Build a reusable Media-Claim Verification Correction & Ed…News & Media Publishers
Replay
Every recorded step of a run, as the agent saw the screen: the frame, the action it took, what it said, and which rubrics the judge's evidence had reached by then. 4 runs recorded with screenshots.
Claude Opus 5.5
- 100%
- Rubric score
- 29
- Steps
- 15/15
- Rubrics passed
Task promptBuild a reusable Media-Claim Verification, Correction & Editorial Traceability System that helps an editorial team evaluate factual claims before publication and manage them responsibly after publication when new evidence appears. The objective is not to decide which publisher is “best” or to rank news organizations. Instead, create a rigorous workflow for determining what a claim actually asserts, what evidence supports it, how strong that evidence is, whether the wording overstates what is known, how conflicting evidence should be handled, and how corrections should be documented and reviewed. Research realistically accessible authoritative sources such as professional journalism organizations, editorial standards bodies, public broadcasters, university journalism schools, academic research, official records, court or government documents where relevant, books on journalism and verification, reputable magazines, newsroom standards manuals, and publicly available publisher correction policies. Prefer primary documents for editorial standards, correction practices, source attribution, and factual records. Use academic and professional research to explain verification methods and common failure modes. Treat social posts, anonymous claims, aggregators, headlines, search snippets, and unsourced summaries as leads rather than automatically reliable evidence. The final deliverable must be a spreadsheet/tracker that records individual claims, publication context, source evidence, verification status, wording strength, contradictions, confidence, correction history, and follow-up requirements. The tracker should distinguish between verified, partially supported, unsupported, contradicted, unresolved, and context-dependent claims without forcing certainty when evidence is incomplete. Make the system useful weeks or months later, when a previously published claim is challenged, a source is corrected, a new official document becomes available, or an evolving story requires the original claim to be reassessed. Preserve the original evidence and decision history rather than overwriting previous judgments. Include 8 clearly labeled fictional stress-test scenarios involving an accurate fact presented with misleading wording, two credible sources that conflict, a secondary source misquoting a primary record, an old fact becoming misleading after circumstances change, a viral claim supported only by repeated copies, an official source containing an apparent inconsistency, a correction that fixes one sentence but leaves the surrounding article misleading, and a claim that cannot yet be verified but is tempting to publish. Include an adversarial audit, explicit Stranger Test, and final quality check so an unfamiliar reviewer can operate the tracker independently without asking follow-up questions.
Build a reusable Media-Claim Verification, Correction & Editorial Traceability System that helps an editorial team evaluate factual claims before publication and manage them responsibly after publication when new evidence appears. The objective is not to decide which publisher is “best” or to rank news organizations. Instead, create a rigorous workflow for determining what a claim actually asserts, what evidence supports it, how strong that evidence is, whether the wording overstates what is known, how conflicting evidence should be handled, and how corrections should be documented and reviewed. Research realistically accessible authoritative sources such as professional journalism organizations, editorial standards bodies, public broadcasters, university journalism schools, academic research, official records, court or government documents where relevant, books on journalism and verification, reputable magazines, newsroom standards manuals, and publicly available publisher correction policies. Prefer primary documents for editorial standards, correction practices, source attribution, and factual records. Use academic and professional research to explain verification methods and common failure modes. Treat social posts, anonymous claims, aggregators, headlines, search snippets, and unsourced summaries as leads rather than automatically reliable evidence. The final deliverable must be a spreadsheet/tracker that records individual claims, publication context, source evidence, verification status, wording strength, contradictions, confidence, correction history, and follow-up requirements. The tracker should distinguish between verified, partially supported, unsupported, contradicted, unresolved, and context-dependent claims without forcing certainty when evidence is incomplete. Make the system useful weeks or months later, when a previously published claim is challenged, a source is corrected, a new official document becomes available, or an evolving story requires the original claim to be reassessed. Preserve the original evidence and decision history rather than overwriting previous judgments. Include 8 clearly labeled fictional stress-test scenarios involving an accurate fact presented with misleading wording, two credible sources that conflict, a secondary source misquoting a primary record, an old fact becoming misleading after circumstances change, a viral claim supported only by repeated copies, an official source containing an apparent inconsistency, a correction that fixes one sentence but leaves the surrounding article misleading, and a claim that cannot yet be verified but is tempting to publish. Include an adversarial audit, explicit Stranger Test, and final quality check so an unfamiliar reviewer can operate the tracker independently without asking follow-up questions.