Ranker

# Ranker prompt — POSITION readability tiebreak (SECONDARY signal only)

> The epistemic contract is prepended to this prompt at dispatch. Do not restate it.

## What this is — and what it is NOT

You are ranking the anonymized POSITION statements below by how well each one
reads as a stance on the task: clarity, internal coherence, and whether it is
grounded in evidence rather than assertion.

This ranking is a **SECONDARY tiebreak signal only.** It does NOT decide the
plan. The plan is assembled from the verified CLAIM_LEDGER (the claim-by-claim
spine); peer-ranking of whole POSITIONs is noisy and gameable, so it is used at
most to break a tie between otherwise-equal options. Rank honestly; do not
treat this as a contest you must win.

## Hard rules

- **Do NOT rank your own POSITION.** If one of the labels below is yours, omit
  it entirely from the ranking (self-vote is excluded). Rank only the others.
- Rank on the POSITION text as written — readability, coherence, evidence-
  grounding. A confident-but-unsupported POSITION ranks BELOW a hedged-but-
  sourced one.
- If the POSITIONs are too similar, or you lack the evidence to separate them,
  do not invent a difference. Output `INSUFFICIENT_EVIDENCE` instead of a forced
  order.
- No commentary, no scores, no justification paragraphs. Output the block below
  and nothing else.

## TASK (for context only — you are ranking POSITIONs, not re-solving it)

{{TASK}}

## POSITIONs to rank (anonymized; one per label)

{{POSITIONS}}

## Output — EXACTLY this shape, nothing before or after