes, then producing a coherent thesis.

Our reasoning pipeline does this in a structured way:

1. Extract material events. What happened in the last 30 days? What's coming in the next 90?
2. Map the bull case. The 3–5 reasons this stock could outperform from here.
3. Map the bear case. The 3–5 reasons it could underperform — including the ones the bull case ignores.
4. Identify catalysts with dates. Earnings, FDA decisions, FOMC meetings, contract awards, IPO lockup expirations — anything dated that could move the stock.
5. Stress-test confluence. Where does the technical setup, fundamental health, and narrative actually agree? Where do they conflict?
6. Produce a verdict and confidence. The verdict is whichever direction has the strongest combined signal. The confidence is calibrated to how many independent indicators agree.

The output is structured. Every analysis returns the same schema: verdict, confidence, summary, key levels (buy zone, danger, 6-month target, 12-month target), bull case, bear case, technicals breakdown, fundamentals breakdown, smart money read, macro context, risk level and explanation, exit plan, and (for Deep Research) catalyst calendar with dated events plus leveraged-product menu.

Every Kasiel analysis is a structured object, not a free-form essay. The fields are stable across every ticker, every depth tier, and every time. This matters for two reasons: it makes the analyses directly comparable to each other, and it makes outcome scoring deterministic.

Key required fields:

• verdict — STRONG BUY / BUY / WAIT / SELL / STRONG SELL
• confidence — 0–100 scalar
• key_levels.buy_zone — entry zone
• key_levels.danger_below — invalidation level (functional stop)
• key_levels.target_6mo — 6-month target
• key_levels.target_12mo — 12-month target
• trading_outlook.setup_rating — STRONG / MODERATE / WEAK / NO SETUP
• risk_level — LOW / MEDIUM / HIGH / EXTREME
• bottom_line — single-sentence TL;DR

When the responsible call is "there's no clean trade right now," the verdict is WAIT and setup_rating is NO SETUP — but the analysis still tells you the specific trigger to watch for. We don't force a directional call when the data doesn't support one.

Every public Kasiel analysis is auto-scored against real market prices at fixed maturity intervals: 7 days, 30 days, 45 days, 60 days, 90 days, 180 days, and 365 days.

The classification is deterministic:

• Right. BUY/STRONG BUY where the stock rose > 5%; SELL/STRONG SELL where it fell > 5%; WAIT where the stock moved < 5% (either direction).
• Close. The direction was right but magnitude was inside the "Right" threshold — or for WAIT, the stock moved 5–15% (outside the WAIT band but close).
• Wrong. Direction was wrong, or magnitude exceeded the WAIT band by a large margin.

Outcomes are computed at maturity and tweeted to @kasielanalysis automatically. No human editing, no curation, no cherry-picking. The full live scoreboard is at /results.

What we explicitly do notdo: backtested performance claims, hypothetical strategy results, "if you had followed our signal" charts. Those are easy to cherry-pick. We only publish numbers from real, timestamped, public verdicts that can't be retroactively edited.

Every claim that makes a numerical or factual statement comes with the source. Earnings figures cite the 10-Q or 8-K. News-driven claims cite the article. Insider buying claims cite the Form 4 filing. The sources block at the end of every Standard and Deep Research analysis is the audit trail.

When sources disagree, the analysis surfaces the disagreement rather than picking one side silently. This is especially important for things like analyst targets (which range widely on the same stock) and earnings estimates (which can be revised mid-cycle).

• Manage money. Kasiel is research output, not a brokerage or RIA. We don't custody assets or execute trades.
• Provide personalized advice. Every analysis is the same regardless of who's reading it. Position sizing, allocation, and risk tolerance are your decisions.
• Time short-term moves. Our Lite tier surfaces fast reads but the strongest claims are multi-week to multi-month directional. We don't do next-day predictions.
• Hide our failures. Wrong verdicts stay public. The track record includes them.
• Use proprietary indicators or hidden methodology. Standard indicators, standard fundamentals, transparent reasoning. The competitive edge is the synthesis quality, not secret math.