# How neutral is local news in Central Appalachia? A panel of three AI models, from three different companies, reads articles from Central Appalachia's local news outlets. Every outlet name and byline is stripped out first. The panel scores each article against a published rubric, and the results ship here. Published by Communities News, which is building Central Appalachia News, a newsroom for Central Appalachia that has not launched or published anything yet. That is why the method is public, the outlets are anonymous to the judges, and every outlet, including a future one of ours, is measured the same way. This is an experiment, labeled that way, with sample sizes shown next to every score. **As of 2026-10-31, 4 of the 4 Central Appalachia outlets on the board, not counting the control, show a near-zero net lean on party politics.** To read the scores: lean runs from -3 to +3 and shows direction only. Quality runs from 0 to 4, and higher is better. Every score sits beside the number of articles behind it. ![How to read a score. Lean. -3. 0. +3. Direction only. Quality. 0 to 4. Higher is better. Never add them together.](/_next/image?url=%2Fimages%2Fgraphics%2Fscorecard-how-to-read-1bbd9602.png&w=3840&q=75) **Latest run: 2026-10-31.** Rubric v0.4.0. Panel: anthropic/claude-sonnet-5, openai/gpt-5.6-terra, google/gemini-3.8-flash. Jump to: [the landscape](https://communities.news/communities/central-appalachia-news/neutrality#field-map) · [the board](https://communities.news/communities/central-appalachia-news/neutrality#roster) · [side by side](https://communities.news/communities/central-appalachia-news/neutrality#comparison) · [fingerprints](https://communities.news/communities/central-appalachia-news/neutrality#fingerprints) · [same story, two tellings](https://communities.news/communities/central-appalachia-news/neutrality#head-to-head) · [access](https://communities.news/communities/central-appalachia-news/neutrality#openness) · [how we measure](https://communities.news/communities/central-appalachia-news/neutrality#methodology) · [plain-English terms](https://communities.news/communities/central-appalachia-news/neutrality#terms) ## The Central Appalachia media landscape, mapped Two measurements separate these outlets more than any others: which worldview their framing favors, and whether official claims get tested or passed along. The map plots one against the other, with each outlet labeled in place. Politically clean but deferent sits bottom-center. Tilted but independent sits upper-left, and advocacy on the far right. The balanced middle is the middle. Positions are this page's own measured numbers and move with the daily data. \[Chart: Every tracked outlet positioned by ideological lean, source independence, and publishing volume, each bubble labeled with its outlet name\] **How to read this:** left and right is ideological tilt. Up is source independence, meaning official claims are tested rather than passed along. Bubble size is local articles per day: publishing volume, not audience reach. Dashed outlines mark the known-lean control and the out-of-market reference. The moose marks our own newsroom. Hover or tap any bubble to spotlight it and see its exact numbers. Tap again or click away to release it. ## The board, and why each outlet is on it Every outlet is on the board for a published reason. Membership is an editorial decision, never a traffic formula. No outlet leaves by quiet deletion: a change to this table is a change someone has to explain. Reach context is a note, never a gate. Each entry is a fact the outlet published about itself, or a figure a measuring tool produced, with its source and date. Search interest, domain rank and circulation are three different measures, and they disagree in useful ways. __Every outlet on the board, its role, and the recorded reason it is here.__ | Outlet | Role | Why it is on the board | Reach context (sourced) | Scored (rubric v0.4.0) | | -------------------------------------------------------------------------------------------------------------------------- | ---- | ------------------------------------------------------- | --------------------------------------------------------------- | ---------------------- | | [Charleston Gazette-Mail](https://communities.news/communities/central-appalachia-news/neutrality/charleston-gazette-mail) | peer | West Virginia's largest daily newspaper, in Charleston. | The most-visited site on this board; Tranco \~54,200 (2026-10). | 10 | | The Herald-Dispatch | peer | Huntington's daily newspaper. | Tranco \~89,700 (2026-10). | first