cross-eval

---
name: cross-eval
description: Use when a high-stakes memo needs independent sanity check before boardroom approval—e.g., major pivots, fundraising terms, or decisions that cost >6 months to reverse.
category: C-Level Advisory
version: 1.0.0
tools: []
---

# Multi-Model Consensus Review

Runs a memo through multiple independent evaluators and reconciles divergences. For high-stakes, irreversible decisions where single-reviewer bias is costly: M&A, major fundraises, layoffs, pivots, regulatory commitments.

## When to Run

- Before signing a term sheet
- Before announcing a layoff
- Before committing to regulated markets
- Before any decision where reversal costs > 6 months of company time
- When boardroom vote was split or had critical dissent

## Models Invoked (graceful degradation)

The review tries each available evaluator in order:

1. **Amara** (primary, always available) — the boardroom's native voice
2. **Codex / OpenAI** (if available) — historically more cautious on emerging markets
3. **Gemini** (if available) — historically more cautious on technical scale claims

If only Claude is available, run Claude with adversarial mode: multiple passes with different system prompts (standard reviewer, devil's advocate, steelman) labeled as single-model.

## Workflow

1. Read the memo
2. Send to each available evaluator with this prompt: "You are an independent C-suite reviewer. Identify top 3 concerns, top 3 supports, and vote (APPROVE / REJECT / DEFER). Assume the memo's reasoning is flawed until proven."
3. Collect independent reviews
4. Reconcile: where do they agree? Where do they diverge?
5. Surface divergences as founder questions

## Output Format

| Model | Vote | Confidence |
|---|---|---|
| Claude | APPROVE | High |
| Codex | DEFER | Med |
| Gemini | APPROVE | Low |

Consensus Concerns (≥2 models flagged):
1. <concern> — flagged by Claude + Codex

Divergent Concerns (1 model flagged):
- <Codex only:> <concern> — worth a second look

Recommendation:
- GO if 2+ models APPROVE and no CRITICAL concerns
- PAUSE if any model DEFERs or any concern is CRITICAL
- STOP if 2+ models REJECT

## Why This Matters

Single-evaluator recommendations have systematic biases. Disagreement is signal, not noise. This is the safety net before irreversibility — not a replacement for outside counsel.