How it works

dissemin8ed is a real-time news aggregator. It pulls from hundreds of RSS feeds, clusters duplicate stories, scores them by signal strength, and surfaces what matters — without an algorithm designed to maximise outrage.

Reading a card

Each card is one story cluster — the same event reported by multiple sources, grouped together so you see the full picture rather than the same article repeated.

• Thumbnail — lead image from the first source to publish.
• Coloured dot — political bias of the lead source (see Bias below).
• Source count — when multiple sources cover the story, a N sources ▼ button expands to show them all. Alongside it, a compact 2L 4C 3R spectrum distribution shows how many sources from each side covered it. The source marked 1st was earliest to report. When only one source exists, the outlet name and its bias dot appear directly.
• Sparkline — a tiny coverage curve next to the source count showing how interest built over time.
• Signal bars — a 0–4 bar indicator of how broadly the story is covered right now.
• X ago — how long since the story first appeared on dissemin8ed. Hover for the exact timestamp.
• ⚡ Breaking / ⚡ Spreading badge — Breaking = coverage rate ≥4 articles/hour; Spreading = ≥1.5/hour. No badge = normal pace.
• ⚖️ Cross-spectrum badge — the story is covered by sources from both sides of the political spectrum.
• 🧵 Thread pill — the story belongs to a continuing situation. Click to filter the feed to that thread.
• 🔒 Paywall badge — all sources covering this story may require a subscription.
• ★ Bookmark — save to your Bookmarks list. Subscribers sync across devices; free users store locally.
• AI summary — an italicised one-liner generated by AI. Only shown when available.
• Source rows — expanding the sources list shows a row per outlet with bias dot, paywall indicator, article title, and timestamp. Subscribers can block a source from its panel.
• Share buttons — 𝕏, LinkedIn, Reddit, copy link. Appear in the card actions row.
• 👍 / 👎 Reactions — rate a story. Select a reason to help surface your preferences.
• Translate badge — appears on non-English headlines. Subscribers get full AI translation; free users see a lock prompt.

Source bias ratings

Every source is rated on a five-point scale based on AllSides and MBFC ratings. These reflect editorial perspective, not accuracy.

Sources not in the bias database show no dot. The ⚖ Cross filter shows only stories covered by sources from both sides of the spectrum.

Category tabs

The broadest lens. The row of tabs at the top of the homepage — All, Politics, Business, Tech, Sports, Culture — filters the stream to stories in that category. Switching tab reloads the feed without losing your other active lenses (sort, bias, paywall).

How categories are assigned

Every article is categorised at ingest time using semantic embeddings. The article’s title and summary are converted to a numerical vector, then compared against anchor vectors for each category using cosine similarity. This catches synonyms, paraphrases, and multilingual variants that keyword lists miss.

Topics

A finer-grained lens than categories. Topics are tags attached to each article; clicking one narrows the stream to stories carrying that tag. A category and a topic can be active at the same time — e.g. Tech + AI shows only AI stories filed under Tech.

How topics are assigned

1. Keyword matching — each article title and summary is checked against curated phrase patterns per topic. Multiple topics can match.
2. AI fallback — articles matching zero keywords are sent to GPT-4o mini, which picks 1–3 topics from the same list.

All topics

The count badge on each pill shows how many stories carry that tag right now.

Trending phrases

The sharpest lens for isolating a single ripple. Trending pills show statistically significant coverage spikes — phrases appearing at 3× their baseline rate in the current 6-hour window. Clicking one filters the stream to every story mentioning that phrase right now, cutting through everything else. They signal accelerating newsroom attention, not social virality. Only 2–3 word phrases (bigrams and trigrams) can trend; single words never appear.

The algorithm

• Window: English articles published in the last 6 hours.
• Baseline: the previous 7 days split into 6-hour buckets (28 windows), averaged.
• Spike threshold: a phrase must be at least 3× its baseline average to qualify.
• Minimum volume: at least 4 articles must mention the phrase in the current window.
• Near-duplicate suppression: morphological variants (“strike Iran” / “strikes Iran”) are collapsed via stem-based matching. A final AI editorial pass then drops templated non-news phrases (sports fixture previews, financial boilerplate) and catches any remaining semantic near-duplicates.

Story threads

A thread is a continuing situation — the same evolving story covered repeatedly over days or weeks. Where a story cluster groups today’s reporting of one event, a thread connects the whole arc: the initial event, follow-ups, reactions, and developments.

Threads appear as a collapsible strip below the category tabs when any are active.

