How the Soar engine works

Soar is a research desk. The engine produces analysis — specific, actionable trade ideas with entries, profit targets, stops, and reasoning. It publishes that analysis up to nine times a day on a fixed schedule, alongside a live track record.

The engine does not manage your money. It does not move funds, decide what you should own, or execute trades — unless you've explicitly connected a Soar bot to a brokerage account you control. Even then, the bot acts on Soar's published signals under the rules you configured. You can pause or kill it at any time.

Think of the engine as a friend who does the research, hands you a written setup, and lets you decide. You can read it, act on some of it, ignore most of it, or just watch the track record.

The engine fires up to nine times every weekday on a fixed US Eastern schedule (eight on weekends). Each session scans the universe from scratch, so a setup that didn't qualify in the morning might qualify in the afternoon once the chart has moved. The windows below are representative of how those sessions span the trading day.

• Early morning
  Pre-open scan

  Catches overnight moves in crypto and Asia/Europe price action. Earliest read on the day.
• Late morning
  US open follow-through

  Lets the first hour of US trading settle, then looks for setups that emerged after the bell.
• Early afternoon
  Midday read

  Captures intraday breakouts, breakdowns, and reactions to mid-session catalysts (earnings, macro releases).
• Late afternoon
  Closing setup

  Looks for swing setups going into the close. Useful for week and month timeframe cards.

This cadence is frequent enough that a meaningful move during the day will surface in the next run, but not so frequent that the engine churns out noise. Crypto signals can fire in any session (24/7 markets); stock and ETF signals only fire during US market hours.

Every Soar trade card carries the same set of fields. They're the difference between "something looks interesting on NVDA" and a setup you could actually act on.

• AssetThe ticker — NVDA, BTC, SPY, GOLD, etc. Drawn from the ~14,000-asset universe Soar tracks.
• DirectionLong (expecting price to rise) or short (expecting price to fall). The engine emits both.
• Entry zoneA price range, not a single number — e.g. $142.50–$144.00. Markets aren't precise enough for single-point entries.
• Target 1 (T1)The first profit objective. When price hits T1, the card is marked T1-hit but stays live, tracking toward T2.
• Target 2 (T2)The second profit objective. When price hits T2, a trailing-stop logic engages to capture additional upside if the move continues.
• Stop lossThe price where the thesis is wrong. Tighter than most traders run, deliberately — small losses, larger wins.
• Conviction scoreA 0–95 indicator of how strongly the inputs agree. Higher means more signals line up. See the next section for what it actually measures.
• TimeframeToday, this week, this month, this quarter, this year, or multi-year. Tells you how long the engine expects this setup to play out.
• ThesisA short paragraph explaining why this setup, in plain English. The engine's actual reasoning, not boilerplate.
• Risk / rewardThe math: distance to T1 divided by distance to stop. Soar's gate requires at least 1.5:1, and most published cards run higher.

Conviction measures how strongly the engine's inputs agree on a setup. It's a relative score — "how confident is the engine in this idea compared to other setups it's looking at right now" — not a probability of profit.

The scale runs roughly 0 to 95. The cap is there because real-world setups never have perfect alignment: the macro might fight the chart, momentum might lead price, news might cut both ways. A score in the high 80s means everything the engine looks at is pointing the same direction. A score in the 60s means the case is decent but mixed.

Conviction is one input among many. A higher score does not guarantee a better outcome — a 75 can win while an 85 stops out. Use it the way you'd use any signal: weight it against your own context and the rest of the card.

The engine reads a wide cross-section of inputs on every asset, every session. No single signal is decisive. The conviction score is the result of how many of them agree.

• Trend. Where is price relative to its 50-day and 200-day moving averages? Is the trend rising, falling, or rolling over? This is the single biggest input for swing-timeframe setups.
• Momentum. RSI (overbought/oversold reading), MACD (trend-following momentum indicator), recent rate of change. Tells the engine whether the move has fuel left or is running on fumes.
• Volatility regime. Are markets calm or stressed? VIX-style measures for stocks; realized volatility for crypto. The engine sizes stops and targets differently in a 12 VIX vs. a 30 VIX environment.
• News + sentiment. Has there been material news on this asset in the last 48 hours? An earnings beat, a downgrade, a regulatory headline? The engine factors this in — sometimes flipping a setup's direction outright if news cuts the wrong way.
• Macro fit. Does the broader regime support this direction? Long oil into a recession scare is a tougher case than long oil into geopolitical stress. The macro layer either reinforces or fights the technical read.
• Multi-timeframe alignment. Does the daily chart agree with the weekly? When timeframes disagree, conviction drops. When they line up, it climbs.
• Historical edge on the asset. Soar tracks how similar setups have resolved on this specific asset over time. Some tickers have stronger edges than others — the engine remembers.
• Risk / reward geometry. Does the setup actually pay? A great-looking chart with the nearest target 0.5% away and the stop 5% away is a non-starter. The engine refuses to publish low-R:R setups even if everything else looks clean.

