pre-write-precision-check

name	pre-write-precision-check
description	Independently fact-check every claim in a paper's bundle against the raw source text (Factored Verification method), flagging any claim the source doesn't actually support at a correctness score below 0.8. Use this before any article drafting begins, immediately after extract-structured-bundle produces the bundle — never skip this even if the bundle looks obviously correct, since research shows even top models hallucinate subtly when summarizing papers.
execution	subagent
prompt	./prompt.md
input	paper_ref (string), bundle (dict)
output	precision_result (dict — failure_type, flagged_claims)
dependencies	{"sops":["spawn-agent","literature-research"]}

Pre-Write Precision Check

Factored Verification (Elicit, arXiv:2310.10627): decompose the bundle into claims, independently re-query the raw paper for each, score correctness. Catches commission errors — claims the source doesn't actually support.

Execution

Subagent — spawned via subagent-spawning/spawn-agent skill.

Why Subagent

Independent fact-checking requires a context that is NOT anchored to the drafting process that produced the bundle — a fresh subagent with no memory of how the bundle was written is what makes "independent" re-querying actually independent (spec §6, §2 point 2).

Critical Constraint (repeated from prompt.md — load-bearing)

This sop MUST query the raw paper fresh for every claim. It must never treat the bundle's own source_anchor field as sufficient evidence — see the newsroom-fact-checker analogy in the design spec (§2 point 2): the checker re-reads the source independently, not the writer's notes about the source.

Available SOPs

Optional, no fixed order; the final leaf is always a sop.

SOP	When to use
spawn-agent	Spawn a customized CC subagent with full MCP tool access.
literature-research	Full-text paper reading via alphaxiv. Import from literature-engine.