Skip to main content
The Postcard Score is a weighted percentage (0–100%) that measures how much a piece of content has drifted from its ground truth. A higher score means the content is well-verified; a lower score signals unreliability or active disputes. The score is computed by combining four forensic subscores, each targeting a distinct axis of credibility, then multiplying by 100 and flooring to an integer for storage.

​
The four subscores

Origin Score (weight: 30%) — URL reachability. Can the original post be accessed at its claimed source? The verifier agent checks whether the URL is reachable and whether platform-specific signals (hostname, structure) match the declared platform. A clean match scores 1.0; a mismatch defaults to 0.5.
Corroboration Score (weight: 25%) — AI confidence. This is the confidenceScore returned by the Gemini-powered corroboration agent after searching trusted domains for independent sources that support or refute the claim. Ranges from 0.0 (no evidence) to 1.0 (strong independent confirmation).
Bias Score (weight: 25%) — Source ratio. Calculated as the ratio of supporting sources to total sources found: supportingSources / totalSources. When no sources are found, the score defaults to 0.5 (neutral) to avoid penalizing posts that are simply too niche to index.
Temporal Score (weight: 20%) — Timestamp alignment. Does the post’s publication date match the narrative it belongs to? The verifier agent checks timestamp coherence against search results. The current implementation defaults to 0.8 when no strong temporal contradiction is detected.

​
Weighted formula

The exact formula from src/lib/postcard.ts: 
const rawScore =
  audit.originScore * SCORING_WEIGHTS.ORIGIN +
  corroborationScore * SCORING_WEIGHTS.CORROBORATION +
  biasScore * SCORING_WEIGHTS.BIAS +
  audit.temporalScore * SCORING_WEIGHTS.TEMPORAL;

const postcardScore = Math.floor(rawScore * 100);
Where SCORING_WEIGHTS is defined in src/lib/config.ts: 
export const SCORING_WEIGHTS = {
  ORIGIN: parseFloat(process.env.POSTCARD_WEIGHT_ORIGIN || "0.3"),
  CORROBORATION: parseFloat(process.env.POSTCARD_WEIGHT_CORROBORATION || "0.25"),
  BIAS: parseFloat(process.env.POSTCARD_WEIGHT_BIAS || "0.25"),
  TEMPORAL: parseFloat(process.env.POSTCARD_WEIGHT_TEMPORAL || "0.2"),
};
The four weights must sum to exactly 1.0. If you customize them via environment variables, verify they add up: ORIGIN + CORROBORATION + BIAS + TEMPORAL = 1.0. Misconfigured weights will silently produce scores outside the expected range.

​
Score interpretation

RangeVerdictMeaning
80–100verifiedMultiple trusted sources confirm the core claims. Strong forensic signal.
50–79inconclusiveSome related sources found, but insufficient evidence to confirm or deny.
1–49disputedEvidence is mixed, or trusted sources actively contradict the claims.
0insufficient_dataNo meaningful sources found. Pipeline may have been blocked by a login wall or rate limit.

​
Verdict states

The verdict is determined independently by the corroboration agent and stored alongside the numeric score. The four possible values are:
  • verified — Multiple trusted sources confirm the core claims.
  • disputed — Evidence is mixed, or trusted sources contradict the claims.
  • inconclusive — Insufficient evidence to determine truthfulness, but some related sources were found.
  • insufficient_data — No meaningful sources found to evaluate the content.

​
How the score is stored

The database stores postcardScore as a real column (0–100 integer range) in the postcards table: 
// src/db/schema.ts
postcardScore: real("postcard_score").notNull(),
The API response divides it back by 100 to return a 0.0–1.0 float: 
// src/lib/postcard.ts
postcardScore: finalRow.postcardScore / 100,
So a stored value of 85 becomes 0.85 in the API response.

​
Configuring weights

Override the default weights with environment variables. All four must be set together if you change any of them.
VariableDefaultDescription
POSTCARD_WEIGHT_ORIGIN0.3Weight for URL reachability score
POSTCARD_WEIGHT_CORROBORATION0.25Weight for AI corroboration confidence
POSTCARD_WEIGHT_BIAS0.25Weight for supporting/total source ratio
POSTCARD_WEIGHT_TEMPORAL0.2Weight for timestamp alignment score
POSTCARD_WEIGHT_ORIGIN=0.4
POSTCARD_WEIGHT_CORROBORATION=0.2
POSTCARD_WEIGHT_BIAS=0.2
POSTCARD_WEIGHT_TEMPORAL=0.2

Build docs developers (and LLMs) love

Get started for freeTalk to us