When the news comes every day, the main question changes: not “what happened” but “where is all of this heading”. Analytics answers it with numbers: how much arrives, how much proves important, what the processing costs, which sources work and which stay silent.
The Home summary
The “Home” screen answers “how are we doing” in five seconds: how much material arrived over the day, how many events, how many of them wait for your attention, how many drafts await approval. It also shows how the quotas are spent — how many AI tasks are used and how many remain until the next day.
Volumes and materiality
- By day — how much material and how many events arrived each day. Both growth and sudden bursts are visible.
- By materiality band — how many events turned out noise, how many “worth knowing”, how many “material” and how many “existential”.
When the share of “existential” suddenly grows — that is a reason to look into what is happening, before the next alert arrives.
AI spending
AI processing is the main cost line of news watching, so it is visible in parts: how much went to summaries, how much to materiality ratings, how much to topics, entities and publication drafts. The by-day breakdown shows whether the spending grows, and on what exactly.
This is a case where the number is not a report for the sake of reporting but a working tool: seeing that spending goes to cheap processing, you can narrow the volume or adjust the materiality rules.
Sources
- Which sources give material — who sends the most, and who sends nothing.
- What is silent — sources that have not brought anything new for a long time. Maybe the feed closed, or you subscribed to the wrong thing.
- What is failing — sources with errors and failure streaks.
This is usually where the cleanup starts: the extra sources are removed, the weak ones are set up again.
Noise
A separate number — how many alerts were not sent, and why. It matches the suppression journal from the Alerts & Noise Protection section: when it says “40 messages suppressed by the rate limit” there, the same forty shows here.
This number is not “loss” but benefit: forty messages you did not read are forty saved minutes.
Archive search
The search runs over the whole collected archive — over all the material, not only today’s. What is found is highlighted, so it is immediately clear why the record made it into the results.
Honestly about the first version: this is a word-by-word search, with no semantics. The query “what they wrote about logistics getting more expensive” finds records containing those words, but will not understand the request phrased differently. It works fast and predictably, but no miracle should be expected of it.
Exports
Data sets can be exported to process them separately: in a spreadsheet, in your own reporting program, wherever is convenient. Useful when the interface’s numbers are enough for watching but you need your own cut — for example, to compare your price behavior with the market’s.
Where the numbers come from
All the numbers are computed on the server in one place. This means the analytics screen and the agent’s answer to “how much arrived this week” carry the same number. Divergence between screens is impossible by construction, not because someone carefully reconciled the balance.
Next → File Configuration