Relational Co-Occurrence Sieve

Two real facts can still compose a lie. After confirming each entity exists, this sieve pulls the subject's real Wikipedia page and checks that every other name, date, and place actually appears on it — catching 'Lincoln used an iPhone in 1865.'

Launch the tool → Open the journal

tools/relational-co-occurrence-sieve/

Live preview · launch for the interactive version

§ A · What it makes visible

three hidden mechanisms

Fig. 01

Level 1 · existence

Every proper-noun phrase is verified as a real article first.

Fig. 02

Level 2 · co-occurrence

The subject's page is read, and each other anchor is checked for presence on it.

Fig. 03

The relational lie

Real entities with no shared record fail — a Type-2 hallucination, exposed.

§ B · How to investigate it

run it like an experiment, not a toy

01 · Predict

Before the audit

Predict whether the entities truly belong together.

guess: iPhone doesn't fit 1865

02 · Change one thing

Swap one relationship

Keep the entities real; change how they relate.

true date → false date

03 · Compare evidence

Read both levels

Passes existence but fails co-occurrence — or both?

L1 pass, L2 fail = relational lie

04 · Name it

Name the lie

Say where it lives: a fake entity, or a fake relationship.

Type 2 = false relationship

§ C · Debrief questions

after the investigation

Why isn't 'all entities are real' enough?

What does co-occurrence on one page actually prove?

Where could this check produce a false alarm?

How would you teach this with the Lincoln example?

§ D · Related

pairs well with · use in context

Pairs well with

Network-Grounded Truth SieveThe Level-1 existence check Confidence Is Not TruthFluency vs. correctness

Use it in context

Optional StudioRun a model-behavior investigation Further ReadingStochastic parrots & more Investigation JournalCapture your evidence