How threads are detected

1. Embedding similarity — every story is represented as a semantic vector. Pairs of stories with cosine similarity above a threshold are connected into candidate clusters.
2. AI validation — each candidate cluster is reviewed by a language model, which confirms whether the stories genuinely form a coherent continuing situation or just happen to sound similar. It also generates the thread’s name and one-sentence description.

Threads are recomputed every 6 hours. Only stories from the past 14 days are considered.

Arcs — mega-narratives

An arc is the highest editorial layer: multiple related threads grouped into a single long-running mega-narrative. Where a thread tracks one evolving situation, an arc connects the whole constellation — the policy debate, the legal case, the geopolitical standoff, and all the sub-stories that orbit it.

Arcs appear in a collapsible ◯ Arcs section on the homepage, between the thread strip and the main feed. They are collapsed by default; click the header to expand.

Arc cards

Each arc card shows:

• Phase badge — how active the arc is right now: Breaking accelerating fast (velocity > 2.5 stories/day), Developing steady ongoing coverage, Fading tapering off, Dormant no new stories in 24h but still active within 72h or too large to be done — displaced, not concluded, Resolved fully quiet across 72h (small arcs only).
• Velocity bar — a visual representation of the arc’s 24-hour story intake rate relative to its peak.
• Thread timeline — the individual threads that make up this arc, listed chronologically. Clicking a thread name filters the feed to that thread.
• Story & source counts — total stories and distinct sources across all threads in the arc.
• ↗ Best story — opens the highest-scored story in this arc.

How arcs are detected

Arc detection runs on the same two-agent pipeline as threads, but at a higher level of abstraction:

1. Embedding similarity — each thread’s name and description are embedded. Threads with cosine similarity above a lower threshold (0.45 vs. threads’ 0.72) are grouped into candidate arcs — intentionally looser, because arcs span broader territory.
2. AI validation — a language model reviews each candidate group and decides whether the threads genuinely form a coherent mega-narrative. It must be at least 70% confident to accept the grouping. If accepted, it generates the arc’s name and one-sentence description.

Arcs are recomputed every 6 hours alongside thread and embedding updates. A snapshot of each arc’s phase, velocity, and thread count is recorded at every run, building a time-series history of how each mega-narrative evolves.

Cascades — escalation ladders

A cascade is a special class of arc where the threads don’t just share a topic — they form a causal chain. Each event provokes or enables the next in a clear provocation → response → counter-response pattern. Trade wars, military exchanges, legal battles, diplomatic crises, and corporate rivalries are common cascade shapes.

Cascades appear as the ⚡ Cascades tab inside the Arcs section on the homepage. Only arcs with at least three threads, where an AI analyst is ≥70% confident a causal chain exists, qualify.

Cascade cards

A cascade card extends the arc card format with a ladder view:

• Rungs — each step in the escalation chain is listed in order. Every rung shows the actor, the target, the type of action (e.g. military strike, sanctions, legislation, statement), a one-line label describing the event, and a severity bar from 1–10. Low severity rungs are grey; medium are amber; high are red.
• Reply-to chain — rungs that are direct responses to an earlier rung note which rung they answer.
• ⚠ Next predicted rung — a dashed amber box at the bottom of the ladder shows the AI’s prediction of what the next escalatory move is likely to be, based on the pattern so far. This is a probabilistic forecast, not a certainty.

How cascades are detected

1. Candidate arcs — only arcs with three or more threads are considered.
2. AI analysis — Claude reviews the threads in chronological order and determines whether they form a genuine escalation ladder (distinct actors, clear provocation–response pattern, events causally dependent on predecessors). It must be ≥70% confident to accept.
3. Rung extraction — if accepted, Claude extracts each rung’s actor, target, action type, severity (1–10), and causal links, and generates the next-rung prediction.

Cascade detection runs every 6 hours alongside arc updates. Arcs that have already been analysed within the past 6 hours are skipped unless forced.

Sort order & scoring

Use the sort buttons in the filter bar to switch modes:

▲ Ranked 🕐 Newest

▲ Ranked (default)

Cards are ordered by a composite signal score recalculated on every page load:

• Freshness — exponential decay with a 24-hour half-life.
• Source count — more outlets covering the same event signals higher importance.
• Cross-spectrum coverage — stories covered by both Left and Right sources rank higher.
• Paywall penalty — fully-paywalled stories score slightly lower.

The ⚡ Breaking filter (bias bar) shows only stories with a coverage rate ≥4 articles/hour. The ⚡ Breaking and ⚡ Spreading chips on individual cards indicate accelerating intake (≥4/h and ≥1.5/h respectively).