run pending | | [Cardinal News](https://communities.news/communities/central-appalachia-news/neutrality/cardinal-news) | peer | Southwest and Southside Virginia's nonprofit newsroom. | Tranco \~159,000 (2026-10). | 9 | | [West Virginia Public Broadcasting](https://communities.news/communities/central-appalachia-news/neutrality/wvpb) | peer | The state's public broadcaster. | Tranco \~140,000 (2026-10). | 10 | | [West Virginia Watch](https://communities.news/communities/central-appalachia-news/neutrality/west-virginia-watch) | peer | The statehouse nonprofit (States Newsroom). | Tranco \~328,000 (2026-10). | 5 | ## Every outlet's fingerprint, drawn The same lean data as a picture: one panel per outlet, every panel on the same -3 to +3 scale. \[Chart: Charleston Gazette-Mail: lean fingerprint across five axes\] \[Chart: Cardinal News: lean fingerprint across five axes\] \[Chart: West Virginia Public Broadcasting: lean fingerprint across five axes\] \[Chart: West Virginia Watch: lean fingerprint across five axes\] _Reading the marks: the solid marker is the net lean. The pale band spans the average per-article swing in both directions, which is strength. A marker in the middle of a wide band means the outlet's articles lean both ways and cancel. A marker riding the band's edge means a steady house tendency. A hollow marker means the row rests on fewer than 5 articles. Direction is position, never color. Hover any row for exact numbers and sample sizes. On a computer, hovering a row also shows them in the panel's corner._ ## Central Appalachia's top outlets, side by side The largest outlets, one column each. The lean rows come first, from -3 to +3, and show direction only. Each shows two numbers: net is the average direction, and strength is how far articles lean, whichever way. The quality rows follow, from 0 to 4, where higher is better. [How we measure](https://communities.news/communities/central-appalachia-news/neutrality#lean-axes) explains every row. __Panel-median scores by outlet, Central Appalachia Media Bias Scorecard. Data last updated 2026-10-31.__ | Dimension | [Charleston Gazette-Mail](https://communities.news/communities/central-appalachia-news/neutrality/charleston-gazette-mail) | The Herald-Dispatch | [West Virginia Public Broadcasting](https://communities.news/communities/central-appalachia-news/neutrality/wvpb) | | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------- | ------------------- | ----------------------------------------------------------------------------------------------------------------- | | [Partisan lean (-3 Democratic to +3 Republican)](https://communities.news/communities/central-appalachia-news/neutrality#term-partisan "Party favoritism. Negative = Democratic-coded framing, positive = Republican-coded, 0 = none detected. Requires actual party-relevant treatment, not vibes.") | 0.1 (near-zero net lean) | \- | 0.3 (near-zero net lean) | | Partisan lean strength (0-3, direction ignored) | 0.1 | \- | 0.3 | | Articles with no partisan lean detected | 9 of 10 | \- | 7 of 10 | | [Ideological tilt (-3 progressive to +3 conservative)](https://communities.news/communities/central-appalachia-news/neutrality#term-ideological "Worldview rather than party: whose assumptions and policy logic the framing favors. Negative = progressive, positive = conservative.") | \-0.2 net, 0.5 strength (n=6) | \- | 0.2 net, 0.4 strength (n=9) | | [Institutional posture (-3 deference to +3 skepticism)](https://communities.news/communities/central-appalachia-news/neutrality#term-institutional "Whether official accounts are passed along or tested. NEGATIVE = deference (official accounts carried as fact), POSITIVE = skepticism (institutions treated as interested parties).") | \-1 net, 1.2 strength (n=9) | \- | \-1.7 net, 1.7 strength (n=10) | | [Establishment-populist framing (-3 establishment to +3 populist)](https://communities.news/communities/central-appalachia-news/neutrality#term-populist "Whose voice is the default lens on power. Negative = establishment (expert, official and incumbent views), positive = populist (outsider and ordinary-person grievance). It is separate from left and right, because populism comes in both flavors.") | \-0.5 net, 0.5 strength (n=8) | \- | \-1.1 net, 1.1 strength (n=9) | | [Conservation-development framing (-3 conservation to +3 development)](https://communities.news/communities/central-appalachia-news/neutrality#term-landuse "On resource and land stories, whose costs get counted. Negative = conservation framing favored, positive = development framing. It is scored only where