15 AI agents.
Every signal you're missing.
One intelligence.

Before a single ticker is evaluated, the engine classifies the prevailing macro regime — risk-on, risk-off, stagflationary, or transitional. This isn't cosmetic. Every downstream score is multiplicatively weighted by the regime classification. A 91-conviction "Strong Buy" in a confirmed risk-off environment carries a materially lower effective weight than the same score in a sustained bull trend. The regime is re-evaluated each session using a composite of VIX term structure, yield curve slope, credit spreads, and momentum breadth — not a manual toggle.

Each ticker is scored across four independent dimensions: momentum (price structure, 20/50-day relative strength, volume confirmation), catalyst (breaking news, SEC filings, congressional transaction clusters, earnings revisions), sector rotation (institutional capital flow trends, sector relative performance), and squeeze probability (short ratio, float utilization, days-to-cover, dark pool activity). The four scores are weighted and collapsed into a single 0–100 conviction number — but every factor is surfaced individually. You can see exactly why a ticker scored 84 versus 61. Nothing is hidden inside a model you can't interrogate.

Stage 1 and 2 outputs are synthesized into a directional call — Long, Short, Accumulate, or Hold — with a precise entry zone, stop level, and a multi-target price ladder (T1, T2, T3). The signal isn't just a direction. It includes the primary catalyst driving the thesis, an explicit bull case, an explicit bear case, and the specific data point that would invalidate the trade. Risk/reward is calculated at each target level. You know the setup before you know the ticker — so confirmation bias doesn't get a seat at the table.

Every recommendation is matched against the historical performance of the same ticker under the same setup conditions — same regime, same signal cluster, same directional bias. You see actual 5-, 10-, 20-, and 60-day win rates and average returns, along with the sample size so you know how much weight to give it. This isn't generic pattern backtesting. It's this ticker, this setup, this market environment. If the historical edge is weak, statistically thin, or contradicts the current thesis, the conviction score adjusts downward automatically — the engine won't let overconfidence compound a bad setup.

Technical setups don't exist in a vacuum. Before a signal is confirmed, it's cross-checked against peer-relative valuations (EV/EBITDA, P/E, P/S versus the sector composite), analyst estimate revision momentum over the trailing 90 days, and a proprietary fundamental grade (A–F) built from earnings quality, balance sheet leverage, and free cash flow trend. A setup that's technically perfect but trading at a 3x sector premium with deteriorating estimate revisions gets flagged — conviction is penalized proportionally. Being right on price action but wrong on valuation is a trade that works until it doesn't, then hurts badly.

Three discrete outcome paths — Bull, Base, Bear — each with an assigned probability weight, a price target, and the specific macro or fundamental condition that unlocks it. Expected value is calculated across all three: (Bull target × Bull probability) + (Base target × Base probability) + (Bear target × Bear probability). If the EV is negative, the signal doesn't surface regardless of conviction score. A DCF sensitivity table is generated for each scenario using live fundamental inputs — revenue growth assumptions, margin trajectory, and discount rate — so the price targets aren't arbitrary. They're derived.

A signal that scores 88 in isolation can still be the wrong trade for your specific book. Stage 7 evaluates each recommendation in the context of your live portfolio — checking concentration (does this push a single position above 20% of portfolio value?), correlation (are three of your top five holdings already moving together?), and tail exposure (what does a 15% drawdown in this ticker do to overall portfolio NAV?). If the signal materially amplifies existing risk, it gets flagged with the specific conflict before it reaches you — not suppressed, but contextualized. You decide with the full picture, not just the ticker view.

All seven stages collapse into a single structured output — ranked by priority score, not alphabetically or by ticker. Each signal surfaces a composite conviction score, the primary catalyst, bull and bear cases, a recommended next step with sizing context, and the exact data point that invalidates the thesis. There's no paragraph of commentary to parse. The output is a decision object: everything you need to act, nothing you need to filter out. Signals are timestamped and fingerprinted — if the underlying data changes and the thesis degrades, the conviction score updates in place. You're always looking at the current read, not a stale one.