Styxis
Company
Knowledge

Reproduce a climate-finance change as a revision plus a new-period result

Reconcile publication vintages before interpreting an increase or decrease as current delivery.

Two reasons a cumulative figure moves

A cumulative result can change because new work was delivered or because an older estimate was revised. Those are different explanations. Truthound can compare an analyst-defined SQL reconstruction with an approved expected table. It cannot establish a program's causal impact or certify that every underlying estimate is accurate.

The publication's explicit distinction

The UK government published its 2026 International Climate Finance results on November 4, 2026. It explicitly distinguishes historical revisions from results newly delivered in 2025/26. Its totals are estimates with methodological qualifications. Reading the difference between two headlines as this year's delivery alone would discard that distinction.

Use the original tables and vintages

Prepare a reconciliation by program, indicator, target year, publication version and revision reason. Use the original numeric tables and their units, not rounded headline figures as exact expectations. Preserve the previous publication, the revised historical value and the latest-period increment. Their relationship should be explainable before a chart is interpreted.

Compare the approved reconstruction

For the climate-finance results reconstruction prepared as a Polars DataFrame, Truthound Core Python SDK 3.3.4's QueryResultMatchesValidator can compare supplied SQL results with expected_df or an owner-written matcher. It can help reproduce the approved decomposition and expose an unexplained difference. The SQL, reference table and tolerance policy are the analyst's responsibility. Passing a reconstruction does not validate the real-world effects represented by the estimates.

Keep schema preparation complementary

Pandera 0.35.0 can support schema preparation for the same source tables, including Polars. A schema check is complementary to the result comparison; it should not be described as performing the identical SQL reconciliation automatically. Keep the established preparation contract where it preserves indicator types and units, then specify the substantive comparison separately.

Test a counterintuitive total

Test a case where the historical estimate falls while the current period adds a positive result. The cumulative change can be negative without implying that support was removed on the ground during the latest year. This is a numerical illustration, not an assertion about a named program. Use it to check whether the reconciliation explanation preserves both components.

Escalate the unexplained remainder

Send unexplained differences to the source owner with the exact indicator, vintage and calculation. Report the revision separately from new-period delivery and retain the approved reconstruction. This creates a reproducible diagnostic trail without converting statistical consistency into a claim of additionality or program success.

Evidence

Event checked
Subject checked
Modified