Overview · Open dataset

Open dataset

The full dataset under CC-BY-4.0. Anyone can reproduce every published number by combining this dataset with the methodology and the reference implementation. If your reproduction differs, file an issue.

methodology v1.0 · sha256 fa87446ef6e023ee60092ffd…

CSV export

incidents.csv

Every incident with the computed impact score, the methodology version that produced it, and archival provenance flags.

↓ download

JSON export

incidents.json

Same rows, plus the original provider payload — enough to re-derive every published number under any methodology version.

↓ download

Schema

id	'{provider_id}:{external_id}'
provider_id	'openai' \| 'anthropic' \| ...
external_id	provider-native incident id
title	incident title from feed
status	'investigating' \| 'identified' \| 'monitoring' \| 'resolved'
provider_impact	'none' \| 'minor' \| 'major' \| 'critical'
uis	0-100, computed under methodology_version
confirmed	true = multi-source gate passed (live only)
confirmation_sources	array of source identifiers
archival_source	null = live · 'provider_api' \| 'wayback'
confidence_tier	'high' \| 'medium' for archival rows
started_at	incident start, ISO 8601
detected_at	when we observed it
resolved_at	null while open
components	array of provider component names
methodology_version	integer (1 = v1.0)

Citation

ai-receipts (2026). Independent AI provider reliability dataset.
Methodology v1.0.
sha256 fa87446ef6e023ee60092ffdcb196ecd7a84f8fe316458c58c875c139c2656bc
CC-BY-4.0. https://aireceipts.org/data