HS Code Versioning

Revisions are not cosmetic renumbering. They alter the shape of the classification tree, and the different kinds of alteration have different consequences for anyone reading data across the boundary.

Split

One code becomes several, because the trade in it has grown or because a subset has acquired policy significance. Pre-revision data exists only at the aggregate; the detail cannot be recovered retrospectively.

Merge

Several codes collapse into one. Detail that existed before the revision stops being collected afterwards. A series can be carried forward only by aggregating the earlier years to match.

Relocation

A good moves to a different heading or chapter, often to reflect a better view of what it is. The code changes even though the good has not.

Creation

A new code appears for goods that previously sat in a residual or "other" category. The new code starts from zero history, and the residual category it came out of falls correspondingly.

Deletion

A code is withdrawn, and the goods it covered are distributed elsewhere. A query against the old code returns nothing after the boundary, which is easily mistaken for trade having stopped.

The WCO publishes correlation tables between adjacent editions, in both directions, so that a code in one edition can be traced to its counterpart in the next. These tables are the authoritative instrument for moving between editions, and they are the starting point for any serious attempt to build a series across a boundary.

What they cannot do is manufacture information that was never collected. Where a correlation is one-to-one, the mapping is clean and a series can be carried across. Where one old code maps to several new ones, the tables identify the successors but cannot say how the historical quantity would have divided between them, because nobody recorded it at that level. Where several map to one, the reverse problem applies going backwards. A chain of correlations across multiple boundaries compounds this: a series carried from an edition two or three revisions back may be resting on assumptions at every step.

Agricultural chapters are affected by revision like any other, and often for reasons specific to the sector. Interest in a particular product form, a distinction between uses, a newly significant processed derivative, or a policy concern that requires separate visibility all tend to produce splits. The result is that long-run series for farm and food products are among the more likely to cross boundaries that matter.

The effect is compounded by how agricultural analysis is usually done. Analysts rarely work with a single subheading; they build baskets — all wheat, all vegetable oils, all pulses — by summing codes. A basket assembled from an edition’s code list and then applied unchanged to a different edition can silently drop a code that was split off, double-count one that was merged, or include a good that was relocated into the range from somewhere else.

• A code list that defines a commodity basket is edition-specific and must be rebuilt for each edition it is applied to
• Apparent jumps or collapses at a revision year should be checked against the correlation tables before being interpreted
• Series compiled by international organisations may already be harmonised across editions — how, and with what assumptions, is stated in their documentation
• National tariff lines below six digits change on their own national schedules, independently of HS revisions

The last point is worth separating out. National extensions beyond the sixth digit are maintained by each jurisdiction and can be restructured at any time, for reasons unconnected to a WCO revision. A national series can therefore break at a year when the international nomenclature did not change at all.

The discipline that follows from all of this is simple to state. An edition is recorded alongside every code, and a code without one is treated as incomplete rather than as a fact. Where a series crosses a revision, the boundary is identified and the correlation is done deliberately, with any apportionment assumption written down rather than absorbed.

AgricultureID applies this to its own content: every classification code published on the platform carries the edition it belongs to, and codes are given at a level of detail the source actually supports. Where a code’s edition is unknown, the code is not published.