The naive gap table, and the instrument that replaces it

1The trap

In the week of 6 July our pipeline joined articles from roughly two hundred and thirty source domains to 3,392 events covered by at least three domains each. Group those domains by TLD and count distinct-domain coverage, and the temptation arrives immediately: sixty-three Polish domains touched 497 events, while seven Spanish domains touched 684. No editorial conclusion survives contact with the fourth column of Table 1.

2The readout

The corrected question is narrower and better: given everything this matrix already fixes, an outlet's weekly volume within its beat, an event's total reach, was this particular absence still unlikely? Fig. 2 is what answering it will look like. Every number in it is invented; publishing real surprises today, before the protocol freezes, would let the design see the answers it is supposed to be judged against.

3The data

The matrix and every count on this page recompute from the downloadable extraction: incidence.csv, cohorts.json, summary.json, matrix-query.sql, and Fig. 2's invented values are marked as such in readout-specimen.json. The cohorts are provisional in the strict sense: membership is a TLD rule. A cohort like .com is mostly wires and international outlets, which is exactly why this page never calls cohorts markets. Earning that word takes a publisher-country audit of the full source panel against a frozen rule sheet, and that audit is scheduled as part of the preregistration, not retrofitted after it.

4Journal

2026-07 · why the interesting half is fake

A mockup that ran the null for real would be a pilot study wearing a costume, and its gaps would quietly shape the preregistration: thresholds picked after seeing which thresholds produce findings. So the real half of this page is confined to what the null treats as fixed, the incidence matrix and its margins, and the inferential half is confined to fiction. The scout pass and the go or no-go decision happen before any real verdict is computed.

2026-07 · what a verdict will never mean

OMITTED·SURPRISING is a statement about a frozen source panel, among events that panel observed, under a null preserving outlet activity and event reach. It is not a statement that an editor erred, that a story was objectively important, or that a country's media failed. Language stronger than that requires the audited source map and a source-availability record, and the eventual paper will say so in its limitations, not its footnotes.

2026-07 · what would kill it

Three things, all measurable. Degree-preserving randomization proving infeasible or degenerate at panel scale, in which case the null is theater. Event clustering precision too low, since under-merge and over-merge both corrupt the incidence matrix before any analysis touches it. And an unmapped residue of the source panel too large for cohorts to mean anything. Each gets a measured answer in the scout report before preregistration.

5Reported results

None yet. Under Rule 1, this experiment cannot appear in an edition until it has a bench entry. The first entry will be the preregistered omission-surprise run, whatever it finds, including nothing.