DAY 2 · CORPUS REALITY CHECK
CIDM 2,231 Corpus QA Dashboard v1.0
We are now testing the database that supports the website's claims, rather than assuming that a working Explorer means every classification is equally reliable.
Strong structural signals: 2,231 records; complete 1995–2026 year coverage; no duplicate current dedup keys; no duplicate PMID extras.
Important finding: classification QA has surfaced historical-label leakage. Fourteen pre-2005 records are currently labelled “Whole-genome sequencing”, and 230 pre-2019 records carry the combined “COVID-19 / respiratory viruses” theme. These are review flags, not automatic errors: the taxonomy/rules need inspection before historical interpretation.
QA metrics
| Check | Value | State | Interpretation |
|---|---|---|---|
| Records | 2231 | PASS | Embedded Explorer population length. |
| Year span | 1995–2026 | PASS | No missing calendar year in the working corpus. |
| Duplicate dedup_key | 0 | PASS | No duplicate current dedup keys. |
| Duplicate PMID extra records | 0 | PASS | No duplicate PMID extras. |
| Duplicate DOI extra records | 1 | REVIEW | DOI duplication needs inspection; dedup key remains unique. |
| Missing DOI | 52 | QUALIFIED | DOI is not universal; missing does not imply invalid record. |
| Missing journal | 499 | REVIEW | Large enough to affect journal-level analyses. |
| Pathogen unspecified/mixed | 1471 | BOUNDARY | Classification coverage is incomplete by design/current rules. |
| Technology unspecified/mixed | 1089 | BOUNDARY | Classification coverage is incomplete by design/current rules. |
| Pre-2005 WGS classification flags | 14 | REVIEW | Likely broad keyword/classification leakage; requires rule inspection. |
| Pre-2019 COVID-theme flags | 230 | REVIEW | Theme label combines COVID-19 / respiratory viruses; historical records need clearer sublabel semantics. |
| Affiliation verification required | 2231 | BOUNDARY | All current records retain this provenance boundary. |
Why this is useful
The corpus can be structurally healthy while a classification layer still needs calibration. This is exactly why iCIDM separates source records, derived classifications, interpretation and validated institutional conclusions.
Open review flags CSV → · Publication Explorer → · High-Impact Claims →