the topic applies, which is why its n is smaller.") | 0.3 net, 1 strength (n=3, too few to read) | \- | 0.6 net, 1 strength (n=5) | | [Evidence fairness (0-4)](https://communities.news/communities/central-appalachia-news/neutrality#term-evidence "0 to 4. Whether each claim gets scrutiny in proportion to its evidence. There is no manufactured balance for unsupported positions, and no burying of supported ones.") | 3 | \- | 2.2 | | [Sourcing (0-4)](https://communities.news/communities/central-appalachia-news/neutrality#term-sourcing "0 to 4. Whether the story shows where its facts come from: named sources, linked records, checked numbers, and the affected people heard. A one-sided account of a contested matter caps at 2.") | 2.2 | \- | 1.8 | | [Source independence (0-4)](https://communities.news/communities/central-appalachia-news/neutrality#term-independence "0 to 4. Whether interested parties' claims are tested rather than passed along. An article can sound perfectly neutral while adopting an official's framing whole, and that scores low here.") | 1.4 | \- | 0.9 | | [Analysis labeling (0-4)](https://communities.news/communities/central-appalachia-news/neutrality#term-labeling "0 to 4. Whether interpretation is marked as interpretation. News, analysis and opinion should never blur inside news copy.") | 3.9 | \- | 3.4 | | [Language neutrality (0-4)](https://communities.news/communities/central-appalachia-news/neutrality#term-language "0 to 4. Whether the wording stays out of the way: no loaded adjectives, partisan cues or smuggled judgment.") | 3.9 | \- | 3.4 | | Articles sampled | 10 | \- | 10 | _Small samples, honestly labeled: these are early numbers, not verdicts. The sample grows daily, and every score above will move as it does._ ## How often a lean is detected at all This table asks the simplest question: in what share of scored articles did the panel find any lean at all? Averages can hide real differences. A mild lean in a third of articles and no lean in any article can produce similar averages. So this table drops the size of the lean entirely. It is the single view of the data that separates outlets most. __Share of articles with any detected lean, by axis, 2026-10-31.__ | Axis | Charleston Gazette-Mail | The Herald-Dispatch | West Virginia Public Broadcasting | | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | ----------------------- | ------------------- | --------------------------------- | | [Partisan lean](https://communities.news/communities/central-appalachia-news/neutrality#term-partisan "Party favoritism. Negative = Democratic-coded framing, positive = Republican-coded, 0 = none detected. Requires actual party-relevant treatment, not vibes.") | 10% of 10 | \- | 30% of 10 | | [Ideological tilt](https://communities.news/communities/central-appalachia-news/neutrality#term-ideological "Worldview rather than party: whose assumptions and policy logic the framing favors. Negative = progressive, positive = conservative.") | 50% of 6 | \- | 44% of 9 | | [Institutional posture](https://communities.news/communities/central-appalachia-news/neutrality#term-institutional "Whether official accounts are passed along or tested. NEGATIVE = deference (official accounts carried as fact), POSITIVE = skepticism (institutions treated as interested parties).") | 89% of 9 | \- | 100% of 10 | | [Establishment-populist framing](https://communities.news/communities/central-appalachia-news/neutrality#term-populist "Whose voice is the default lens on power. Negative = establishment (expert, official and incumbent views), positive = populist (outsider and ordinary-person grievance). It is separate from left and right, because populism comes in both flavors.") | 50% of 8 | \- | 100% of 9 | | [Conservation-development framing](https://communities.news/communities/central-appalachia-news/neutrality#term-landuse "On resource and land stories, whose costs get counted. Negative = conservation framing favored, positive = development framing. It is scored only where the topic applies, which is why its n is smaller.") | 100% of 3 | \- | 80% of 5 | ## Same story, side by side Sometimes two tracked outlets cover the same story, and we hold both texts. Then the panel also ranks the two blinded accounts against each other. This is the most sensitive comparison we have, because differences that average scores smooth over show up when the event is identical. An adjudicator matches story pairs and records a one-sentence reason for every match. There are 3 same-story comparisons so far. A win counts only when a majority of the three judges agree. __Head-to-head wins by category, 2026-10-31. Appearances: West Virginia Watch 2, West Virginia Public Broadcasting 3, Charleston Gazette-Mail 1.