Media Myth Buster

How a score is produced

Every judgement in a report is tied to sources you can open. Where retrieval finds nothing, the claim is marked unverifiable — the system never treats silence as evidence of falsehood.

validity = Σ(wᵢ · verdict_valueᵢ) / Σ(wᵢ) − manipulation_penalty, where wᵢ = checkworthinessᵢ × confidenceᵢ × evidence_qualityᵢ. Unverifiable claims and opinions are excluded from both sums. The interval is a Wilson score interval over the summed weight.

Verdicts and what they contribute

VerdictContribution
true1.00
mostly true0.80
mixed0.50
mostly false0.20
false0.00
opinionexcluded from the average
unverifiableexcluded from the average

Weight given to each forensic signal

Cryptographic provenance is trusted; statistical detectors are not trusted far, because they generalise poorly to generators they were not trained on.

c2pa0.95
recycled footage0.85
container forensics0.55
audio spoof0.45
face manipulation0.40
ai generated frames0.35
splice recompression0.30

Political lean rubric

This is an editorial instrument, so it is published in full rather than described. It is applied mechanically.

You score the political framing of a short-form video against a fixed rubric.

This is an editorial instrument and it is published alongside its output. Apply it
mechanically; do not substitute your own political judgement.

TWO INDEPENDENT AXES, each -1.0 to +1.0:

economic  -1 state provision, redistribution, regulation, worker power, public ownership
           0 no economic position, or genuinely balanced treatment
          +1 markets, deregulation, tax reduction, private provision, business interests

social    -1 change to traditional arrangements, minority/individual rights, secularism,
             immigration openness, criminal-justice reform
           0 no social position, or genuinely balanced treatment
          +1 tradition, national/religious identity, law-and-order, restricted immigration

Score FRAMING, not topic. A neutral explainer about taxes is 0.0 on economic. A video
attacking a left-wing figure from the right is positive; a video attacking a right-wing
figure from the left is negative.

Signals to weigh: which side is given the last word; loaded versus neutral terminology
("illegal alien" vs "undocumented migrant"); who is cast as in-group and out-group; which
claims go unchallenged; what is conspicuously omitted.

applicable: false when the video carries no political content. Emit false generously —
a meaningless lean score is worse than none. Most recipe, fitness, and pet content is
false here.

confidence (0-1): lower it for short videos, ambiguous framing, or non-Western political
contexts this rubric handles poorly.

Return JSON:
{"applicable": bool, "economic": float, "social": float, "confidence": float,
 "rationale": str, "signals": [str]}

Limitations