Styxis
Company
Knowledge

An available reference database does not preserve an old analysis

Long-term use of life-sciences references requires versioned snapshots and relationships, not continued access alone.

An open address is not a preserved analysis

A reference database can remain online while an older analysis becomes difficult to explain. Its current entries and links may differ from those originally used. Truthound belongs at the relationship between a result table and a chosen reference snapshot, rather than being asked to decide biological truth.

European support for continuity

On October 6, 2026, the European Commission earmarked EUR 65 million for GENCODE, UniProt, Europe PMC and the wider life-sciences data ecosystem. The support concerns maintenance and continuity, with grant-agreement evaluation still required. An earmarked amount should not be described as funding already disbursed.

Make the snapshot part of the result

Keep the reference snapshot name, version, acquisition time and identifier policy beside the result. An updated analysis can use newer data without erasing what the earlier analysis saw. Design continuity of access and continuity of analytical evidence as separate responsibilities.

The same identifier can change state

An identifier may persist while its description changes, its entry is withdrawn or its replacement link is revised. A result that fails to match today might reflect an input error or a legitimate version difference. Without separate old and current snapshots, those causes become indistinguishable missing references.

The reference set supplied to Truthound

Truthound Core Python SDK 3.3.4, operating on a prepared Polars table, provides ReferentialIntegrityValidator to compare values with the supplied reference_data and reference column. MultiColumnUniqueValidator can check a composite accession-and-version key. The caller must provide the reference data and keys. These interfaces do not establish the biological meaning of a remote entry or acquire a release automatically.

Keep the useful join boundary

An existing pandas 3.1.1 merge(validate=...) can remain useful for 1:1 or 1:m join cardinality. That relationship check differs from membership of an approved snapshot. Instead of removing the existing join, decide where version and unmatched-reference review belong. Compare the two validation boundaries without claiming that either resolves scientific interpretation. Keep missing accessions separate so pandas matching null keys on both sides cannot appear to confirm a reference entry.

Do not erase withdrawn history

Deleting a withdrawn entry from the historical snapshot makes the original result harder to explain. Preserve its earlier use and attach withdrawal state, replacement identifiers and change evidence in a later version. Whether an entry should be used now and whether an earlier dependency must remain explainable are different decisions.

Spend storage on explainability

Where storage is constrained, prioritise the reference versions and relationship tables needed to explain results rather than retaining every intermediate file indiscriminately. Literature links can change on their own schedule. The objective is recoverable dependency, not a large collection of URLs.

Approve the next refresh with its effects

Approve a refresh after its effects are understood: newly introduced entries, changed relationships and unresolved references in current results need distinct review. Continued access becomes a durable analytical asset when the team can preserve yesterday's explanation as well as use today's data.

Conditions

Provide a lawfully obtained reference snapshot and explicit identifier/version rules.

Alternatives

Retain pandas 3.1.1 join-cardinality checks while assessing membership of the chosen reference snapshot separately.

Limitations

  • Reference checks do not determine biological interpretation or clinical validity.

Evidence

Event checked
Subject checked
Modified