How we choose

How Huntaegis decides what to publish

Every score, filter and rule that picks, ranks or hides a story, in plain words, with the settings in force today.

These measure tone and language, not truth

These measures describe the tone and wording of the text. None of them measures whether a story is true. A calm, accurate report of something terrible can score as negative; a confident falsehood can score as neutral. Read the numbers as a description of how a text is written, never as a verdict on whether it is right.

For what the reading score and the objectivity dial show on each story, see how the scores work.

1. Which sources get looked at

Before anything is judged, the pipeline decides which feeds to read this time. None of this looks at what a story says.

Which sources are read

Sources per sweep

What it measures: How many feeds are read in one pass. The list is worked through in a fixed rotation, so every source gets its turn.
Range: A whole number
Setting on Huntaegis: 60 sources per pass, out of 354 in the list, one pass every 7 minutes.
What it cannot see: Which sources are good. The rotation is blind to quality; a source is in the list or it is not.

Which sources are read

Newest entries per feed

What it measures: How many of a feed's most recent items are considered each time it is read.
Range: A whole number
Setting on Huntaegis: The newest 3 items of each feed, skipping any we already published.
What it cannot see: Anything older than the newest items, even if it is important.

2. What a story must pass to be a candidate

These are filters on the page itself, not judgements of the story. A story that fails one never enters the pool.

Filter

Duplicate check

What it measures: Whether we have already published this headline, or this headline with the same text.
Range: Yes / no
Setting on Huntaegis: Anything we have already published is skipped.
What it cannot see: The same story told by a different outlet with different words: that is a new candidate.

Filter

The page must be readable

What it measures: Whether we can fetch the article's full text from the source. If a page cannot be fetched, it is remembered as unreachable for a few days.
Range: Yes / no
Setting on Huntaegis: Several fetch methods are tried in turn; a failing address is not retried for 3 days (7 days if it was not found).
What it cannot see: Whether the text we got is the whole article rather than a teaser.

Filter

Paywall wording

What it measures: Whether the page text contains stock paywall phrases such as "subscribe to continue reading".
Range: Yes / no
Setting on Huntaegis: 5 phrases are checked.
What it cannot see: A paywall that uses different words; a story that merely quotes one of the phrases.

Filter

Minimum length

What it measures: How much text was extracted.
Range: Characters
Setting on Huntaegis: At least 200 characters; if the page looks like a bot-check, at least 800.
What it cannot see: Whether short text is complete: a good short item and a truncated long one look the same to a length test.

Filter

List-article filter

What it measures: Short pages that read like "7 things you need to know".
Range: Yes / no
Setting on Huntaegis: Dropped when under 800 characters.
What it cannot see: A long list article; a short, serious item that happens to be phrased as a list.

Filter

Machine-written markers

What it measures: Whether the page announces itself as automatically generated.
Range: Yes / no
Setting on Huntaegis: 3 phrases are checked.
What it cannot see: Machine-written text that does not say so (Sentinel, below, estimates that but never blocks).

Filter

Language check

What it measures: Which language the text is in, from its first characters. Only English text is scored for tone and objectivity; other languages are still read and are translated to English at publication.
Range: A language code, or "unknown"
Setting on Huntaegis: Needs at least 50 characters; looks at the first 1,000.
What it cannot see: Mixed-language text, or very short text: those are left unscored rather than guessed.

3. Scores computed on every candidate

These numbers describe the language of the text. They are stored and shown beside each story. They measure tone and wording, not whether anything in the story is true.

Score

VADER sentiment

What it measures: How positive or negative the wording is, word by word, using a fixed list of emotionally loaded words.
Range: -1 (very negative) to +1 (very positive)
Setting on Huntaegis: Computed, not thresholded. Only the first 29,999 characters are read.
What it cannot see: Truth, sarcasm, irony, context, or negative news reported neutrally. A calm account of a disaster scores as negative; an angry opinion with mild words scores as neutral. English only.

Score

TextBlob subjectivity

What it measures: How much of the wording is opinion-like rather than factual-sounding, again from a word list.
Range: 0 (factual-sounding) to 1 (opinion-like)
Setting on Huntaegis: Computed, not thresholded. TextBlob's polarity is not used anywhere; VADER is our only positive/negative measure.
What it cannot see: Whether a claim is correct. A confident false statement can read as factual; a careful true one can read as opinion. English only.

Score

Objectivity score

What it measures: One minus subjectivity, as a percentage: how neutral the wording sounds.
Range: 0 to 100 (higher = more neutral wording)
Setting on Huntaegis: Computed from the subjectivity score. Used for the dial on each story and to rank the daily email digest.
What it cannot see: Fairness, balance or accuracy. It rates tone, not truth.

Score

Reading difficulty

What it measures: How hard the prose is to read, from sentence length and word difficulty.
Range: 0 to 100 (higher = harder)
Setting on Huntaegis: One standard formula (Flesch-Kincaid grade), divided by 20 and capped at 100. Not used to choose stories.
What it cannot see: Quality. Dense can mean careful or padded; plain can mean clear or shallow.

4. How the one story per pass is chosen

Each pass publishes at most one story. Candidates are matched against a list of topics ("directives"), each with a fixed priority number.