__ | Category | Wins | | ------------------------------------ | ------------------------------------------------------------------------------------------ | | fairest evidence treatment | West Virginia Watch: 1 · Charleston Gazette-Mail: 1 · West Virginia Public Broadcasting: 1 | | best sourced | West Virginia Watch: 2 · Charleston Gazette-Mail: 1 | | most independent of official framing | West Virginia Watch: 2 · Charleston Gazette-Mail: 1 | | most editorialized | West Virginia Watch: 2 · Charleston Gazette-Mail: 1 | _Small numbers, honestly labeled. Head-to-head pairs exist only when outlets cover the same story in the same window, which is rarer than it sounds. The counts grow daily._ ## Access and distribution practices These are facts about each outlet's publishing choices, and anyone with a terminal can check them. They are not neutrality or quality scores. Blocking AI crawlers is a legitimate choice, not a journalistic failing. Full-text RSS means the outlet's journalism arrives without a site visit. The AI column counts the crawler user-agents that robots.txt turns away entirely. __Openness facts by outlet, checked 2026-10-31: paywall, RSS, and crawler policy.__ | Outlet | Paywall | RSS | Full-text RSS | AI crawlers blocked | Readable without JS | | -------------------------------------------------------------------------------------------------------------------------- | ------- | ------------------------------------------------ | ------------- | ------------------- | -------------------------------------------- | | [Charleston Gazette-Mail](https://communities.news/communities/central-appalachia-news/neutrality/charleston-gazette-mail) | metered | unreachable from our collection point (HTTP 429) | no | 12 | yes | | The Herald-Dispatch | metered | unreachable from our collection point (HTTP 429) | no | 12 | not checked (the site rate-limited us today) | | [Cardinal News](https://communities.news/communities/central-appalachia-news/neutrality/cardinal-news) | none | yes | yes | 0 | yes | | [West Virginia Public Broadcasting](https://communities.news/communities/central-appalachia-news/neutrality/wvpb) | none | yes | yes | 0 | yes | | [West Virginia Watch](https://communities.news/communities/central-appalachia-news/neutrality/west-virginia-watch) | none | yes | yes | 11 | not checked (our crawler got HTTP 403) | ## How we measure **How the numbers are computed.** Each judge scores each article once. The per-article score is the median across the three judges. An outlet's number is the mean of its per-article scores, shown to one decimal. Per-article scores and judge disagreement are in the raw data download. One boundary, stated plainly: the panel assesses the text, not the world. It scores whether claims are attributed and whether the article shows its checking. It does not verify facts, and it produces no factuality rating. A model reading one article about a brand-new local event has nothing independent to check it against. Pretending otherwise would be the exact overreach this page exists to avoid. ### The five lean axes and the quality scores Since rubric v0.3, lean is measured on five separate axes, because partisanship and ideological tilt are different things. An outlet can favor no party while still favoring one worldview's institutions, sources and framings. Alongside party favoritism, the panel scores ideological tilt, deference or skepticism toward institutions, conservation or development framing, and establishment or populist framing. Each is scored per article, in both directions. Every axis reports two numbers. Net is the average, which can hide opposite leans that cancel. Strength is the average size of the lean per article, whichever way it points, which cannot hide them. Read them together. A net near zero with high strength means the outlet swings both ways on that axis, story by story. Net and strength nearly equal means a steady house tendency. Partisan lean also reports how many articles show no lean at all. Lean is direction only, not quality. A lean-free article can still be badly sourced, and that is what the quality dimensions measure, from 0 to 4, where higher is better. Evidence fairness rewards scrutiny in proportion to the evidence, not equal time. Source independence asks whether official framing was tested or passed along. Judges may mark a dimension not applicable where it has no meaning: a weather brief is not penalized for lacking opposing sources. The judged sample covers each outlet's own local journalism: news, politics, business and opinion, sorted by the same rule for every outlet. National wire stories, sports and syndicated lifestyle columns are left out. So an outlet's numbers describe its journalism, not its content mix. ### The panel and the blinding **Partial blinding.