Social reporting, source perspective and post framing

Open Advanced / Investigate → Social. The reader now separates source political/editorial context, post reporting style and text-framing cues. They answer different questions and can disagree or remain unassessed.

Choose a useful order

• Reporting first · diversified is the new Social default. It uses available reporting-style cues, original publication age and literal link presence, then spreads adjacent publisher families and repeated links. It preserves all matching records. It is a reading order, not a truth, popularity or confidence score.
• Source-perspective mix rotates through the source-context groups actually present in the selection: left-associated, right-associated, liberty-associated, mixed, explicitly centrist, institutional and unclassified, where available. Each group keeps publisher diversity and its own reporting order. It creates a varied reading queue, not a representative political survey or a guarantee of factual balance. Unclassified sources are not counted as centrist.
• Latest first, Global / publisher balance and Severity first remain available. News workspaces retain their existing default and behavior.

The ordering selector is directly visible in Social. Mix source perspectives is also a shortcut. Existing source-leaning, source-type, platform and topic filters remain available; region/language/date controls retain their separate scope.

What affects reporting order

An original source publication within seven days receives the current-publication band. Unknown, future, update-only, referring-article-only, older and explicitly retained copies do not acquire a new publication date. Within each band, reporting style contributes a bounded weight and recency decays with a24-hour half-life. A supplied literal outbound URL adds a small link-presence weight; it does not establish the linked page's contents or reliability. Engagement, likes, followers, the reader's identity and political labels do not affect the base score.

Unsupported language receives the reporting baseline weight rather than a low-quality penalty. In ordinary reporting order, an80-record look-ahead applies publisher-family and exact-link/text repetition penalties. No record is deleted. Perspective mix separately round-robins the available source-context groups, so a small group is not hidden behind a large unclassified bucket. Current-publication records remain ahead of the older/unknown band.

Read source leaning correctly

The Source: badges and Source political / editorial leaning filter describe the publisher or organizational account. The existing catalogue and its source classifications are retained. They are not labels for every article, a journalist, an individual reader or a community member.

Under Post framing & source evidence → Source leaning evidence, distinguish:

• Cited organizational context: an assigned description has a usable source reference, such as the publisher's self-description.
• Earlier catalogue grouping: an inherited Watchtower category, with its earlier basis retained. It is not a new external bias rating.
• Unclassified / institutional: no supported political assignment, or a source whose institutional role is separately described.
• Assigned without a usable citation: a description lacks a usable citation.

Source details retains citations and review dates. A cited self-description is evidence of how an organization describes itself, not independent proof of bias or factual accuracy. National political terminology differs, and broad labels overlap.

Inspect the post itself

Open Source details → How this post was assessed. The method is social-reporting-v1, a bounded, deterministic ruleset—not a trained or calibrated political-bias model. It analyzes the supplied headline/excerpt, not a complete article, video transcript, private feed or the account's entire timeline.

Post reporting style distinguishes reporting cues, analysis/explanation, opinion/commentary, advocacy/action, institutional updates, community discussion and unassessed style. A reporting cue does not verify a reported event.

Post text-framing cues can identify redistributive/labour, market-oriented, civil-liberties/limited-state, socially progressive, socially conservative or security/restriction wording. More than one cue family is shown as multiple frames; it is not averaged into a centrist score. Each match identifies:

1. The exact source field and excerpt.
2. Its surrounding sentence and rule identifier.
3. Whether it is an explicit policy-position cue or evaluative-language cue.

For example, “We support a wealth tax” contains an explicit policy-position cue. “Government overreach” is an evaluative-language cue; it does not by itself turn the whole report into an opinion piece. A party name, country, ordinary policy topic or mention of an election is insufficient to infer a post's political frame.

Detected quotations, speech attribution, denials, questions, hypotheticals and satire/fact-check context are withheld from endorsement cues. Details can show why wording was excluded. Repeated headline/excerpt text is analyzed once, and detected channel/subscription boilerplate is removed. Referring-publisher video entries are explicitly assessed as referring-article excerpts, not as the video's content.

Coverage and abstention

The text rules currently cover limited English wording. A declared post language takes precedence over the feed language; those declarations can themselves be wrong. Other languages remain unassessed for post framing while their existing organizational source context, original text and source dates remain accessible.

The rules deliberately abstain when wording is insufficient. In the initial seven-day qualification selection, none of the6,552posts met the narrow post-framing rules; source-perspective sorting still used the existing67classified source feeds. That outcome is an abstention, not a finding that those posts were neutral or unbiased. This release chiefly improves source-perspective sorting, transparency and reporting-style inspection; it does not claim comprehensive semantic bias detection. Authored positive/negative regression cases verify mechanics, not independently adjudicated classification accuracy.

Compare, save and export

Compare shows source context and post assessments side by side. Source details explains matched/suppressed text and ordering. Reading-list saves preserve the assessment version/time; they do not silently refresh old labels. Saved views keep the chosen filters. Page exports include source metadata, the assessment and method description. Existing native exact-record Pro Brief selection remains available.

Social pages use a native snapshot tied to source content, profile revision, selection and assessment minute. If it changes between pages, Load latest first page restarts the selection rather than mixing rankings. A snapshot token is not a historical archive. Related coverage still uses bounded lexical similarity; multiple posts and cross-platform repeats are not independent corroboration.

The public catalogue remains479configured feeds plus2directory-only references. Availability and publication coverage change normally. Collection schedules, provider pauses, account-local saved-state ownership and replay controls remain native. Public queries do not fetch arbitrary URLs, infer private political profiles, train on reader behavior or trigger paid models.