Skip to main content
After a forensic analysis completes, Postcard displays a report broken down into a headline score, a verdict, four subscores, a primary sources list, and two audit logs.

The Postcard Score

The Postcard Score is a single number from 0–100% that measures how much a post’s content has drifted from the ground truth. Higher is more credible. The score is a weighted combination of four subscores:
WeightSubscore
30%Origin Score
25%Corroboration Score
25%Bias Score
20%Temporal Score
Score ranges to guide interpretation:
RangeIndication
75–100%Strong corroboration from trusted sources; origin verified
50–74%Partial corroboration or minor inconsistencies
25–49%Mixed or refuting evidence found; treat with caution
0–24%Disputed, unverifiable, or access-blocked content

Verdict

The verdict is determined by the corroboration agent after analysing search results from trusted domains. There are four possible states:
VerdictMeaning
verifiedMultiple trusted sources confirm the core claims in the post
disputedEvidence is mixed, or trusted sources directly contradict the claims
inconclusiveSome related sources were found but are insufficient to confirm or deny
insufficient_dataNo meaningful sources were found, or the platform blocked access

Subscore breakdown

Origin Score

Measures whether the post is reachable at its claimed origin URL and whether that URL matches the expected platform. A score of 1.0 means the origin check passed cleanly; 0.5 indicates partial or ambiguous results.

Corroboration Score

The confidence score returned by the corroboration agent (0–1). It reflects how strongly the search results from trusted domains support or refute the post’s claims. A high corroboration score means the agent found multiple credible sources that align with the content.

Bias Score

Calculated from the ratio of supporting to total primary sources found:
biasScore = supportingSources / totalSources
If no sources were found, the bias score defaults to 0.5 (neutral). A low bias score indicates that most sources found were refuting rather than supporting the claims.

Temporal Score

Measures whether the post’s timestamp is consistent with what the auditor found in search results. A high temporal score means the reported date aligns with when related content appeared in the public record.

Primary sources

The Primary sources section lists the individual articles and pages found by the corroboration agent. Each source is classified as:
  • Supporting — the source confirms or aligns with the post’s claims
  • Refuting — the source directly contradicts or debunks the claims
  • Neutral — the source is related but neither confirms nor denies
Sources are drawn from a curated list of trusted domains including major news outlets (Reuters, AP, NYT, BBC, Guardian) and dedicated fact-checkers (Snopes, PolitiFact, FactCheck.org, FullFact).

Corroboration log

The Corroboration log records every step the corroboration agent took: the search queries it executed, how many results each query returned, and the final verdict decision. Use this log to understand why the agent reached its conclusion and which queries produced results.

Audit log

The Audit log records the results of the origin and temporal checks performed by the forensic auditor. It shows whether the source URL was reachable, how the platform signal was interpreted, and the raw text response from the auditing agent.

Sharing a report

Every completed report has a stable shareable link:
https://postcard.fartlabs.org/postcards?url=<encoded-url>
Postcard generates a dynamic Open Graph image for each completed report, so the link unfurls with a forensic summary card — including the Postcard Score, verdict, and post details — when shared on Discord, X, or any OG-aware platform.

Insufficient data

A report shows insufficient_data when the pipeline could not retrieve enough content to run an analysis. Common causes:
  • The platform requires a login to view the post (Instagram, some X content)
  • A Cloudflare or bot-protection page was returned instead of post content
  • The scraped content was fewer than 50 characters
  • The platform URL was not recognised
In these cases the Postcard Score is 0% and no API credits are consumed for corroboration or auditing.

Build docs developers (and LLMs) love