Why this page exists

HappeningNow uses automation heavily, and readers are entitled to know which parts of a page were produced by a language model and which were calculated. This page draws that line explicitly rather than leaving it to be inferred.

Where language models are used

Language models are used in a narrow set of places, all of them summary text derived from reporting the platform has already collected:

• Short story summaries shown alongside a story, generated from the publisher-provided title and description for that story.
• Short summaries of a group of reports covering the same event.
• Text in Intelligence briefs, which are built from already-collected story records.

Where language models are not used

The parts of the platform that make things appear, connect, or count are rule-based code, not model output:

• Reporting Context: every value is provenance from the source or a calculation over story records.
• Grouping related reports: fixed-threshold, rule-based comparison.
• Categories and tracked topic names: published taxonomy and curated lists.
• Coverage counts, distinct-domain counts, timestamps, and freshness windows: computed arithmetic.
• Which sources the platform reads: a maintained list, not a model decision.

What automation is not permitted to do

Summaries are constrained to condensing text the publisher already published. Automation on HappeningNow does not add facts that are absent from the source report, does not resolve disagreements between publishers, does not rate publications, does not score claims for truth or bias, and does not present predictions as outcomes.

Editorial judgement about what a story means is not delegated to a model. Where the platform has nothing deterministic to say, it says nothing.

How to recognize automated text

Generated summaries are labelled where they appear, and every summarized story links to the original report so the publisher's own words are one click away. Anything not labelled as a summary is either quoted provenance from the source or a calculated value described in the Methodology.

Known limits

Automated summarization can compress away qualifiers, attribution, and uncertainty that matter in the original report. A summary is a pointer to reporting, not a substitute for it, and the linked publisher report always takes precedence over a summary of it.

Summaries are generated without a human reading each one before it is published. When a summary misrepresents its source report, that is an error in HappeningNow's own output and is handled under the corrections process.