Skip to content

Headline–Body Consistency Check

Consistency check — instantiates Summary-Substance Alignment Audit

Isolates the single claim a headline asserts and tests whether the body actually supports it — at that scope, tense, and force — before it goes out.

A Headline–Body Consistency Check is the fast, pre-publication gate on the shortest and most-seen surface there is: the headline. It extracts the one proposition the headline asserts and the proposition the body actually supports, and tests whether the first is entailed by the second — not just on topic, but at the same scope, tense, and force. Its whole discipline is narrowness and speed: it does not build a claim-by-claim matrix or weigh evidence, it renders a single verdict on a single dominant claim — supported, or the headline outruns its story — because a headline that says more than the body earns is a distortion at the scale of everyone who reads only the headline.

Example

A sports desk is about to publish the headline "City Sign Star Striker." The consistency check extracts the headline's claim — the club has completed a signing (completed action, done deal) — and the strongest claim the body supports — the player has agreed personal terms, but the clubs are still negotiating the fee and a medical is pending. Tested for entailment, the headline fails on tense and force: it asserts a completed transfer the body says has not happened. The verdict is unsupported (premature completion), and the headline is corrected to "City Agree Personal Terms With Striker, Fee Talks Continue" before it hits the wire — the same event, told at the strength the body can carry.

How it works

  • Extract the headline's single asserted proposition — the summary surface's claim, reduced to actor, action, object, scope, and tense.
  • Extract the body's strongest supported proposition on the same topic.
  • Test entailment, not overlap: does the body support the headline claim at its scope (all vs some), tense (done vs proposed), and force (is vs may)?
  • Classify any mismatch — unsupported, scope-inflated, wrong actor, or premature — and hand it back for a one-line fix.

Because it works on a single claim it is cheap enough to run on every headline; the whole-document mechanisms are not.

Tuning parameters

  • What counts as "the" headline claim — the single dominant proposition, or every clause including the deck and subhead; wider capture catches subhead drift at more cost.
  • Scope/tense/force sensitivity — how strictly the headline must match the body's quantifier, timing, and modality; this is the dial that separates a real mismatch from ordinary compression.
  • Entailment strictness — must the body support the headline, or merely not contradict it.
  • Gate timing — a hard pre-publication block or a post-publication spot-check.

When it helps, and when it misleads

Its strength is a cheap, fast check on the surface with the widest reach and the least room to qualify — catching the headline that says more than the story before it ships. Its degenerate case is clickbait: a headline engineered for the click rather than for fidelity to the piece beneath it.[1]

Its limit is that a headline can be literally consistent and still mislead by emphasis or selection — the check passes it, because it judges the dominant claim, not the tone. And a headline written after a check can be tuned to pass the entailment test while still steering the reader. The discipline is to test scope, tense, and force rather than topic alone, and to fold the deck and subhead into the surface being checked, since readers take them as part of the headline.

How it implements the components

This check fills the fast two-surface comparison at the front of the audit:

  • summary_surface_claim_set — the proposition the headline asserts, reduced to actor / action / scope / tense; the summary surface, extracted.
  • substantive_surface_claim_set — the strongest proposition the body actually supports, extracted so the headline claim can be tested for entailment against it.

It stops at a single dominant claim: it does not build the evidence-traced, claim-by-claim matrix for a whole document (Abstract–Full-Text Alignment Review, Summary-Claim Traceability Matrix), nor judge certainty or causal inflation specifically (Certainty & Causality Inflation Check), nor scan for dropped qualifiers (Qualifier-Drop Scan).

References

[1] Clickbait — a headline optimized for the click rather than for faithfulness to the story it fronts; the consistency check is the guardrail that keeps a headline earning its claim from the body. ↩