Data Provenance: Tracing a Reported Number Back to a Tree

A verifier reads your monitoring report, points at a figure, and asks how it was derived. The answer requires walking backwards through every transformation the number went through until you reach a person standing next to a tree with a tape measure. Projects that can do this in an afternoon pass verification. Projects that cannot spend months reconstructing, and reconstruction looks identical to fabrication from the outside.

Each link is a transformation, and each transformation can lose the connection to the one before it.

Most projects can produce steps 1 and 2. The break is usually between 3 and 5, which often involves a spreadsheet where plot data was entered, cleaned, corrected, and re sorted, with no record of what changed.

The cleaning step. Field data always contains errors: transposed digits, impossible diameters, missing species. Correcting them is necessary and correct. Doing it in place, overwriting the original, destroys the link to what was measured.

The fix costs nothing: keep raw and cleaned as separate files, and log every correction with its reason. The cleaned file is what you calculate from; the raw file is what proves you did not simply choose convenient values.

The undocumented re-run. A calculation is redone with a corrected parameter and the new output replaces the old. Three months later nobody can say which version generated the number in the report.

The departed analyst. One person understood the workbook. The formulas are live, the logic is in their head, and they have left.

Raw field records retained and never edited. A documented cleaning step with a change log. Calculation performed by something re runnable, such as a script rather than a chain of manual copy paste operations, so the same inputs reproduce the same outputs. Version stamps tying a report to the exact data and code that produced it.

None of that is exotic. It is ordinary data discipline, applied to an activity where the records must survive decades and be defensible to a hostile reader.

Pick a number in your own last monitoring report at random. Trace it back to individual tree records. Time yourself.

If it takes more than a day, your verification will be expensive and your findings list will be long. If you cannot do it at all, you have a problem that is cheaper to fix now than at the next verification, and far cheaper than at the one after, when the same gap has propagated through another reporting period.

Registries and buyers are converging on the same demand for traceability, for the same reason: the credits that have damaged market confidence were rarely wrong in their arithmetic. They were unverifiable in their inputs.

A project whose numbers can be traced is making a claim that can be checked. One whose numbers cannot is asking to be trusted, and the market has become noticeably less willing to do that.