Skip to content
SingularityOBSERVATORY
Menu

PUBLIC METHOD AND TRUST AUDIT

Claims remain traceable

This read-only record is generated from canonical claims, corrections and assessments; it does not grant source or editorial approval.

5
PERIODIC RECONSTRUCTIONLOWEST HASH PER DIRECT MODULE

Sampled claims

  • agent-horizon/claim/gpt-5-4-p50: reconstructed · lineage be89de0e78c4
  • chip-supply/claim/us-share-forecast: reconstructed · lineage 0733095a1d3f
  • compute/claim/ai-2027-superhuman-coder-context: reconstructed · lineage fe1d8770c866
  • energy/claim/global-demand-2030-updated: reconstructed · lineage 87c455f46228
  • research-automation/claim/no-aggregate: reconstructed · lineage aa0f66a745fd
2
CANONICAL CORRECTIONSTIME-MODEL REVISION RECORDS

What changed

  • research-automation/claim/rexbench-capability · corrected 2026-08-13T09:42:11.000Z
    Before: RExBench reports that the best result, even with human-written hints, is below 40% on realistic research extensions.
    After: RExBench v3 reports that the best result, even with human-written hints, is below 44% on realistic research extensions.
  • agent-horizon/claim/evidence-state · corrected 2026-08-23T00:52:22.230Z
    Before: Agent Horizon shows two compatible reliability levels, P50 and P80, for the same public frontier cohort. They are concurrent choices of success requirement and do not constitute a time series; suite coverage and the 16-hour boundary limit generalisability.
    After: Agent Horizon shows 4 compatible P50/P80 estimates across 2 public model cohorts under the same TH1.1 contract. They are concurrent model and reliability observations, not a time series; suite coverage and the 16-hour boundary limit generalisability.
7
ASSESSMENT REVIEWFRESHNESS IS SEPARATE FROM CANONICAL STATUS

Assessment evidence

  • assessment/singularity/proximity: mixed / published-baseline · audit current · supporting 2 / contradicting 2
  • assessment/scenario/ai-2027: watch / published-baseline · audit current · supporting 2 / contradicting 2
  • assessment/milestone/agent-task-horizon: partial / published-baseline · audit current · supporting 1 / contradicting 1
  • assessment/milestone/automated-ai-rd: watch / published-baseline · audit current · supporting 1 / contradicting 1
  • assessment/milestone/superhuman-coder: unconfirmed / published-baseline · audit current · supporting 2 / contradicting 2
  • assessment/physical-deployment/constraints: mixed / published-baseline · audit current · supporting 2 / contradicting 1
  • assessment/scenario/ai-2027: watch / reviewed · audit current · supporting 2 / contradicting 2

Historical assessment revisions: unassessed; no canonical assessment-history registry is available.

UNASSESSED
REVIEW DISAGREEMENTFAILS OPENLY, NEVER SILENTLY

No canonical disagreement registry is recorded

Absence of a registry is reported as unassessed, not as agreement or resolution.

VERSIONED EXPORT58 claim provenance records

This is a deterministic structural reconstruction, not independent source re-analysis or editorial approval. A missing, malformed or unresolvable lineage set or reference fails closed as missing-lineage; it is not evidence that a claim is false. No canonical reviewer-disagreement registry exists yet, so disagreement status is unassessed rather than resolved.

Download trust-audit-v3 JSON →