Used in the pick

Topic keyword match

What it measures: Whether the title or text contains any of a topic's keywords as a whole word.
Range: Yes / no per topic
Setting on Huntaegis: 15 topics, each with its own keyword list.
What it cannot see: Meaning. A story that mentions a keyword in passing matches; a story about the topic in other words does not.

Used in the pick

Topic priority

What it measures: A fixed number we set for each topic: the base score of any story that matches it.
Range: A number per topic
Setting on Huntaegis: Highest to lowest: Active Threat Campaigns (98); Ransomware and Extortion (97); State-Sponsored Cyber Operations (96); … down to General Cybersecurity News (10).
What it cannot see: How important or newsworthy this particular story is. The priority belongs to the topic, not the story.

Used in the pick

Tone term in the pick

What it measures: On this site the pick adds the strength of the VADER sentiment (0 to 1, positive or negative) to the topic priority, so among stories on equally ranked topics the more emotionally worded one wins, and a strongly worded story can outrank a plainer one on a topic one step lower.
Range: Adds 0 to 1
Setting on Huntaegis: Applies to every topic on this site (no topic sets its own profile).
What it cannot see: Importance or accuracy. It favours charged wording.

Used in the pick

Topic rotation

What it measures: Topics used by the most recent posts are set aside for this pass, so the feed does not run one topic in a row.
Range: A count of recent posts
Setting on Huntaegis: The last 10 posts' topics are skipped, unless nothing else matches.
What it cannot see: Whether a repeated topic deserved the slot.

Used in the pick

Ties

What it measures: If two stories score the same, the one that appears first in our list order is taken.
Setting on Huntaegis: No randomness, no recency, no quality measure breaks a tie.
What it cannot see: Anything: a tie is settled by list order.

Used in the pick

Reader submissions

What it measures: Stories a signed-in reader submits skip this selection entirely and are published as submitted (private by default unless the reader chooses otherwise).
Setting on Huntaegis: No selection score applies.

5. Added after publication

These run on stories that are already published. They add information for readers and never decide what is published or where it appears.

Added after publication

Red / Blue / Purple analyses

What it measures: Three written perspectives on the story produced by a language model: the facts, a balanced summary, and pattern analysis.
Range: Text
Setting on Huntaegis: Run on demand and in the background.
What it cannot see: Errors of the model. The analyses are a model's reading of the text, not verification.

Added after publication

Sentinel

What it measures: A language model's estimate of whether the text was written by an AI.
Range: 0 to 1 confidence, and a verdict
Setting on Huntaegis: Verdicts: Human, Likely Human, Likely Synthetic, Synthetic, Uncertain.
What it cannot see: Whether the claims are true. AI detection is unreliable on short or edited text.

Added after publication

Anti-pattern scan

What it measures: Whether the text uses any of 48 named rhetorical patterns (for example appeals to fear or false dilemmas).
Range: Named patterns found
Setting on Huntaegis: Reported in the Purple analysis.
What it cannot see: Intent. A pattern's presence is a prompt to look closer, not a finding of bad faith.

6. What readers and channels see

Where the scores above end up, and the only places a score affects anything.

Where it shows up

Daily email digest

What it measures: The most neutral-sounding stories of the last day, ranked by the objectivity score. Only for people who opted in.
Range: Top stories
Setting on Huntaegis: Top 10 of the last 24 hours (newest 200 scanned), sent around 7:00.
What it cannot see: Importance. A neutral-sounding minor item can outrank a vital but forcefully worded one. Unscored (non-English) stories are never ranked.

Where it shows up

Feed order

What it measures: The order of stories on the home page.
Setting on Huntaegis: Newest first. No score sorts or hides anything.

Where it shows up

Social posts

What it measures: Which published stories are posted to social accounts.
Setting on Huntaegis: Newest unposted public story first, after its counter-analysis is ready (or after a wait). No post-rate limits are set on this site.
What it cannot see: Importance: posting follows publication order, not merit.

Where it shows up

Narration

What it measures: Which stories get an audio version.
Setting on Huntaegis: Newest first among public English stories that have analysis and at least 1,500 characters of source text; one at a time.
What it cannot see: Importance: narration follows publication order.

Filter

Comment moderation

What it measures: Whether a reader's comment is too hostile to publish, using the VADER score of the comment text.
Range: -1 to +1
Setting on Huntaegis: Rejected below -0.7; between -0.3 and -0.7 is allowed and logged.
What it cannot see: Whether a comment is true, civil or on topic. Strong disagreement in mild words passes; a blunt true statement in harsh words may not. This is the one place a score blocks content.

Filter

How long stories stay

What it measures: Age at which a story is removed from the site.
Range: Hours
Setting on Huntaegis: 720 hours after publication.
What it cannot see: Whether the story is still relevant.

7. What we do not measure

Truth or fact-checking: we do not run an automated fact-check on any story.
Source reliability: sources are not scored or ranked for trustworthiness.
Popularity: clicks, shares and reader reactions do not influence which story is chosen.
Political lean: no score tries to place a story on a left-right scale.

Settings read from the live site configuration on 2026-10-01

Return to About