Sovarion research is built around one uncomfortable rule: a report is not published just because the system can generate text. It must first produce a coherent investment case, a usable target price, and enough evidence quality to survive deterministic checks.
The current long-horizon pipeline covers 7d, 30d, 60d, and 90d research. The 30d/60d/90d engine is intentionally separate from the day-trading scanner. Short-term opportunities care about active movement from a start price; long-term research cares about whether a broader evidence stack supports one clear Bull or Bear case.
Stage 1: Candidate selection. We do not automatically publish a report for every asset. The selector looks at market activity, watchlist demand, recent forecasts, and horizon-specific quotas. This keeps research focused and scalable as the asset universe grows.
Stage 2: Evidence stack. The engine reads fundamentals, technical context, analyst targets where available, recent news, asset profiles, sector relationships, macro regime, and market-understanding impacts. The market-understanding layer is internal: it maps events through topics such as policy, commodities, energy infrastructure, rates, currencies, supply chains, and sectors before the impact reaches an asset.
Stage 3: Target engine. The target engine builds an internal envelope from valuation, analyst context, technical anchors, volatility, mean-reversion signals, and winning thesis evidence. This envelope is not shown as bull/base/bear scenarios to customers. It is used to choose one final case.
Stage 4: Single final case. Sovarion publishes one case per report: Bull or Bear. The report must have one direction, one rating, one target price, an expected return, risks, invalidation, and a reason why the opposite case was rejected.
Stage 5: Pre-gate. Before any expensive report writing, a deterministic gate checks relevance. It blocks weak reports: tiny expected moves, thin target evidence, low confidence, low signal quality, target-direction contradictions, macro conflict, or market-understanding conflict.
Stage 6: Quality gate. After generation, a second deterministic gate checks direction consistency, target consistency, narrative completeness, target-data quality, macro alignment, and rating alignment. If the report does not pass, it is held back from customers.
This matters because research software earns trust by saying no. A blank research list can look frustrating, but publishing low-quality analysis would be worse. Sovarion now tracks both sides: reports that became visible and reports held back by the quality gate.
The goal is not to claim certainty. The goal is to make structured research cheaper, broader, more transparent, and measurable.