skills/screen-stocks-etfs/references/modes.md

# Panel Modes and Evidence Budget

Choose the smallest mode that can answer the decision reliably. Mode controls research depth; presentation remains Decision View unless the user explicitly requests scores, calculations, methodology, or an audit.

## Quick Scan

Use for one named security, a narrow follow-up, or an explicit quick scan.

- Select the three or four voting roles most capable of changing the conclusion.
- Reuse a compact evidence packet; do not run broad scouting or collect unrelated macro data.
- Trigger the compact market-history packet only under `market-history.md` rules.
- If a fast-moving geopolitical or macro event is central, research the direct transmission channel but skip historical event studies unless they could change the conclusion.
- Show every selected voting member's decision plus a separate Chief Risk Officer line.
- Skip formal debate unless a hard risk or close split emerges.

## Full Panel

Use for rankings, sector selection, standardized model comparisons, or comparisons of two or more meaningful candidates unless the user explicitly requests Quick Scan.

- Run all six voting roles plus the non-voting Chief Risk Officer.
- For multi-candidate ETF/sector comparisons, run the compact 1M/3M/6M/YTD/1Y market-history packet once for all candidates.
- Scout outside alternatives only for expanded/open universes.
- Use only macro/geopolitical modules with plausible causal impact.
- Score combinations separately from components.
- Keep raw scores and arithmetic in Audit View.

## Panel Debate

Use after a Full Panel when the user requests debate, finalists are close, disagreement is large, or a contested assumption drives the ranking.

- Freeze blind baseline votes.
- Let the strongest evidence-backed defender answer the Chief Risk Officer's strongest challenge.
- Permit one evidence-based vote revision per affected member.
- If the contested assumption is historical/regime-based, escalate the relevant history module rather than merely restating opinions.
- Reaggregate and show changed preferences; detailed score deltas belong in Audit View.

## Natural-language controls

- `quick scan` → minimize depth and latency
- `full panel` → all six voters + Chief Risk Officer
- `panel debate` → rebuttal and one revision round
- `show dissent` → expand strongest minority case
- `show scores` / `audit this` → reveal labeled score matrix and arithmetic without rerunning current research
- `value-first` → reweight before rescoring
- `stress <scenario>` → rerun scenario-sensitive roles
- `swap A for B` → update candidate-dependent evidence and votes
- `change horizon` → rerun fit, weights, market-history depth, catalysts, and implementation
- `rerun <role>` → refresh one role with current evidence

Do not ask the user to choose a mode unless the tradeoff materially matters.