Half of our 12-story sample centred on programmes, plans or guidance

A count of our own coverage

Six of the twelve briefs in this audit were primarily about an institution's programme, plans or guidance. Four led with reported figures, one with an experiment and one with a modelled projection. We reached those counts by classifying a fixed set of GlobalPulse articles, then recording a primary label for each. The result describes a small part of our own publication. It also reveals an editorial problem worth addressing: several kinds of evidence can arrive in the same stream of official updates, looking equally settled when they are displayed as matching news cards.

We selected the twelve institutional-source briefs whose original publication dates run from November 2 to November 4, 2026. The cohort contains the twelve records already selected for this edition before this notebook and its companion comment were added. Earlier October briefs are outside the cohort. The two new first-party pieces are outside it too. The downloadable table identifies every included record, so later changes to homepage order will not silently change the denominator. Reproducing the count means using those twelve identifiers rather than selecting whichever twelve articles happen to appear first.

This was a study of our selection and our descriptions. We read the local article records, including their headlines, excerpts, factual paragraphs and analysis, and assigned the label that best described the principal development. We did not draw a random sample of institutional communications. Publication language, available sources, our section mix and previous editorial choices all helped determine which articles entered the set. A percentage from these twelve records cannot establish the frequency of announcements in international news, the quality of a particular institution's work, or readers' exposure to any topic.

Four primary labels

The four labels are working definitions. Reported figures means that the lead development is organised around a numerical account. Institutional action, programmes or guidance covers a policy process, operating plan, guidance document or programme update. Experiment means that a described test is the central development. Scenario projection means that an estimate of possible outcomes does most of the explanatory work. We selected one primary label per record to make a simple tally possible. That choice reduces a complicated article to one field, which is why the definition and a short coding reason belong beside every row.

The reported-figures group contains four entries. Its name must be read carefully. A numerical report can contain an observed historical series, a publisher's financial account, estimated future requirements and recommendations in the same document. Giving it this label does not certify every number as an independently established measurement. Our decision concerns what the short article leads with. Readers checking a row should still ask about the period covered, the unit, the denominator and the organisation responsible for the figure. A precise-looking total can have a broad boundary or an important qualification.

Six entries fall in the institutional-action group. The group includes guidance and programme accounts as well as announced plans. Some describe activity that had already happened or was under way when the source published its account. The label therefore does not mean six promises about the future. Nor does it mean six completed public benefits. Its practical use is to signal that an institution's account of an activity needs a different follow-up from a historical numerical series. The next relevant record might be a decision, a revised programme, a delivery milestone or an evaluation.

One entry was coded as an experiment because a test supplied the main development. One was coded as a scenario projection because possible benefits estimated under assumptions supplied the main development. These labels explain the emphasis of those two briefs. They do not form a ranking in which a test automatically outranks a model. A test has a setting, an instrument and operating constraints; a model has inputs, relationships and assumptions. Each can answer a useful question within a defined scope. Neither label by itself supplies an effect size, a causal explanation or a guarantee of performance elsewhere.

Overlap and sensitivity

Overlap is the central difficulty in this exercise. A programme announcement can cite participation figures. A projection can incorporate historical data. A numerical assessment can end with recommendations. We resolved overlap by asking what a reader would identify as the article's principal new development, then recording a reason for that decision. Another editor could reasonably select a different emphasis for a borderline record. Keeping the row-level explanation visible is preferable to presenting the four totals as if they had emerged from a perfectly objective sorting mechanism. The tally is an inspectable editorial judgment.

The small denominator makes recoding consequential. Moving one record changes a group's share of the sample by roughly 8.3 percentage points. If a reported-figures entry were recoded as a projection, the respective counts would become three and two, while the other groups would retain their counts. If an institutional-action entry were recoded as a numerical report, those groups would contain five entries each. These are sensitivity examples, not additional observations. They show why a single headline percentage should travel with the definitions, the complete table and the reader's ability to inspect individual decisions.

What the labels can establish

We did not use these labels to score truthfulness, importance or editorial independence. A programme update may concern a consequential decision, and a numerical release may still require careful scrutiny. We also did not count readership, search visibility or public impact. Those variables are absent from the records used here. The experiment offers a narrower result: our own selected stream contains several evidence types, and their presentation should help readers distinguish them. Treating that result as a verdict on institutional publishing would claim considerably more than this dataset can support.

Source and site dates need their own columns. An organisation may publish an account after an event, an observation or a reporting period has ended. GlobalPulse may prepare its article later still. Those are separate clocks. We retained the source's supplied date precision rather than creating publication hours. The table's inclusion rule uses source publication dates, while the article records retain their own editing history. That separation makes the cohort reproducible and prevents an editorial revision from turning an older development into an apparently new event.

Review and reproduction

Our coding was prepared with AI assistance and checked against the twelve local article records. It was not a blinded exercise with multiple independent coders. We have no agreement statistic to publish and no reason to manufacture one. A stronger future comparison could give the same frozen records and definitions to another reviewer, retain both sets of decisions and explain disagreements. That would test how consistently the definitions can be applied. It would still leave the selection limits intact: two reviewers agreeing about this corpus would not make it representative of all news.

Reproduction should be straightforward. Download the table, check that it contains twelve unique article identifiers, confirm the date range, and group the rows by their primary label. The counts should sum to twelve. Then read the coding reasons alongside the linked GlobalPulse articles. A reproduction that produces another total should first check whether it used the same cohort and label definitions. A substantive disagreement should identify a row and an alternative rationale. That creates a traceable editorial correction instead of an argument about the appearance of the overall chart.

An editorial baseline

The immediate editorial change is a better follow-up record. An article about reported figures should preserve the period, units and source of the account. A programme story should identify the next decision or delivery record a reader could look for. An experimental story should preserve the test setting and the next operating constraint. A projection should identify assumptions and distinguish estimated gains from realised outcomes. These are four different jobs. Giving all four the same closing paragraph about uncertainty would discard the information the classification was designed to reveal.

A later recoding should preserve the earlier file as a dated version and publish a note identifying the changed row. Silent replacement would leave the headline count intact as a historical claim while removing the evidence behind it. This is a publication rule for this notebook: corrections should allow a reader to reconstruct both the original tally and the revised one.

This audit is small enough for a reader to challenge in full. That is an advantage for an initial notebook: every inclusion and every label can remain visible. We would rather publish a modest, reproducible account of twelve editorial choices than attach a grand claim to a thin denominator. The table supplies a baseline for our own next review. A later review can compare a fresh, explicitly selected cohort with this one, while keeping changes in definitions and selection visible. The work begins with exposing our decisions to inspection.