TruthScore vs Ground News
Both products try to help a reader reason about media bias. We built TruthScore on five commitments that map directly to our methodology — a transparent ledger, a published 0 — 10 formula, three kinds of source disclosure, a first-class coverage-gap signal, and an open-fetch scorer behind no paywall. This page names where the two products differ on each of those five.
| Dimension | TruthScore | Ground News | Why it matters |
|---|---|---|---|
| Transparent ledger of every verdict | Full audit trail of every moderation action: actor, target, decision code. Outcomes are reproducible from the recorded change set. | Single editorial posture per outlet. The decision record is internal; the public surface is the verdict, not the reasoning. | Enables third-party scrutiny — anyone can retrace how a single score was set, and file a correction request. |
| Score and band come from a published formula | 0–10 heuristic with a baseline of 5 and five weighted signals: Length (penalty), Sensationalism (penalty), Hedging, Attribution, Structure. Landed in one of five named bands. | Bias and factuality ratings with a published methodology, but not as inspectable source: no signal weights to check, no per-signal breakdown on a tile. | Reproducibility — the same body text returns the same breakdown every time, without editorial discretion in the middle. |
| Three disclosure kinds, not one black box | Engine (daily confidence-weighted against the trending corpus), Curated (hand-set baseline with a Catalyst badge), and Hybrid (engine-first, curated fallback when no rows land for the day). | A single editorial posture per outlet, surfaced as a bias rating and a factuality rating. No separate concept of curated baseline vs. engine-derived score. | Lets users reason about confidence — a Hybrid outlet score is genuinely different from a hand-set curated one, and the tile says so. |
| Coverage-gap signal is a first-class surface | /underreported and /trending list credible stories against a 13-outlet reference set (Reuters, AP, BBC, NYT, Bloomberg, Guardian, NPR, PBS, WaPo, WSJ, FT, Economist, Hacker News) and surface stories appearing in fewer than four of them. | BLINDSPOT surfaces under-covered stories across the political spectrum. Shipped on a paid tier and not every plan exposes the underlying check. | Closing the gap between what happened and what got reported is the actual job — somewhere a reader can verify the comparison, not just trust a marketing page. |
| No paywall, no login, no LLM inference | Open fetch with a 500 KiB body cap, 3-hop redirect cap, 5 s timeout, DNS + private-IP rejection, public http(s) only, fixed TruthScoreBot/1.0 UA, no LLM calls on the scorer path. | Bias comparison and gap features are subscription-gated; the historical archive and BLINDSPOT sit behind paid plans. | Open scrutiny vs. closed editorial call — anyone can re-run the fetch, anyone can re-state the heuristic, and the verdict comes back without a login wall. |
A transparent ledger of every verdict
Every moderation action in TruthScore is recorded with its actor, target, and decision code, and every score is the deterministic output of the same public heuristic. The verdict a reader sees and the change set that produced it are both inspectable, and a third party can retrace how a single tile was set. This is the difference between “we trust this number” and “anyone can retrace how this number was set” — the second is the one we’re betting on.
Score and band come from a published formula
Each article URL runs through a deterministic heuristic that starts from a baseline of 5 and sums five weighted signals: Length, Sensationalism, Hedging, Attribution, and Structure. The result is clamped to 0 — 10, rounded to one decimal, and landed in one of five named bands. Every signal’s raw value and contribution shows up on the article page, so the breakdown is what you see.
- 0 — 2Hoaxworthy
- 3 — 4Cautious
- 5 — 7Mixed signals
- 8 — 9Largely credible
- 10God's Honest
The 5-signal heuristic and the 5-band scale are spelled out in our methodology. The same heuristic runs on individual article URLs (score a URL) and on outlet homepages (look up an outlet), so the band you see on the News Wall and the per-URL result come from the same formula.
Three disclosure kinds, not one black box
On the News Wall, every outlet tile declares how its score was set. The three kinds we publish are Engine (daily confidence-weighted against the trending RSS corpus), Curated (hand-set baseline, marked with the Catalyst badge), and Hybrid (engine score with a curated fallback when no trending rows land for the day). The disclosure kind rides on the tile, so the reader can tell a hand-set curated baseline apart from an engine recompute without reading the methodology.
Coverage gap is a first-class surface, not an upsell
The other half of TruthScore’s mandate is the coverage gap — credible stories that the reference outlets haven’t picked up. Every day we compare the stories our reference outlets cover against stories that look credible but appear in fewer than four of them, and surface the gap on /underreported. The same 13-outlet reference set — Reuters, AP, BBC, NYT, Bloomberg, Guardian, NPR, PBS, WaPo, WSJ, FT, Economist, and Hacker News — drives both /trending and the gap rail. The mechanism is open and inspectable: no paid plan gates the comparison.
No paywall, no login, no LLM on the scorer path
The scorer is just an HTTP fetch: public http(s), TruthScoreBot/1.0UA, a 5-second timeout, a 500 KiB body cap, three redirect hops max, and DNS-resolution + private-IP rejection. Pages with paywalls aren’t fetched for adjudication; the fetch cap is on what can reach the scorer, not just what it returns. There is no login requirement to score a URL or look up an outlet, no query history is stored against a user, and no LLM call sits on the scorer path — a score is the deterministic output of the heuristic, not of an inference call.
Where Ground News still wins
Acknowledging what we don’t do well is part of being trustworthy. Three honest strengths of Ground News that’ve earned their place: the historical archive is deeper because they’ve been covering media bias longer, the BLINDSPOT feature on their paid plan surfaces under-covered stories across the political spectrum, and their reader community is larger and more active than ours. If you want curated discovery or a community around media-bias discussion, the right answer for today is Ground News — that’s not a knock we need to paper over, it’s a real difference in product maturity.
See the methodology in action
Read the full methodology, or score any public article URL and look up an outlet on the News Wall.