DeepResearchForecast

Use this skill during research and forecast runs when sourced structured artifacts should become decision-relevant figures. It renders local deterministic charts from quantitative.json, timeline.json, prediction_markets.json, actors.json, and sources.json; use it for comparable market/technology/policy benchmarks, published forecast revisions, dated inflection points, market-implied probabilities, industrial-chain relationship maps, and evidence diagnostics. It rejects fabricated data, incompatible denominators, proxy-score charts, and decorative visuals.

Use this skill in the DeepResearchForecast/DeerFlow forecasting pipeline when research must seed an ontology, knowledge graph, and actor simulation. It produces an actor-centric, ontology-ready dossier with deeply profiled key actors, directed typed and valenced relationships, historical evolution, and the behavioral fields needed for tailored runtime roles. It builds on the deep-research skill's source tiering, evidence grading, and verification discipline and adds a multipass actor/relationship workflow with an AI-judge quality gate.

Use for web research, comparisons, explanations that depend on external evidence, and every forecast investigation. Provides the always-on core for KIQ decomposition, source grading, evidence-yield stopping, disconfirmation, forecast inputs, provenance, and phase-aware handoffs. Detailed tradecraft and the final-dossier contract live in lazy references so repeated slash activation stays token-efficient.

Use this skill whenever a research question is a forecast — anything asking about probability, outcomes, elections, policy decisions, macro events, or "will X happen". It teaches the prediction_market_search tool (Polymarket's public Gamma API — no API key, no wallet) — how to derive short high-recall queries that match how markets are actually titled, which markets to trust (open, priced, liquid), how to self-filter for relevance, and how to write the surviving anchors into a "Prediction Market Signals" table. Market-implied probabilities are calibration anchors, never ground truth.

Use this skill instead of a bare web search for ANY question requiring web research. Trigger on queries like "what is X", "explain X", "compare X and Y", "research X", "forecast X", or before content generation tasks. Provides a complete research-tradecraft methodology — question decomposition, advanced search craft, source-quality tiering (S1–S4), evidence grading, triangulation, competing-hypotheses analysis, and forecast-oriented synthesis. Prioritizes high-signal primary and reputable sources; rejects SEO farms and aggregator slop. Use proactively whenever the answer depends on online information.

Use this skill whenever a DRF-2 forecast could be calibrated against real prediction markets. It teaches the prediction_market_search tool (Polymarket's public Gamma API — no API key required) — how to derive short high-recall queries, which markets to trust (open, priced, liquid), how to keep the snapshot diverse across events, and how to use market-implied probabilities as calibration anchors rather than ground truth, including the divergence-explanation rule.

Use this skill for the DRF-2 forecasting pipeline whenever the research output must seed an ontology, a knowledge graph, and an actor-based simulation — i.e. any "forecast X" / "who wins / what happens to X" prediction run. It specializes the deep-research tradecraft toward an ACTOR-CENTRIC, ONTOLOGY-READY dossier — identify the real key actors (demoting mere reporters/outlets/sources), profile each in depth (role, values, beliefs, incentives, goals, constraints, resources, vulnerabilities, relational roster), map their directed, typed, valenced relationships, and trace how the cast evolved over time. Runs a multipass workflow with an AI-judge quality gate that loops until the dossier is excellent. Builds on (does not replace) the deep-research skill.

Use this skill in the DRF-2 pipeline whenever you build or query the temporal knowledge graph through the kg_* MCP tools (Graphiti + FalkorDB engine). It teaches WHEN and HOW to call each tool — episode chunking and batching discipline, bitemporal reference-time anchoring, entity-resolution hygiene against the canonical cast, causal-edge admission criteria, and the retrieval playbook (typed search, multi-hop causal traversal, cascade tracing, centrality).

Use this skill in the DRF-2 pipeline to assemble the final deliverable — a three-part Bridgewater-style forecast brief. It encodes the calibration rubric (at least 10 independent binary forecasts with objective metric+threshold+date resolution criteria, contrarian framing mix, conviction spread with stdev above 0.12), the scenario spine, prediction-market anchoring with explained divergences, and the provenance discipline that separates research evidence from simulation roleplay.

Use this skill in the DRF-2 pipeline to design the knowledge-graph ontology (entity types + edge types) from the research report and the actor dossier, before any graph construction. It encodes the proven methodology — domain-adaptive templates (social-opinion vs general-forecast), dossier-seeded type budgets, archetype and simulation-tier tagging, edge family/valence/direction semantics, causal edge design, and reserved-attribute rules — and produces the exact ontology.json contract the graph stage applies to the KG engine.

Use this skill in the DRF-2 pipeline to design the multi-agent OASIS simulation from the actor dossier and knowledge graph — dossier-grounded persona design (identity/views/incentives/relations/red-lines), the shared world brief, seed posts from researched stances, activity profiles and round mapping, scheduled events from the dated timeline, and the initial follow graph with valence-aware clustering. The output is the simulation_config.json contract consumed by the simulation engine's sim_start tool.