CalCOFI CTD QA/QC release v2026.08.03
Rule results

Counts are computed over the full result set; the findings table below is capped for display, so a large count is never understated.

Cruise context
Casts on this cruise; the selected one is ringed. Click one to switch to it.
Cast span — every variable this cast recorded

Full-resolution min/max across both directions, with the declared valid_min / valid_max beside it; anything outside is red. Orientation, not a check: this is recomputed from the values we published, so a corrupted value just widens the span and cannot disagree with itself. The real cross-check is against the processor's own YYMM_span_###-###.csv , which is not yet ingested (question Q15).

Scans

Click a scan in the plot to select it in the table, or a table row to highlight it in the plot. A flagged scan is ringed in red.

File
Column mapping

Every column, and what it became. Unmapped columns are the point of this table — a renamed sensor or a new instrument shows up here first. Raw voltages are unmapped deliberately: which sensor a V0 belongs to depends on the wiring.

Rule results

The same registry the release is checked with, run unchanged — every rule targets obs/sample, so projecting the upload into that shape is all it takes. A rule whose input this file does not carry reports skip, never pass.

Download findings (CSV)
Recorded verdicts
Registry

Rules are data, not code — they live in the workflows repo at metadata/qc_rules/ so they version with the pipeline that produces the data they check. Parked rules are listed with the reason they cannot run yet.

For what each rule means — the reasoning, the threshold and how it was derived, the known limitations, what the source documentation says about the corrections behind these values, and how this compares with QARTOD and oce — see the QA/QC protocol, which is generated from this registry.

Quality codes are dataset-scoped. The CTD files use 0/blank = good, 1 = use the primary sensor, 2 = use the secondary, 8 = questionable, 9 = bad or missing. 1 and 2 are sensor-selection instructions, not quality grades — and because 9 covers "bad", a 9-flagged row that carries a number is bad data, not a contradiction. The bottle database uses a different set (including 6, which has no CTD meaning).