Data methodology
A Fresh Pricing Snapshot Can Still Contain Old Prices
A new dataset build timestamp proves that a file was regenerated, not that every underlying price was verified at that moment. Source retrieval, observation period, price-effective date, and verification date answer different questions. ByteCosts now exposes dated reference observations without relabeling them as current quotes. In particular, the initial ECB reference snapshot retains its 10 June 2026 observation date even though the page was retrieved on 5 September 2026.
A label saying updated today can describe a successful build around a failed or partial refresh. The distinction is important whenever a cost estimate combines sources with different release schedules or retains an earlier value when a request fails.
The correct response is not to discard every older observation. It is to preserve what the date means and avoid promising a level of freshness that the evidence does not support.
Four dates that should not be collapsed
A build timestamp records when an artifact was generated. A retrieval timestamp records when a request obtained source material. An observation period describes the time represented by a statistic. A price verification or effective date connects a quoted price to the source or billing rules that support it.
| Metadata | What it establishes | What it does not establish |
|---|---|---|
| Build time | The output file was generated | Every source was contacted successfully |
| Retrieval time | Source material was obtained | The material describes today’s conditions |
| Observation period | The period represented by the data | A customer can buy at that value now |
| Verification or effective date | Evidence for a rate or its applicability | All other rows share that evidence |
These fields can agree, but they should not be forced to agree. A successful response from a statistics API can legitimately return a six-month reference period. A pricing page retrieved today may contain a future-effective change that should not be applied to yesterday’s invoice.
A concrete reference-data example
The initial ByteCosts reference export preserves an ECB observation dated 10 June 2026. The source page was retrieved during implementation on 5 September 2026, but it exposed the June values. The dataset explicitly describes them as historical reference rates, not September rates or bank transaction quotes.
The electricity references use different periods. The selected Eurostat household observations describe the second half of 2025; the initial EIA residential observation describes December 2025. Their retrieval date does not change those periods.
The dataset also retains the original units. The EIA observation is 17.24 cents per kilowatt-hour, which ByteCosts normalizes to 0.1724 USD per kilowatt-hour by dividing by 100. Preserving the original value makes that transformation auditable.
Old does not automatically mean useless
A historical reference can support a dated example, a sensitivity analysis, or a comparison with an earlier budget. It should not silently become the default answer to a current purchasing question.
For a production budget, use a current provider quote, the applicable contract, or your actual electricity tariff. Household averages are not commercial data-center tariffs. Nor should the selected U.S. monthly observation and European half-year observations be presented as a synchronized country ranking.
A useful interface makes the relevant date visible next to the rate. Hiding the caveat several pages away in a methodology document makes it too easy to copy a number without the information needed to interpret it.
Handle failed refreshes without inventing certainty
Retaining the last valid observation can be preferable to replacing it with null, provided the original observation and verification dates remain intact. Mark the refresh failure separately. Do not overwrite the row’s evidence date with the current clock merely because a surrounding file was saved.
If a source response changes units or loses a required field, reject the observation rather than treating the missing field as zero. Zero is a real numeric claim. The new reference-data parsers check series dimensions and units before publishing a normalized value.
The same principle applies to price changes. A newly discovered difference can be a correction, a new SKU, or a changed region, rather than an actual provider price cut. Explain the evidence before turning every diff into a headline.
A practical check before sharing a cost estimate
Read the source date and billing unit, verify the exact product configuration, and keep the assumptions with the result. When the estimate mixes periods, say so. A precise-looking total cannot make its inputs more current than they are.
The source page shows the reference periods and native units; the methodology explains how to interpret pricing evidence. Those details are part of the result, not optional decoration.
Sources and method
This article explains the metadata in the committed ByteCosts reference dataset. It does not claim a fresh September ECB quote. The reference values retain their original units, periods, and attribution; normalization is identified as a ByteCosts calculation.
A Fresh Pricing Snapshot Can Still Contain Old Prices. ByteCosts. Updated 2026-09-05. https://bytecosts.com/blog/pricing-snapshot-date-vs-evidence-date/