** Before any model sees an article, a rule-based pass with no AI in it strips outlet names, bylines, self-references and boilerplate. A guard refuses to score any article where a known identity string survives. This removes explicit identity, not every clue. House style, subject matter or distinctive passages could still hint at the source. So we call it partial blinding, rather than claiming the judges cannot know. **The panel.** Three models from three vendors (anthropic/claude-sonnet-5, openai/gpt-5.6-terra, google/gemini-3.8-flash), each scoring on its own at temperature 0\. The headline number is the panel median. Disagreement between judges is data we keep, not noise we hide. No single human rater sits above the panel. One person's read is one set of priors nobody can audit, while the panel's disagreement is published. ### Validation checks and limitations **Can our method detect lean at all?** A judge that scores everything "center" is broken. This board has no known-lean control outlet of its own yet. The same panel and rubric read a control outlet daily on the [Alaska board](https://communities.news/communities/alaska-news/neutrality). Choosing one for Central Appalachia is an editorial decision still to be made. ### How the rubric changed, and why - **v0.3, 2026-09-20: lean split into five axes.** Story-by-story human audits found that a single left-right axis under-reads outlets whose house style is institutionally neutral. Their language carries no partisan tells, while the framing still favors particular institutions and worldviews. So v0.3 separates party favoritism from ideological tilt and scores the worldview axes directly. - **v0.3, 2026-09-20: local journalism only.** The judged sample narrowed to each outlet's local journalism. A feed that is one-third syndicated advice columns and national wire pulls every score toward neutral. - **v0.4, 2026-09-29: headlines judged too.** Feed-collected outlets had been scored on article bodies alone, while page-fetched ones included their headlines. The gap surfaced when one outlet's partisan headlines sat atop more restrained bodies. - **v0.4, 2026-09-29: a defined scale.** The lean axes gained a ladder defining what separates a 1 from a 2 from a 3\. The v0.3 scores showed judges almost never used the ends of the scale without a definition. Each change came from a measured gap. The axes were added to measure tilt the old rubric could not see, not tuned toward any expected result. Every outlet is scored by the same prompt, still blinded. Scores from different rubric versions are never mixed, so each version's numbers build fresh from its start date. Every version's final numbers stay published below: a version change restarts the count, never the history. **Calibration baseline, and where it came from.** The panel was first run against a frozen 14\-article reference set. Its provenance matters, so here it is in full. The reference audit was produced by an AI assistant and reviewed by a human in conversation. The rubric's scoring anchors were derived from that same audit. So agreement with the seed set shows the judges reproduce a documented read, not independently established truth. Every article scored since then is one the rubric has never seen, which is the ongoing check that matters. Each seat's average miss on bias against the seed set, its mean absolute error over 2 runs per seat: anthropic 0.2, openai 0.1, google 0.1. Any change of models or rubric re-scores this set, and the break is published. **Collection ethics.** Our crawler identifies itself honestly and respects robots.txt. It never gets around a paywall, and we republish no competitor text. Scores describe the text, never anyone's intent. **Conflicts, stated plainly.** Published by Communities News, which is building Central Appalachia News, a newsroom for Central Appalachia that has not launched or published anything yet. We operate this page. The safeguards are the blinding, the cross-vendor panel and a public method. One panel seat runs a model from Anthropic, whose models also draft our newsrooms' articles. The other two seats exist so that seat never rules alone. **Raw data.