Data Methodology & Guardrails
Open documentation of data connectors, LLM boundary guardrails, and verification scores.
Release v20260807_031925 is assembled from 10 resources exported
from the OpenFly database by scripts/sync_release_data.py. Each resource carries named source
references with retrieved-at timestamps and content hashes, so claims link back to a specific page and crawl time.
Guardrails
- Connector links are only recorded if the source URL validates against an allow-list of known domains; anything else is rejected at collection time.
- No claim is invented: summaries, versions, licenses, and event timestamps are copied verbatim from the referenced source, never patched by models.
- Detail pages explicitly surface "Unknown" when a fact has not been verified rather than fabricating a placeholder.
- Release data is a static, hash-pinned artifact; the deployed site does not call backend endpoints at runtime.
Source Tier Distribution
| Tier | Definition | Resources |
|---|---|---|
| A | Official upstream or regulator | 10 |
| B | Maintained community / vendor | 0 |
| C | Aggregated or secondary | 0 |
| D | Unvetted candidate | 0 |