collect-evidence

name	collect-evidence
description	Collect source-backed investment evidence at the start of analyst workflows. Use for research intake, source lists, fact versus assumption separation, and missing-evidence tracking before analysis.

Collect Evidence

Use this skill at the start of an investment workflow.

Before using external data, apply read-only source, provider, as-of, and coverage checks.

Universe and source posture:

• Identify the investment universe before collecting evidence: public equity, ETF/index, crypto public market, macro/rates/FX/commodity, cross-asset overlay, credit signal, or unsupported/unclear.
• For public equity, the detailed evidence shape usually includes company filings/IR, transcripts/presentations, market data/estimates, internal/user notes, portfolio/model/tracker context, and news.
• For other universes, collect only source categories that are actually available and relevant; label missing installed workflows or unavailable source routes as support gaps.
• Record source/as-of or retrieved-at timestamps for market-sensitive data.
• Keep support files such as source indexes, raw exports, normalized CSVs, and logs secondary to the evidence pack unless explicitly requested.

Expected output:

• Universe and workflow type
• Company or asset identifier
• Source list
• Source trust notes: official primary source, management claim, market-derived evidence, secondary news, stale evidence, or unsupported assumption
• Filing, news, price, and market context references
• Facts versus assumptions
• Missing evidence
• Source/as-of posture and support gaps

Decision quality fields when applicable:

• evidence_grade, source_freshness, source_quality
• conflict_status, decision_readiness, confidence
• source_trust_notes
• forecast_required, forecast_allowed, forecast_block_reason
• contrary_evidence, update_triggers, invalidation_conditions

Quality floor:

• Apply the shared artifact quality floor.
• Tag material narrative claims as [factual], [inference], or [assumption].
• Include source dates or retrieval dates when available.
• Include provider/tool names, query parameters, warnings, and credential or coverage failures for external sources.
• Separate verified facts, source claims, assumptions, and analyst inference.
• Weight official primary sources above management claims, secondary news, stale sources, and unsupported assumptions.
• Flag stale, missing, or conflicting evidence.
• Label the evidence pack factual-baseline, screen-grade, or not-decision-ready when source gaps limit downstream use.
• Do not fabricate source dates, prices, filings, metrics, or tool output.
• Include confidence: high, medium, or low, with one reason.
• When writing markdown, include context summary, handoff state, confidence, missing-evidence, next-recipient, blocked-action, and source-snapshot metadata in frontmatter.

Write evidence packs under trading/research/.