Styxis
Company
Knowledge

Why summing material categories changes the count of 46 strategic projects

Maintain project counts separately from material-and-stage contributions when allocating support.

Repeated rows can be legitimate

A project contributing to several materials and processes should appear in several relationship rows. Adding those rows and calling the result the number of supported projects can inflate a budget argument. Truthound's role is not to remove every repeated identifier. It is to check the different keys used by the project register and its contribution tables.

A selection is not an operating plant

On October 9, 2026, the European Commission selected 46 new critical-raw-materials Strategic Projects in 16 Member States. 8 are integrated projects combining extraction with processing or processing with recycling. Strategic selection and an operating production facility remain different stages.

Separate the register from relationships

Give each project one identifier in the master register. Represent material and process participation in a separate relationship table, allowing one project to contribute to several categories. Keep selection, environmental assessment, construction permission and operating state distinct rather than collapsing them into a single in-progress value. The differences identify the work or support still required.

Which key must be unique?

Uniqueness therefore depends on the table. The register may require one project_id; the relationship table may require a unique project_id/material/stage combination. Removing every repeated project_id from the second table loses legitimate integrated contributions. Counting all its rows as projects produces the opposite error.

Protect the aggregation boundary

Truthound Core Python SDK 3.3.4, operating on a prepared Polars table, provides MultiColumnUniqueValidator for a declared composite key and DistinctCountValidator for expected cardinality bounds on a column. The operator must establish the expected project population and relationship scope from the official list. A filtered subset should not inherit the same expected count as the full selection.

Do not delete a duplicate candidate blindly

An existing pandas 3.1.1 register can keep DataFrame.duplicated with explicitly chosen subset and keep settings. The important step is to mark candidates rather than delete them automatically. Preserve enough context to distinguish a duplicate master record from a legitimate material relationship. Compare Truthound when that definition needs to become a reusable validation contract.

Capacity needs another state boundary

Adding capacity requires another boundary review. Processing capacity and extraction capacity are not interchangeable totals, and a selected project's planned capacity is not realised output. Preserve material-specific units, project stage and the reference date before deciding which sum answers the investment question. Avoid defaults that promote an unbuilt plan into an operating result.

Give decision-makers two numbers

Give the budget review two linked outputs: the number of projects receiving resources and their contributions by material and stage. One answers where support goes; the other shows which capability gaps remain. Traceable relationships make those questions explainable without concealing their differences inside one attractive total.

Conditions

Reconcile the official selection list with verified project, material and stage relationships before aggregation.

Alternatives

Keep an existing pandas 3.1.1 register with duplicated checks against the appropriate business keys.

Limitations

  • Selection is not operation or delivered production, and validation cannot guarantee future supply.

Evidence

Event checked
Subject checked
Modified