Intelligence

Signals are the heart of Mediastilo's pipeline, the scored feed of material it has pulled from your Sources and ranked against your Brand. It's where you see what Mediastilo found, how on-brand it judged each item to be, and what it decided to do with it.

What's in the feed

Every pipeline run reads your active sources and adds the items it finds here. Each signal carries:

The original article, its title, source, and link.
An alignment score, how closely the item matches your Brand. Higher scores are more on-brand.
A decision, what Mediastilo did with the item (see below).
Topics: the themes Mediastilo detected, used for filtering.

Items are listed newest-first.

States: what's happening with each item, and why

Every item shows one explicit state, and Mediastilo always records why (the score and reasons for anything it evaluated, or a plain explanation when it didn't):

Included: the item became (or contributed to) a draft. Related items about the same story share one draft. See Drafts & Publishing.
Set aside: Mediastilo scored the item and didn't draft it; either it scored below your alignment threshold, or its story was already covered. It keeps its score and reasons so you can see why, and you can always Force include it.
Needs review: a Set aside item that scored just under your threshold, close enough to be worth a human call. This is the queue to clear first.
Analyzing: a run is in progress and this item is being scored right now.
Not reviewed: the item is older than your freshness window, so the engine never auto-considered it. The state shows the exact window (e.g. "older than 36h"). You can still Force include it.

You set the alignment threshold, the auto-generate threshold, and the freshness window in your pipeline configuration. See Account & Settings. A wider freshness window lets Mediastilo consider older stories (up to a 5-day maximum) but analyzes more items, which uses more credits.

Linked documents

On a lot of sites, the article is the announcement and the document is the news: the piece says the city "secured funds", and the decision it links to says 49,887.50 EUR. So when Mediastilo ingests an article, it also picks up the files linked from it, PDF, DOCX and XLSX, and reads them as part of the source material.

Documents are read only for stories that pass your alignment threshold — the ones you're actually about to write from. A file hanging off a story that scored 12 never costs you anything.
Their contents go to the writer along with the article, so figures, dates and names in a draft can come from the decision itself instead of the summary of it.
A document Mediastilo couldn't read (a scan with no text in it, a file above your plan's page ceiling, a dead link) is never guessed at. It's declared as a gap on the draft, naming the document, so you know exactly what to open.
The same file linked from several articles, or followed by several people, is read once. You're not billed twice for one council decision.

Reading a PDF is metered, because a one-page decision and a 200-page annex are not the same job. See Credits & Billing. Spreadsheets and Word documents are read without AI, so they cost nothing.

What the chips on a story mean

Every document an article links to shows up as a chip under the story, with its file name, its length, and what Mediastilo did about it. Clicking a chip opens the original file.

Chip	What it means
read	Its text went to the writer. A figure in the draft can have come from here.
being read	It's in the queue. It'll be read before the draft is written.
not read	Nobody asked for it — the story didn't pass your threshold, so the document was never opened and never billed.
couldn't read	Mediastilo tried and failed (a scan with no text, a page ceiling, a dead link). Hover the chip for the reason. This is the one that becomes a gap.

A couldn't read chip is worth a look: it's the article whose numbers nobody has verified. Open the file yourself and fill the gap on the draft.

Grounding stories with Google

Linked documents answer "what did the source actually say?". Grounding answers the next question — "what does the live web say?" — by running a Google Search and handing the writer a short, cited answer to work from.

On plans that include it, you can switch two per-brand toggles on in Settings so the overnight pipeline does this automatically:

Ground with Google researches each story before it's written, adding sourced context the article itself didn't carry.
Fill gaps with Google takes the gaps a fresh draft declares and tries to source each — turning "the grant amount is missing" into a cited figure where the web has one.

Both are off by default, because each run spends metered credits, and both obey the same rule as everything else here: only a cited fact is used. A search that comes back without a real source is left as a gap, never written in as a guess. You can also ground a draft by hand, or ask the assistant to — see Drafts & publishing.

Filtering and searching

The feed is built for triage. You can narrow it by:

Search: free-text across titles and content.
Source: search your sources and pick one to show items from that source only.
Score range: a minimum and/or maximum alignment score.
State: needs review, included, set aside, or out of window.
Topics: one or more detected themes.
Date range: items ingested between two dates.
Sort: by recency or by score.

The filter bar shows live counts per decision bucket and the list of distinct topics, so you can see the shape of your feed at a glance.

Overriding Mediastilo's decisions

You're always in control. On any item you can:

Force include: generate a draft from an item regardless of its score. Use this when you spot something valuable that scored lower than your threshold.
Force exclude: set an item aside and teach Mediastilo to deprioritise similar content in future runs.

Force-including an item generates a draft, which uses credits. See Credits & Billing.

Submitting an article yourself

Found something off your feeds that you want covered? Submit a URL and Mediastilo will ingest that single article, score it against your Brand, and add it to the feed like any other item, ready to force-include or let the pipeline pick up. The article is read in full — the same way a scraped or syndicated one is — and the documents it links to come with it.

Submitting queues a pipeline run for you, so you don't have to wait for the next scheduled one. If your Brand ran very recently, the item is scored at the start of your plan's next refresh window rather than immediately — it stays queued either way, so nothing you submit is lost.

You can also paste a document. A link straight to a PDF, DOCX or XLSX (a decision, a council agenda, a budget sheet) becomes an item that is that document, and it's read right away: submitting it by hand is you asking for it. There's no article to score, so force-include it to write from it. Reading a PDF this way is metered like any other, so it costs credits.

Manual submissions are rate-limited to keep the pipeline healthy. If you hit the limit, wait a moment and try again.

How decisions improve over time

Your include/exclude choices aren't one-offs, they're signal. Excluding items similar to ones you don't want, and including the ones you do, sharpens which material Mediastilo surfaces on future runs, alongside refinements to your Brand.