MagPunk
EN

Help & Frequently Asked Questions

The most wondered things and practical tips when using MagPunk daily. For a short summary of what the app does, see the home page; if you are curious about the calculation behind the scenes, see Understand MagPunk.

What makes MagPunk different?

MagPunk is not an ordinary news reader, it is a content analysis tool running on your device. It has no server, no account, and no data collection; classification, scoring, clustering, and reports are all done on your phone, offline. It groups sources that provide the same event and separates the one who breaks the news first from the follower, catches rapidly spreading events, and shows the reason for every label it gives word by word. So it both protects your privacy and leaves a serious analysis power in your hands.

Is my data safe?

No account, no telemetry; your reading history and settings stay solely on your device. The only thing that leaves your device is the search word that goes out when you say "Search web". See the Privacy Policy page for details.

Does MagPunk consume my battery and internet a lot?

All analysis (classification, scoring, reports) is done on your device and does not require internet. The internet is only used for two things: pulling items from sources and source discovery with "Search web". The background scan is periodic and lightweight; you can reduce usage by lowering the scan frequency and retention period from Settings.

Does it work when I have no internet?

Yes. Reading downloaded items, Saved, history, Reports, and all analysis work offline. The internet is only needed to pull new items and for Smart Discovery.

How are recommended sources determined?

Recommended is a curated catalog of 435 hand-prepared sources embedded in the app; it is organized by category and language. This is not an algorithmic ranking, but a maintained list. You can add and remove with a single tap, and use Smart Discovery for a topic or site not in the list.

Is MagPunk free? What is Pro?

Everything required for MagPunk's daily use is free: reading, analysis, scoring, the core of the Reports, watched words, the Daily Brief, and source management. The optional Pro subscription unlocks advanced features: comparison, PDF/CSV export, archive longer than 7 days, Daily Brief and Pulse customization (sources to show and threshold score), control over scoring settings, protected items, and unlimited tag/word tracking (5 of each in the free version). No matter what tier you are in, there are no ads in the app; Pro is a form of support and ensures the app stays ad-free and privacy-first. You can review the subscription from the "Upgrade to Pro" section in the menu.

Who do the export features serve, and why?

You can export your archive and reports as CSV or PDF. This feature is especially useful for researchers, journalists, and analysts: for those who want to keep records, examine the data in their own spreadsheet program, or share a finding. All export is done locally on your device. Export is a Pro feature.

How are related items determined?

MagPunk extracts a "fingerprint" consisting of keywords from each title and measures the common word ratio (Jaccard similarity) of two items. If the ratio is 50% and above, it is considered a copy of the same item, if it is between 30-49%, it is considered a "similar item". Comparison is only made in the same language, between different sources, and in a 24-hour window; you can narrow and widen these thresholds from the "Score & Matching" section in Settings.

Why does it gather the same item under a single card?

We match sources that provide the same event and gather them in a single card; thus, you do not get overwhelmed by seeing the same item dozens of times. You can tap the card and see from the "related items" section which sources provided the event and when.

What is the Critical label?

Critical means an item needs to be seen without delay: events directly concerning life safety, disaster, or public order. A Critical item starts from the highest score and decays over time. The red "CRITICAL" badge drops 3 hours after publication; this does not mean the item has lost its importance, it merely means it is no longer considered "new". Critical items are privileged in notifications: your category, language, and source filters do not weed them out. For an issue to be considered Critical, it is not enough for the event itself to appear in the title, expressions verifying the event must also be present — "earthquake happened" is Critical, "did an earthquake happen?" is not.

What is the Viral label?

Viral means an item spreads to many sources in a very short time. For an item to be considered Viral, it is required that it is fresher than three hours and has at least one exact copy in other sources; merely "similar" items are not enough to be considered Viral. This label does not show that the item is correct or important, it only shows that it spread rapidly. The Viral badge remains active for 4 hours.

What is the Low Quality label?

Low Quality means the title is written not to inform you, but to get a share from search engines or clicks: "who is...", "watch live", "when", "what is the latest", in English "how to" or lists starting with a number. An item found Low Quality receives a score penalty and drops down in ranking; if you wish, you can completely hide it from the feed from the "Content" section in Settings. This is not a grade given to a source, it is a grade given to a single title; a serious source also receives this label when it publishes a service title.

What is the Sentiment label based on?

Sentiment looks at the outcome of the event, not whether the words in the title are individually positive or negative. The only question is this: what did this event change compared to yesterday's situation? If a new harm or grievance emerges, it is negative (attack happened, price spiked), if an existing harm ends or a grievance is resolved, it is positive (missing found alive, gang busted), if only a process is running and there is no clear outcome, it is neutral (investigation launched, appointment made). Because of this, "theft was committed" comes out negative, "thief was caught" comes out positive. Wherever there is indecision, it is called neutral; furthermore, items with political and ideological context are consciously labeled neutral, because calling the routine operation of the state "good" or "bad" would be teaching the model a political judgment.

How accurate are the item labels?

Labels are a statistical estimate; they are not flawless. MagPunk is tuned with the principle of "prefers missing over pressing a wrong label", meaning it prioritizes not showing an item erroneously as "Low Quality" or "Critical". The accuracy measured in a real feed is roughly in the 70-75% band; the exact numbers are in Honesty on Display on the home page, and the details of the method and its limits are in Understand MagPunk.

I found a label wrong, can I correct it?

Labels can be mistaken. You cannot change a label by hand inside the app — that is deliberate: a correction on a single device does not train the model, it only distorts that one item. You can report the error to us instead. Tap the "report error" link in the reading panel, mark which label is wrong or which one is missing, and press "Continue". The report is copied to your clipboard and a submission page opens in the browser; nothing leaves your device until you press "Send". You can also see the decision word by word with the "why this label?" explanation of every item. Reports that reach us are reviewed by hand; those that help improve the model go into the next training round.

Why are only titles scanned?

This is a conscious privacy and speed preference: processing only the title allows the decision to come out instantly on your device without sending anything to the internet. Furthermore, while RSS feeds provide the title reliably, the full body of the item is often incomplete or inconsistent. The tradeoff for this is that a misleading title can also mislead the model; see Understand MagPunk for details.

Why does the item summary not show completely?

The reading panel limits the summary to the first 60 words so it stays quickly scannable. To read the entirety of the item, you can switch to the "Web page" view; you can select the default view from the "Content" section in Settings.

Which settings can I change myself?

Retention period, matching thresholds (Score & Matching), notification score/interval/count, watched words, Daily Brief hour, and Pulse threshold — all are under your control from Settings. Theme selection is not in settings, but in the palette icon in the side menu. The full scope of settings is listed under the Settings heading of the app's "Features" section.

How does it scan items when the app is closed?

MagPunk conducts its background scan with Android's WorkManager system; for this, it does not need a persistent "running" notification constantly staying in your notification drawer. The scan is periodic (at least every 15 minutes) and lightweight. Your phone may delay the scan in some cases to protect the battery; if notifications are delayed or not arriving at all, you can increase reliability by turning on the battery optimization exemption (and auto-start on some brands) from the "Is the app not running in the background?" card at the very bottom of Settings.