🕐 Newest

Cards are ordered by when the story first appeared on dissemin8ed — i.e. when the earliest article in that cluster was discovered. This is the arrival time into the system, not the article’s own publication timestamp (which is set by the publisher and can be back-dated or future-dated). Use this mode to browse the latest incoming stories rather than whichever ones have accumulated the most signal.

Filters & controls

Topic pills

Click any topic pill to filter the feed. All resets to the full feed. The country pill shows geo-detected local news.

Trending pills

The row above the topic pills shows phrases with accelerating coverage in the last 6 hours — each prefixed with a ↑ arrow indicating a rising coverage rate. Clicking one searches all stories mentioning that phrase.

Bias filter

All ● Left ● Lean L ● Center ● Lean R ● Right ⚖ Cross — filter cards by the lead source’s political leaning.

Language

A language dropdown appears when sources in multiple languages are active. All langs is the default.

Paywall toggle — 🔒 / 🔓

🔓 (default) — paywalled stories are shown dimmed with a 🔒 badge. Click to switch to 🔒 which hides paywalled articles entirely.

Personalisation — Bookmarks & Reactions

★ Bookmarks

• Click ★ on any card to bookmark it. A ★ pill in the toolbar opens the slide-in Bookmarks panel.
• The panel lists all saved stories with their headlines. Click any to open the story; click ★ again to remove.
• Subscribers — bookmarks are stored in your account and sync across devices and browsers.
• Free users — bookmarks are stored in your browser’s localStorage and are device-specific.

👍 / 👎 Reactions

• The bottom action row of every card has a 👍 Like and 👎 Dislike button.
• After clicking, a reason picker appears (e.g. important, misleading, clickbait). Picking a reason records your feedback and labels the button.
• Reactions are stored locally and are not sent to the server.

Source blocking ★

Subscribers can mute specific outlets so their rows are visually suppressed wherever they appear in the feed.

How to block a source

1. Click any source name in an expanded source list to open the source panel.
2. Click Block source. The button turns red and the outlet’s rows across all story cards are struck through and dimmed immediately.
3. The story itself remains visible — other sources that cover the same event are unaffected.

Managing blocked sources

• Click the 🚫 pill in the toolbar (visible once you have at least one block active) to open the Blocked Sources panel.
• Click ✕ next to any source to unblock it. Unblock all clears the list with one click.
• Blocked source lists are stored in your account and sync across devices.
• Blocking is also accessible from your account menu (desktop dropdown or mobile drawer).

View & text size

Layout

⊞ Auto — fills available width (default)   ▦ 2-column fixed   ▬ Single column

Text size

A− decrease   A+ increase — four steps from small to extra-large. Resizes titles, summaries, and AI summaries. Your choice is remembered.

Search

The search box in the header searches across all story headlines and summaries in two phases that fire simultaneously:

1. CDN phase (~50ms warm) — pre-built index shards are fetched from the CDN and matched against your query words. Results appear almost instantly.
2. API phase — a full-text database query runs in parallel; when it resolves, CDN-only results are upgraded with full story data and any database-only results are merged in.

Results are ranked by relevance (term weight × source count), not by recency. The bias filter applies to search results too.

Typing a source name in the search box shows a quick-select suggestion to filter the feed to that outlet’s chronological coverage directly.

Subscriptions

A free account gives you the core feed, category tabs, topics, trending, and basic bookmarks. A subscription unlocks the full platform:

• Full feed depth — free users see a capped feed; subscribers get the complete stream with infinite scroll.
• All arcs & threads — free users see a limited arc strip and thread chips; subscribers see every active arc and all thread chips.
• AI translation — non-English headlines show a language badge. Subscribers get one-click AI translation; free users see a subscription prompt.
• Bookmarks sync — bookmarks are tied to your account and sync across every device and browser.
• Source blocking — mute specific outlets so their rows are struck through across the entire feed (see Source blocking).

To subscribe, click Subscribe in the account menu or tap the unlock prompt that appears at the bottom of a gated feed.

Live feed

The /live page shows a real-time stream of every article as it arrives — unfiltered, unranked, in pure discovery order. New items slide in at the top as they are ingested; a ↑ N new items banner appears when the page is scrolled down so you don’t lose your place. Scroll to the bottom to load further back in time.

• Pause / Resume — freeze incoming items while you read. They queue up and flush when you resume.
• Infinite scroll back — keep scrolling to browse historical raw intake.
• There is no clustering, scoring, or filtering on the live page — it is the raw firehose.