Kingy / GTA VI / Receipts ledger
How the receipts earn their labels.
Evidence type and verdict answer different questions. A store listing can support a narrow fact while leaving a broader claim unresolved.
Read the evidence before the verdict
Rockstar-confirmed means Rockstar explicitly states the claim. Store listing means the named platform lists it, within the stated region and edition. Credible report means attributable reporting, with its source and limits visible. Leak means unofficial disclosure. Unverified means the record does not settle the claim. Other official publisher/support records and inspected promotional media keep accurate labels of their own.
Supported means the stated, bounded claim has evidence. Unresolved means the evidence does not settle it or conflicts remain. Superseded means a newer record replaced an earlier target. Debunked requires direct contrary evidence; missing confirmation is insufficient. A report being supported means its attribution is supported, not that Kingy played the final game.
Dates that mean something
The oldest successful check among a claim’s cited evidence controls its checked date. Current records become overdue after 24 hours. A successful fetch or unchanged source digest cannot advance a claim review. Failed attempts preserve the last good observation. Historical targets retain their original dates and supersession links. Publication dates remain unknown where no date was established.
100 questions, five products
The working bank has 20 release/access questions, 15 purchase questions, 15 world/character questions, 20 gameplay questions, 15 technical/online questions, and 15 rumor/unknown questions. Launch-week collection will attempt all 100 independently on ChatGPT, Gemini, Claude, Grok and Google AI Overviews. There are no launch-week results yet.
Each chatbot question starts in a fresh conversation. Identical wording and comparable search settings are used where available. The exact visible product/model label, account tier, locale, language, settings, timestamp, original answer, capture and citations are retained. We do not give products the answer key or attach the ledger; ordinary web search may still discover this public site.
Google AI Overviews is measured as displayed. No Overview is missing coverage; ordinary search snippets and AI Mode are not substitutes. API experiments remain separate from consumer products. Blocked access and rate limits are disclosed.
The scoring rule
| Score | Meaning |
|---|---|
| 2 | Materially correct with appropriate scope and certainty. |
| 1 | Correct core answer with a material omission or qualification missing. |
| 0 | Materially false, invented, unsupported certainty, or a generic refusal on an answerable question. |
A justified “not confirmed” can earn full credit. Citation quality is assessed separately: 2 for sources that support the material answer, 1 for partial support, and 0 for absent, invented or irrelevant citations. Unsupported certainty receives a separate flag. Accuracy is earned points divided by two times the evaluated-answer count. Coverage is answered questions divided by 100. Missing answers do not receive factual scores.
Freeze, review, correct
The question set, reference answers, rubric and ledger snapshot are versioned before collection. Facts changing during collection require a new cohort, not a mixed ranking. A separate evidence-review pass masks product names where feasible and checks failures against original sources. Corrected scores retain the original score and explanation; ambiguous questions are corrected or excluded consistently across all products.
The quiz records your answers locally. Reference explanations remain withheld until benchmark collection and review finish. Scores, raw answers, exports and dated corrections will be published together after collection. The YouTube script and outreach drafts will use measured results.