What quality flags tell a customer buying ocean data

Ocean observations arrive with a second product attached: information about how much confidence the supplier places in each measurement. A customer can lose that information while preserving every measured value. The resulting file may load perfectly into an analysis application and still support a less defensible decision than the original delivery.

For a business buying a continuing marine-data service, quality flags affect usable coverage, review workload and the meaning of a completed delivery. They deserve attention during product evaluation, before a contract turns a monthly observation count into an acceptance measure. The question is how the service distinguishes observations that passed defined checks, observations awaiting assessment, unusual observations requiring attention and records that failed.

Read the named scheme before the numbers

The June 2020 QARTOD flag manual provides a concrete scheme. Its Table 2 assigns 1 to data passing critical real-time checks and suitable for preliminary use; 2 to unevaluated data or unavailable quality information; 3 to suspect observations or observations of particular interest; 4 to failures of critical checks; and 9 to missing data. The manual also encourages additional flags that explain individual checks.

Those distinctions have immediate purchasing consequences. An unevaluated observation has a different evidential status from one that passed. A flag drawing attention to an unusual observation is not automatically a reason to discard it. A preliminary pass should not silently become a promise that no later assessment can change the result. These are differences in what the supplier knows, rather than different decorative colours for the same value.

The numbering belongs to a named scheme. A receiving system should not assume that a larger number always means worse quality, or that another programme uses the same numbers. Ask for the scheme name, version, applicable variables and any local extensions. If a supplier translates several upstream schemes into one customer vocabulary, the translation itself becomes part of the product documentation.

Preserve the reason as well as the summary

A single summary status helps customers filter a large delivery quickly. It can also hide why a record needs attention. Commercially, that matters because different reasons imply different follow-up work. Missing quality information may require a conversation with an upstream provider. An observation outside an expected range may require scientific review. A measurement affected by a documented instrument problem may need replacement or exclusion.

The CF Conventions version 1.11 describes associations between measurements and ancillary information, including uncertainty and quality flags. It distinguishes individual checks from aggregate flags and provides metadata for interpreting flag values and combinations of conditions. This supplies a way to carry meaning with the data; it does not establish that a particular commercial delivery contains all the evidence a customer needs.

A buyer can ask the supplier to demonstrate one ordinary record and one record requiring review in the actual export. Can the recipient identify the observation, see its summary status, inspect the relevant supporting status and recover the definition without returning to a proprietary dashboard? If the answer depends on a human explaining a screenshot, the handover is incomplete for an automated data product.

This is closely connected to marine-data handover standards. A familiar file format makes transfer easier, but acceptance should include the associations between measurements and the metadata that qualifies them.

Define what counts as a delivered observation

Consider a hypothetical subscription supplying environmental observations for a commercial port study. The invoice counts all records transferred. The research team counts only observations passing its agreed quality policy. Both counts can be accurate while describing different quantities. If the purchase order says only “monthly data delivery,” the disagreement may not emerge until the customer tries to compare months.

A useful service definition separates expected records, received records, records evaluated, records passing the agreed checks and records withheld or rejected. The denominator matters as much as the percentage. A provider reporting a high pass rate among evaluated records might still have substantial gaps in evaluation. Conversely, a provider that reports unusual data openly may appear worse than one that omits it before delivery.

The buyer should choose the measure that reflects the purchased outcome. A raw-data subscription may reasonably include flagged observations because the customer has its own specialists. An analysis-ready subscription may include further review and explanation. These are different offers with different labour requirements. Comparing their prices without distinguishing the promised state of the data rewards ambiguous packaging.

A coverage report should also preserve time and variable distinctions. Good temperature coverage does not establish good coverage for every other measurement in the same file. A monthly total can conceal a concentrated gap during a period central to the study. The commercial review can examine those differences without prescribing how a vessel or sensor should operate.

Decide who owns the review queue

Quality flags create work only when someone can act on them. The contract should identify whether the supplier investigates flagged observations, whether the customer provides scientific judgement, and how unresolved cases remain visible. A promise to deliver flags alone is different from a promise to explain them within an agreed reporting cycle.

The review queue needs a stable observation reference, the reason for attention, the current owner and the date of the latest decision. Otherwise the same record can be repeatedly questioned in different customer teams, generating unnecessary support work. A concise explanation attached to the observation may be more valuable than an additional summary chart.

Review timing should reflect the product. A preliminary feed may prioritise prompt availability, with a later assessed edition following on a stated schedule. A historical dataset sold for comparative analysis may need its review completed before acceptance. The customer should understand which edition supports which decisions, and the price should include the corresponding work.

Instrument history can inform that assessment, but it is a separate layer of evidence. Our guide to marine-sensor calibration services examines the records surrounding an instrument. The quality flag describes an observation within a data process. Buyers should be able to connect those layers where relevant without treating a calibration certificate as a blanket approval of every observation.

Keep earlier decisions reproducible

A later assessment may change a flag or introduce a revised measurement. The customer needs to distinguish a corrected delivery from the version used in an earlier report. Simply downloading the same filename again can make a past decision impossible to reconstruct if the supplier has replaced its contents.

An effective commercial handover gives each delivered edition an identity and explains material changes. The change record should say whether values, flags, definitions or coverage changed. Those distinctions matter to customers deciding whether to rerun an analysis, update a published chart or merely retain the new documentation for future work.

The supplier and customer should agree how long earlier editions remain obtainable. Keeping every internal processing intermediate may be unnecessary for a small service, but retaining the exact delivered product and its explanatory metadata is a practical foundation. Access after the subscription ends is also relevant: the customer may need to substantiate a report long after it stops purchasing new observations.

A strong offer makes the path from observation to customer decision visible. It defines quality states, preserves their explanations, measures coverage honestly and prices the human work associated with unresolved data. That gives a marine-data business a defensible basis for selling reliability, while giving its customer a clear understanding of what has actually been assessed.