A plain explanation of the canonical data model behind every publication here -- including two real errors the verification process caught this month, before they reached print
lucabindi.com is built on a canonical data model with 320 registered series across 16 official providers, 284 of them currently populated with 156,605 real observations. Every article must name its exact evidence base and separate observation from interpretation. This piece explains that design plainly, including two real cases this month where the platform's own verification process caught genuine errors before publication: a three-source unit contamination in a US inflation series, and a code bug that had silently substituted a regional aggregate for two countries' individual growth data (caught because the resulting growth rates were suspiciously identical). Both were fixed and documented rather than quietly patched.
Every claim on this platform traces back to a specific, queryable number in a database, not to a language model's memory or a generic web search at the moment of writing. That distinction is the entire point of how lucabindi.com is built, and it is worth explaining plainly rather than leaving it implicit. As of this month, the platform holds 320 registered data series across 16 official providers (central banks, statistical offices, and international institutions), of which 284 are populated with real observations -- 156,605 individual data points in total. Every published article is required to name exactly which of these series it used, and to state clearly which parts of its argument are observation versus interpretation.
Each data series on the platform has a single canonical identity (for example, US real GDP, or the Euro Area's 10-year government bond yield) that can, in principle, be fed by more than one official source -- useful when one provider updates faster or covers a longer history than another. This flexibility comes with a real discipline requirement: every source feeding a given series must report the same underlying quantity in the same unit. When that discipline breaks, the result is not a subtle error but a detectable one, and this platform has caught and corrected two such cases in its own recent history rather than leaving them undiscovered.
First: a US inflation series was found to be simultaneously fed an index level, a year-on-year percentage, and a differently-based index from three different official sources under one canonical identity -- a genuine data-model defect. It was diagnosed, the incompatible sources were deactivated, and the contaminating historical rows were removed, leaving a single clean source matching the series' stated unit. Second, and more instructive: while preparing a piece comparing growth across several major economies, two countries' growth rates came out identical across thirteen consecutive quarters -- not a plausible coincidence for two different national economies. That similarity was the signal. Investigation traced it to an ingestion route that had silently substituted a regional aggregate for country-specific data. The bug was fixed at the code level, the incorrect historical data was deleted, and the two series were re-collected correctly before any article was published using them. In both cases, the error was caught before publication, not after -- because the editorial process requires checking the numbers against each other before writing about them, not after a reader notices something looks wrong.
A deterministic, evidence-first process reduces certain failure modes -- fabricated statistics, unsourced claims, silent copy-paste errors -- but it does not eliminate every risk. Official data sources themselves get revised, sometimes substantially: this platform has directly observed a major US external-position figure revised by more than five trillion dollars for the same reporting date between two official releases, and has learned to state which vintage of a number it is using rather than treat any single figure as permanently final. Some series are also thinner than others -- a handful of newly added series have only a few weeks of history -- and the platform treats that as a stated limitation on any trend claim built from them, not a reason to avoid using genuinely new data.
Every published piece on this platform is expected to show its evidence base explicitly -- which series, how many observations, how current -- state what could be wrong with the analysis, and separate observation from interpretation from forward-looking scenario. The two corrections described here are shared publicly rather than quietly fixed, because a research platform's credibility rests on being verifiably careful, not on appearing to have never made a mistake. Coverage is deliberately uneven in places -- deep for the United States and the Euro Area, thinner for some emerging markets and for a few specific indicators -- and that unevenness is documented rather than papered over with confident-sounding but unsupported claims.