** Every number on this page comes from [one downloadable JSON](https://communities.news/data/central-appalachia-media-scorecard.json), with aggregates and sampling frames but no article text. It was last updated 2026-10-31, and the page and the download update together. The data and the analysis on this page are licensed [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/): reuse them freely, with credit to the Central Appalachia Media Bias Scorecard by Communities News. [Back to the Central Appalachia News profile](https://communities.news/communities/central-appalachia-news) · [Central Appalachia News](https://centralappalachia.news) ## The terms, in plain English Every dotted-underlined term on this page lands here. One rule covers most of it: the lean axes measure **direction**, and the quality scores measure **craft**. Never add them together in your head. A story can point nowhere and still be built badly, or point somewhere and be built well. On every lean axis the signs are written out below. A signed number you have to guess the direction of is a number we have not really published. Panel The three AI models that score every article. They come from Anthropic, OpenAI and Google, and each scores on its own, so no single vendor's model decides. Blinded The judges do not see who published the article. Outlet names, bylines and self-references are stripped first, so the panel judges the text, not a reputation. The blinding is partial, because style or subject can still hint at the source. Rubric The published scoring rules. When the rubric changes, scores from different versions are never mixed. Each version's final numbers stay published in the lineage table. Sample size (n) How many articles stand behind a number. A bigger n means more confidence. A small n is an early read, not a verdict. Net The average direction across articles. Articles leaning opposite ways cancel out, so a net near zero can mean level or can mean both ways. Strength tells you which. Strength How far articles lean on average, whichever way they lean. High strength with a near-zero net means the outlet swings both ways. Low strength means its articles are level one by one. Partisan lean Party favoritism. Negative = Democratic-coded framing, positive = Republican-coded, 0 = none detected. Requires actual party-relevant treatment, not vibes. Ideological tilt Worldview rather than party: whose assumptions and policy logic the framing favors. Negative = progressive, positive = conservative. Institutional posture Whether official accounts are passed along or tested. NEGATIVE = deference (official accounts carried as fact), POSITIVE = skepticism (institutions treated as interested parties). Conservation-development framing On resource and land stories, whose costs get counted. Negative = conservation framing favored, positive = development framing. It is scored only where the topic applies, which is why its n is smaller. Establishment-populist framing Whose voice is the default lens on power. Negative = establishment (expert, official and incumbent views), positive = populist (outsider and ordinary-person grievance). It is separate from left and right, because populism comes in both flavors. Evidence fairness 0 to 4\. Whether each claim gets scrutiny in proportion to its evidence. There is no manufactured balance for unsupported positions, and no burying of supported ones. Sourcing 0 to 4\. Whether the story shows where its facts come from: named sources, linked records, checked numbers, and the affected people heard. A one-sided account of a contested matter caps at 2. Source independence 0 to 4\. Whether interested parties' claims are tested rather than passed along. An article can sound perfectly neutral while adopting an official's framing whole, and that scores low here. Analysis labeling 0 to 4\. Whether interpretation is marked as interpretation. News, analysis and opinion should never blur inside news copy. Language neutrality 0 to 4\. Whether the wording stays out of the way: no loaded adjectives, partisan cues or smuggled judgment. Local An outlet's own journalism: its news, politics, business and opinion sections, sorted by the same rule for every outlet. Wire stories, sports and syndicated lifestyle columns are left out. Local means the outlet's own reporting, not stories set in Central Appalachia, which the coverage map counts separately. Near-zero net lean The label for a net lean closer to zero than 0.5\. From 0.5 up to 1.5 the label is "leans", and 1.5 or more names the side outright. These are cutoffs for labeling, not statistical tests, so read the number beside the label: a -0.4 and a -0.5 differ by a tenth.