Collection process & sampling bias.
ClearBrief is an independent learning and portfolio project built around public NASA Aviation Safety Reporting System (ASRS) reports.
Where the reports come from
This library uses a small selection from NASA’s published topical report sets. Each report links to the original PDF and its starting page.
How we expanded the collection
- Start with the five original NASA PDF snapshots, each containing 50 full records. Keep every record regardless of length, apparent severity, or how interesting it sounds.
- Extract each primary accession number, month, aircraft, flight phase, complete narratives and NASA synopsis. Normalize line-wrap whitespace only; retain anonymized identifiers and original wording.
- Merge exact duplicate primary accession numbers, retaining every source membership. Keep similar narratives with different accession numbers: they may describe the same event.
- Verify record counts and ordinal headers, required fields, page links and source hashes. Compare the original 40 accounts with their saved text. Automated checks do not verify the truth of accounts or replace manual transcription review.
Bias we reduce—and bias that remains
- Our selection: the original 40 favored concise accounts. Including all records removes that preference within these five files, but the original choice of five topics still limits coverage.
- Voluntary reporting: awareness, willingness to report, recall and self-presentation affect who reports and what they describe. Missing reports and reporting probabilities are unknown.
- NASA topic selection: these are pre-screened topical sets, not a random sample of the ASRS database. Each set contains 50 records by design; equal counts do not mean equally common safety problems.
- Time and operational coverage: topic sets cover different months and years. These saved PDFs are not current incident surveillance and cannot support time-trend comparisons.
- Overlapping events: accession deduplication prevents counting the same record twice, but multiple accounts can describe one event. Counts are records, not confirmed unique incidents.
- Interpretation and extraction: accounts are unverified, anonymization removes context, PDF extraction can introduce errors, and AI briefs can omit or misstate facts.
No statistical correction is claimed. We do not know inclusion probabilities or the number of flights exposed. Weighting these records would create false precision. Use them for qualitative exploration, not population estimates, event rates, risk rankings or causal conclusions.
Download collection metadata · Download all records and source variants · Download the inclusion and deduplication log
What AI contributed
The original 40 records retain their AI-written titles and briefs, marked as awaiting human review. Newly included records display NASA’s own synopsis and an identifier-based title. No new AI brief is implied. Collection-topic filters and charts now use the same source-membership rule for every record; the earlier AI theme labels are preserved in the downloadable dataset but are not used in these charts.
How search and related reports work
Search matches words, common aviation abbreviations, and selected synonyms across report text and summaries. Related reports share source collections and keywords. Neither feature assesses severity or determines a cause.
What the charts mean
All charts reflect the current search and filters. A report may have several source memberships or flight phases. This curated sample is not representative of all aviation events, so counts cannot measure incident rates or aviation risk.
Read the original
ASRS reports are voluntary accounts, sanitized by NASA to remove identifying details. NASA does not verify or validate the reports. AI summaries may omit context or make mistakes; compare them with the source narrative.
Visit the NASA ASRS database ↗