evidence-board

name

evidence-board

description

Answer a question or frame a decision with MULTIPLE evidence-backed viewpoints instead of a single verdict. Pulls multiple credible documentations via source-ingest (deliberately including disconfirming sources), then has Board persona lenses + a Senior Project Manager reason over the SAME cited corpus to produce a multi-vision brief — 2–4 distinct options, each with rationale, supporting AND opposing citations, confidence, and the lens that holds it — plus a Senior-PM synthesis (strongest-supported option, live tensions, what evidence would change the answer). Use when the user says "give me options / multiple viewpoints", "don't give a single answer", "what are the perspectives on", "multi-vision", "evidence-based options", "pressure-test with evidence", or any question where a one-sided answer would be misleading. Options-out sibling of ceo-board (which is decision-out). NOT for fiction.

Evidence Board — multi-vision, evidence-grounded

The job is NOT to give the answer. It is to surface the real spectrum of defensible positions, each tied to evidence, so the decision is informed. Single-viewpoint output is the failure mode this skill exists to prevent.

Hard rules

1. Multi-source, multi-viewpoint — dissent fetched INDEPENDENTLY. Never reason from one document. Gather ≥3 credible sources via source-ingest, and DELIBERATELY seek disconfirming / opposing evidence. Pull each dissent as its OWN separately-fetched source — never trust a confirming source's summary of the view that rebuts it (a publisher framing its own critics softens the conflict). Verified in practice: a meta-analysis claimed "fear never backfires," but the independently-fetched rebuttal found a net-negative effect — the conflict only surfaced because the dissent was pulled on its own. If every source agrees, you haven't searched hard enough.
2. Evidential intelligence. Every claim a lens makes is tagged: [evidence: <source>] (cites a stored Tier-1/2 file), [prior] (reasoning, no source), or [gap] (no evidence found — flag it). No unmarked assertions.
3. Options, not a verdict. Output preserves 2–4 distinct positions. A recommendation is allowed, but framed as "strongest-supported given current evidence" alongside the live tension and the evidence that would flip it.
4. Tier discipline (from source-ingest): only Tier-1/2 anchors a fact; Tier-3 is context, traced back before use.

Stage 1 — Senior PM frames (orchestrator)

The Senior Project Manager reads the question and emits:

• The real decision/question in one sentence.
• The fault lines — 2–4 dimensions where reasonable parties would DISAGREE (these become the viewpoints).
• 3–6 sharp empirical questions the evidence must answer (each ends in ?). Save as brief.md.

Stage 2 — Evidence gathering (multi-source, via source-ingest)

Run source-ingest for each empirical question. Pull a CORPUS, not one hit: aim ≥3 credible sources spanning the fault lines, including the strongest source AGAINST the obvious answer. Each lands in the vault Sources/ with credibility tier + citation. Compile a evidence.md index listing every source by ID, tier, and which fault line(s) it speaks to. If a fault line has no evidence, mark it [gap] — do not paper over it.

Stage 3 — Parallel lenses over the SAME corpus

Dispatch parallel sub-agents (see parallel-delegate), one per Board lens relevant to the question — draw from ceo-board personas: Contrarian (mandatory — steelmans the opposing view), Technical Architect, Market Strategist, Revenue/Compounder, Custom Oracle (domain expert), and any fault-line-specific lens. Each lens:

• reads the shared evidence.md corpus (not its priors first),
• argues its position, citing specific sources by ID with the [evidence:]/[prior]/[gap] tags,
• names what evidence would change its mind. Each returns a structured position (stance, key citations for, key citations against, confidence).

Stage 4 — Senior PM synthesis → multi-vision brief

The Senior PM does NOT pick a winner and bury the rest. It produces multi-vision.md:

1. Question + fault lines.
2. Evidence base — the cited corpus (source IDs, tiers, links).
3. The options (2–4) — for each: the position; who holds it (which lens); the rationale; citations FOR; citations AGAINST; confidence (high/med/low) and why.
4. Tensions — where the evidence genuinely conflicts or is thin.
5. Senior-PM recommendation — the strongest-supported option as of this evidence, explicitly NOT the only one, with the tradeoff accepted and "what would change this" (the finding that would flip the recommendation).
6. Confidence + gaps — overall confidence and the [gap] items worth resolving next.

Output

multi-vision.md (+ brief.md, evidence.md) in the working folder; store a copy in the vault Decisions/ if it informs a ratified call. Every viewpoint must trace to the evidence base. Hand to ceo-board only if the user then wants the options collapsed into ONE ratified decision.

Composition

source-ingest (evidence) → this skill (multi-vision options) → optionally ceo-board (decision-out). Reuses ceo-board persona definitions (ceo-board/references/board-members.md).