Gold Picks are the engine's highest-conviction setups. A card qualifies as gold only when three conditions are met at once:

• Conviction score in the upper band (roughly 80+).
• Multi-timeframe alignment — daily and weekly charts agree on direction.
• Risk/reward of at least 2.5:1 (the potential gain is 2.5× the size of the risk).

The engine caps Gold Picks at a maximum of two per session. The cap is intentional. If gold were unbounded, "gold" would lose meaning. Two-per-session forces the engine to pick its best, not just every setup that clears a number.

On most sessions there is one gold or none. Some sessions produce two. The engine doesn't manufacture gold to fill the slot.

Some setups don't clear the trade-card bar but are close enough to be worth tracking. Soar marks those as "watching."

A watching card is a setup within roughly 20% of the conviction threshold — typically with one specific thing missing (the chart looks right but RSI is mid, or momentum looks right but the macro doesn't fit). The card shows what the engine is watching for: a tighter stop, a cleaner pullback, a momentum confirmation.

When the missing piece arrives in a later session, the watching card can get promoted to a full trade card. When the setup invalidates — price moves the wrong way, the catalyst window closes — the watching card simply ages out. They're a window into the engine's near-term attention.

A trade card has four possible end states. The engine tracks each card live until it lands in one of them.

• T1 hit (card stays live)When price reaches Target 1, the card is marked T1-hit but does NOT retire. The thesis is partially confirmed — the engine keeps it on the dashboard with a 'PAST T1' tag, tracking the move toward T2.
• T2 hit (trailing exit armed)When price reaches Target 2, the engine arms a trailing-stop. If the move keeps running, the card captures more. If price reverses past a defined threshold, the card retires as a T2 winner.
• Stopped outIf price hits the stop loss before reaching T1, the card retires as a loss. The thesis was wrong. The stop is published exactly to make this verifiable.
• Aged out (timeframe ceiling)Every card has a maximum age tied to its timeframe — a 'today' card that hasn't resolved in a few days expires; a 'week' card has more rope; a 'multi-year' card has the most. Aging out is neither win nor loss — the setup ran out of time.

Carry-forward and the Day-N counter

If a setup is still valid across multiple sessions, the engine carries the card forward rather than re-publishing it. You'll see a "Day N" counter — Day 1, Day 2, Day 3 — telling you how long this exact setup has been on the board.

That carry-forward logic means a high-conviction setup doesn't disappear just because a new session ran. It stays live with the original levels until it resolves.

Honesty about limits is more useful than marketing. Here's what the engine explicitly does not do:

• Predict the future. No model can. The engine reads the present better than most, but every setup can fail.
• Guarantee an outcome. Conviction is a relative score, not a probability. Stops exist because losses happen.
• Know your portfolio. The engine has no view into what you already own. A setup it publishes might be a great add for one user and a duplicate for another. That's your call.
• Account for your taxes. Short-term capital gains, wash-sale rules, account types — none of that enters the analysis. Talk to a tax professional.
• Replace your judgment. The engine is a research tool. The decision to act, how much to size, and when to exit on your own terms is always yours.
• Trade illiquid microcaps. The universe is US-tradeable names and major crypto. Penny stocks, OTC, and thinly-traded foreign names aren't covered by design — published setups need enough liquidity to actually fill.
• See private data. No insider tips, no leaked filings, no private order flow. Everything the engine sees is public.

Most brokerage research desks publish human-written notes on an ad-hoc cadence — when an analyst has something to say, or after a notable event. The output is qualitative, often without specific entries or stops, and rarely tracked publicly for accuracy.

Soar runs the opposite playbook. The engine fires up to nine times a day on a fixed schedule. Every published idea carries an exact entry, exact target, exact stop, and a thesis you can audit. Every outcome is recorded against the published levels. The track record is on the platform, not buried in a quarterly performance memo.

Both approaches have value. Human analysts catch nuance an algorithm can miss; a consistent engine catches setups a stretched-thin analyst skips. Soar isn't trying to replace judgment — it's trying to be the most consistent voice in the room.

Every published card is tracked against live market prices on a regular schedule. When price touches the stop, the card is marked stopped. When price touches T1 or T2, the card is marked hit. When the timeframe ceiling is reached, the card ages out.

The full record lives at /dashboard/recent-trades— closed trades, win rate, average gain, average loss, profit factor, the best winner, the worst stop-out. Soar publishes the whole picture, including losers. There's no cherry-picking, no "featured wins", no hidden cohorts.

Outcomes are computed by comparing the published levels to observed market prices on schedule. They are simulated — they don't include the slippage, fees, or execution latency of a real account. Real-trading results will differ. Use the track record to judge the engine's historical edge, not as a predictor of your own returns.

On US market holidays — Memorial Day, Independence Day, Thanksgiving, Christmas, and other full-day closures — the engine adapts. Stocks, ETFs, and US-traded commodities are skipped entirely (nothing to trade if the market is closed).

Crypto runs 24/7, so on holiday sessions the engine still emits crypto cards. The session log will note "US market holiday — crypto only," so you can see why the board looks different from a normal Monday.

On weekends, the same rule applies — the schedule shifts toward crypto and major macro context, and stock-side cards stay quiet until Monday's open.